AI Automation Agency: Can You Really Make Money With It?

Yes, people are making money selling AI automation services—but that doesn't mean starting an "AI automation agency" is easy money. There is measurable demand for freelancers who can build AI agents, automate workflows and integrate AI into businesses. The difficult part isn't getting access to AI tools. It's finding businesses with problems worth paying to solve, building systems that work reliably, and convincing clients to trust you with important parts of their operations.

The evidence-based answer: AI automation is a real freelance and consulting market. What the evidence does not prove is the much stronger social-media claim that a beginner can learn a few no-code tools, send automated cold emails and reliably build a $10,000-a-month agency in a few weeks.

Table of Contents

• What Is an AI Automation Agency?
• Is There Real Demand for AI Automation Services?
• Where Is the Income Proof?
• How Much Can You Realistically Make?
• What Do Businesses Actually Pay For?
• Can You Do It Without Coding?
• What Does It Cost to Start?
• The Hardest Part Isn't AI
• Is the Recurring-Revenue Model Real?
• Why AI Automation Agencies Fail
• Who Has the Best Chance of Making Money?
• What If You're Starting From Zero?
• Is an AI Automation Agency Worth Starting?
• Bottom Line
• Frequently Asked Questions

What Is an AI Automation Agency?

An AI automation agency is essentially a consulting or service business that helps other businesses automate work using artificial intelligence and related software.

Despite the futuristic name, many projects don't involve creating a new AI model.

An agency might connect existing tools so that a business can automatically:

• Respond to common customer questions.
• Qualify incoming sales leads.
• Summarize calls or meetings.
• Extract information from documents.
• Route emails to the right employee.
• Update a CRM after a customer interaction.
• Generate drafts of routine communications.
• Schedule appointments.
• Search an internal knowledge base.
• Turn information from one business system into an action in another.

Some of these workflows use generative AI heavily. Others are conventional business automation with an AI component added to interpret text, speech, images or documents.

The agency gets paid because the client doesn't want to figure out how to connect, test, maintain and troubleshoot all of this themselves.

Is There Real Demand for AI Automation Services?

This is where the AI automation agency idea has considerably more evidence behind it than many online side hustles.

Upwork reported that gross services volume from AI-related work exceeded $300 million on an annualized basis in the fourth quarter of 2025. More specifically, its AI Integration & Automation category grew more than 90% year over year. That is marketplace spending rather than a survey asking businesses what they might someday do. ([Upwork](https://investors.upwork.com/news-releases/news-release-details/upwork-reports-fourth-quarter-and-full-year-2025-financial))

Earlier Upwork marketplace research found AI-related work growing 25% year over year in the first quarter of 2025. It also reported that freelancers doing AI-related work received an average hourly rate premium of more than 40% compared with freelancers doing non-AI-related work. ([Upwork Research Institute](https://www.upwork.com/research/ai-impact-work-categories))

Fiverr has seen a similar shift. Its Spring 2025 Business Trends Index reported an enormous increase in searches for freelancers specializing in AI agents. By June 2026, Fiverr was still reporting businesses hiring specialists to automate workflows, build agents and integrate newer AI tools. ([Fiverr](https://www.fiverr.com/news/spring-bti-2025))

So the demand is real. Businesses are demonstrably spending money on AI implementation and automation. The questionable part isn't whether a market exists. It's whether an inexperienced new agency can capture enough of that market to produce the income advertised on social media.

Where Is the Income Proof?

Search for AI automation agencies online and you'll encounter claims of agencies reaching $10,000, $20,000 or even $100,000 per month.

Some may be genuine.

But a revenue screenshot or founder case study is weak evidence of what a typical beginner earns.

There are several reasons to be skeptical.

Revenue Isn't Profit

An agency reporting $15,000 in monthly revenue might be paying for contractors, APIs, automation platforms, voice services, hosting, sales software and advertising.

The amount the owner actually keeps could be substantially lower.

One Successful Agency Doesn't Tell You the Failure Rate

Imagine 1,000 people attempt an AI agency and ten succeed spectacularly.

If you only see interviews with those ten people, the business model can look almost foolproof.

Without knowing what happened to the other 990, you cannot calculate the probability of success.

Some People Make Money Teaching the Business Model

There is also an important incentive problem.

A person selling a course, coaching program or agency blueprint benefits financially when you believe starting an AI automation agency is unusually easy or profitable.

That doesn't automatically make the person's claims false. It does mean income claims deserve independent scrutiny.

Case Studies Can Be Real Without Being Typical

A technically experienced founder with an existing business network may reach $10,000 per month quickly.

A beginner with no portfolio, sales experience, technical ability or professional network is starting from a completely different position.

Be careful with "income proof." A screenshot showing revenue proves, at most, that an account displayed a particular number. Good evidence should also explain the time period, expenses, source of customers, work performed and whether the result is repeatable.

How Much Can You Realistically Make?

There is no trustworthy universal income figure for an AI automation agency.

That's an unsatisfying answer, but inventing a range would be worse.

Income depends on several variables:

Factor	Why It Matters
Technical skill	More difficult integrations can justify higher prices and face less commodity competition
Sales ability	A brilliant automation has no commercial value if you cannot find a buyer
Industry knowledge	Understanding a client's workflow makes it easier to identify valuable problems
Portfolio	Demonstrated results reduce the risk perceived by prospective clients
Client size	A workflow worth $200 to a solo business could be worth thousands to a larger organization
Project complexity	Simple automations are easier for competitors and clients to reproduce
Reliability	Businesses pay more for systems they can trust with important operations
Recurring support	Maintenance and monitoring can create continuing revenue after implementation

A beginner might make nothing for months.

A capable freelancer might sell individual projects without ever building what most people would call an "agency."

An experienced consultant with a strong niche could build a substantial business.

All three outcomes are compatible with the evidence that AI freelance demand is growing.

What Do Businesses Actually Pay For?

Businesses generally don't care that you have an AI automation agency.

They care about an expensive or irritating problem.

Consider the difference between these pitches:

Weak pitch: "We build cutting-edge AI agents for businesses."

Better pitch: "Your staff manually answers the same appointment questions hundreds of times each month. We can automate the routine ones and send complicated cases to an employee."

The second identifies a measurable problem.

Services with clearer economic value can include:

• Customer-support automation.
• Lead qualification and follow-up.
• Appointment and intake automation.
• Document processing.
• CRM automation.
• Internal knowledge assistants.
• Email triage.
• Sales workflow automation.
• Voice agents for appropriate business use cases.
• Automated reporting and data extraction.

The more directly a project saves employee time, captures missed revenue or improves response speed, the easier its business case becomes to explain.

Can You Start an AI Automation Agency Without Coding?

Yes—for some types of work.

No-code and low-code automation platforms have dramatically reduced the technical barrier to connecting applications and building workflows.

Generative AI has lowered the barrier further because it can help users write code, understand APIs and troubleshoot integrations.

But "no coding required" can be misleading.

You may eventually encounter APIs, authentication, webhooks, structured data, databases, rate limits, permissions, error handling and software that doesn't behave exactly as the tutorial demonstrated.

More importantly, client systems contain real data and real consequences.

A demonstration that works five times on your laptop is not necessarily a reliable business system.

No-code doesn't mean no technical skill. You may not need to become a professional software engineer, but understanding how systems exchange data, how failures occur and how to troubleshoot them can separate a useful consultant from someone who simply knows how to copy an automation template.

What Does It Cost to Start?

An AI automation service can have relatively low initial costs compared with a traditional physical business.

You may already own the most expensive basic equipment: a computer and internet connection.

But operating costs can grow as your projects become more sophisticated.

Potential expenses include:

• AI model/API usage.
• Automation software subscriptions.
• CRM or sales tools.
• Voice and telephone services.
• Cloud hosting.
• Domain and website costs.
• Database or vector-search services.
• Monitoring and logging tools.
• Contractors or developers.
• Professional services and business expenses.

A sensible beginner usually doesn't need subscriptions to every AI tool recommended by influencers.

Learn enough to build a working solution first. Buy additional infrastructure when a genuine project requires it.

The Hardest Part Isn't AI

This is probably the most important reality missing from many "start an AI agency" videos.

The hard part is getting clients.

AI has made building simple demonstrations remarkably easy.

That means it has also made it remarkably easy for thousands of other aspiring agency owners to build similar demonstrations.

You still have to answer:

• Why should a business trust you?
• Why does it need this automation?
• What measurable problem does it solve?
• Why can't an employee simply set it up?
• What happens when the AI makes a mistake?
• Who maintains the workflow?
• What happens when an API or software product changes?
• How is sensitive customer or company information handled?

Those are business questions, not prompt-engineering questions.

Is the Recurring-Revenue Model Real?

It can be.

An agency might charge an initial implementation fee and then a recurring amount for hosting, monitoring, maintenance, support or ongoing optimization.

That can create monthly recurring revenue.

But recurring revenue needs recurring value.

Charging a client every month for an automation that requires no maintenance and creates no ongoing expense may eventually lead the client to question the arrangement.

Real recurring work can include:

• Monitoring failures.
• Updating workflows when software changes.
• Managing API usage.
• Improving prompts and knowledge sources.
• Reviewing AI accuracy.
• Adding new workflows.
• Providing support.
• Maintaining integrations.

The strongest recurring model isn't "get the client to keep paying." It is "keep doing something the client considers worth paying for."

Why AI Automation Agencies Fail

Even with strong market demand, several things can derail the business.

They Sell AI Instead of a Business Outcome

"AI agent" sounds exciting to people who follow technology. A local business owner may care much more about reducing missed calls or processing invoices faster.

They Choose a Solution Before Finding a Problem

Learning an automation tool and then searching for someone to buy whatever you built reverses the normal business process.

Start with the expensive problem.

They Underestimate Reliability

AI can produce unpredictable output. Business automation therefore needs safeguards, testing and human escalation when appropriate.

They Depend Entirely on Cold Outreach

Sending thousands of generic AI-generated emails doesn't create a durable competitive advantage—especially when thousands of other agencies can do exactly the same thing.

They Have No Industry Expertise

An accountant who understands accounting workflows and learns AI automation may have an advantage over an automation generalist trying to understand an accounting firm's problems from scratch.

They Believe the Tool Is the Skill

Today's popular automation platform may eventually be replaced.

Understanding workflows, customers, APIs, data, reliability and business economics transfers much better between tools.

Who Has the Best Chance of Making Money?

The strongest candidate may not be the person who knows the most about AI.

Consider someone who has spent ten years working in dental offices. They understand scheduling, insurance verification, patient reminders, missed appointments and repetitive administrative work.

If that person learns enough AI and automation to solve one expensive dental-office problem, they have something powerful:

AI skill + domain knowledge.

The same applies to people with experience in:

• Accounting.
• Real estate.
• Insurance.
• Restaurants.
• Healthcare administration.
• Legal services.
• Home services.
• E-commerce.
• Recruiting.
• Sales operations.

Industry knowledge helps you recognize problems outsiders don't even know exist.

What If You're Starting From Zero?

Don't begin by designing an agency logo.

Begin by proving you can solve something.

1. Choose one business problem. Avoid trying to automate everything.
2. Learn the minimum tools needed to solve it.
3. Build a working demonstration.
4. Test failure cases. Find out what happens when inputs are incomplete or the AI gives an unexpected response.
5. Talk to people who actually experience the problem.
6. Find out whether solving it has financial value.
7. Get one real customer before worrying about scaling.
8. Document the result. A genuine before-and-after case study is more persuasive than another certificate saying you completed an AI course.

If nobody will pay for the first solution, that's useful information. Change the offer before spending months trying to scale it.

Is an AI Automation Agency Worth Starting?

It can be worth testing if you enjoy solving business problems, are willing to learn technical concepts and can tolerate the uncomfortable work of finding clients.

It is less attractive if what interests you most is the promise of passive income.

Good Reasons to Try It	Bad Reasons to Try It
You understand a particular industry	A YouTuber says everyone is making $10K/month
You enjoy solving workflow problems	You want passive income with little client interaction
You are willing to learn integrations and troubleshooting	You think ChatGPT will do all the technical work
You can identify a measurable business problem	You want to sell "AI" without knowing what problem it solves
You are comfortable selling or developing that skill	You expect automated cold email to find all your customers

Bottom Line: Can You Really Make Money With an AI Automation Agency?

Yes. There is credible evidence that businesses are spending increasing amounts of money on AI integration, agents and workflow automation.

Upwork's marketplace data shows strong growth in AI-related freelance work, including particularly rapid growth in integration and automation. Fiverr's marketplace data independently shows businesses actively searching for specialists who can implement AI agents and automation.

That's considerably better evidence than a TikTok video showing a Stripe dashboard.

But growing demand does not make this an easy business.

The tools are becoming easier to use, which means the barrier to entry is falling for your competitors too. The durable advantage is increasingly likely to come from understanding a business, identifying valuable problems, building reliable solutions and earning enough trust that someone will pay you to implement them.

The opportunity is real. The "easy money" version is the questionable part. If you approach an AI automation agency as a genuine consulting and technical-services business, it can make money. If you approach it as a shortcut to passive income because someone showed you a $20,000 revenue screenshot, your odds probably look very different.

Explore More AI Side Hustle Guides

• AI Side Hustles That Actually Pay: A Realistic Guide
• The Best Free AI Tools for Education and Side Hustles
• What Jobs Will Get Replaced by AI?
• Why AI Hasn't Taken Your Job Yet—and What Comes Next

Frequently Asked Questions

Can you actually make money with an AI automation agency?

Yes. Freelance marketplace data shows businesses are spending money on AI integration, automation and AI-agent expertise. However, that proves a market exists; it does not mean every new agency will find clients or become profitable.

How much can a beginner make with an AI automation agency?

There is no reliable typical-income figure for beginners. Some may earn nothing, others may sell occasional projects, and experienced specialists can build substantial businesses. Be skeptical of income claims that don't disclose expenses, experience, customer acquisition and how representative the result is.

Can I start an AI automation agency with no coding experience?

Simple workflows can be built with no-code and low-code tools, so professional programming experience isn't mandatory for every project. However, understanding APIs, data, troubleshooting, security and error handling becomes increasingly valuable as client projects become more complicated.

What AI automation services can I sell?

Examples include customer-support automation, lead qualification, CRM workflows, document processing, appointment intake, internal knowledge assistants, email triage, reporting and appropriate voice-agent applications. The best service is usually one tied to a measurable business problem.

Is an AI automation agency passive income?

Usually not. Client acquisition, implementation, testing, support, troubleshooting and maintaining integrations require work. Recurring revenue is possible, but it generally comes with recurring responsibilities.

Do I need an expensive AI automation course?

No course can guarantee customers or income. Free documentation, tutorials and hands-on experimentation can teach many of the technical basics. A course may be useful if it provides structured learning, but the important test is whether you can build a reliable solution that a real customer values.

Is the AI automation agency market already saturated?

The barrier to offering generic AI automation services is increasingly low, so competition is real. Specialized knowledge can create differentiation. Someone who understands both AI automation and a particular industry's workflows may be better positioned than another general-purpose "AI agency."

What's the biggest mistake beginners make?

Building an AI solution before confirming that a customer has a sufficiently valuable problem. Start with the business problem, not the AI tool.

Income disclaimer: This article discusses business trends and potential opportunities for informational purposes. It does not guarantee income, clients or profitability. Business results vary substantially based on skills, experience, market conditions, expenses and execution.

Which Medical Specialties Are Safest From AI?

If you are choosing a medical specialty and wondering which ones are safest from AI, the reassuring answer is that AI is much more likely to change doctors' work than eliminate most doctors. Specialties built around hands-on procedures, unpredictable emergencies, complex patient relationships and physical examination generally have more protection from full automation. Psychiatry, family medicine, emergency medicine, surgery and procedure-heavy specialties are among the stronger candidates, while image- and data-intensive fields such as radiology and pathology are likely to experience some of the deepest AI-driven workflow changes.

But "most affected by AI" and "most likely to disappear" are not the same thing. Radiology is a perfect example: it is already one of medicine's biggest AI deployment areas, yet radiologists remain responsible for integrating findings, recognizing errors, communicating with clinicians and patients, and making consequential medical judgments.

Short answer: The safest medical careers are generally those in which the physician must combine human interaction, physical examination, procedures, unpredictable situations and legal or clinical responsibility. AI can automate tasks inside these specialties without automating the physician.

Table of Contents

• Medical Specialties Safest From AI: Quick Comparison
• What Makes a Medical Specialty Hard for AI to Replace?
• Psychiatry
• Family Medicine
• Emergency Medicine
• Surgery
• Obstetrics and Gynecology
• Physical Medicine and Rehabilitation
• Internal Medicine
• Is Ophthalmology Safe From AI?
• Is Dermatology Safe From AI?
• Is Radiology at High Risk From AI?
• What About Pathology?
• Which Medical Specialties Will Be Most Affected by AI?
• Will AI Eventually Replace Doctors?
• Should Medical Students Choose a Specialty Based on AI Risk?
• The Safest Doctor May Be the One Who Knows How to Use AI
• Bottom Line
• Frequently Asked Questions

Medical Specialties Safest From AI: Quick Comparison

Medical Specialty	Relative AI Replacement Risk	Why
Psychiatry	Low	Relationship, nuanced communication, behavioral observation and complex judgment remain central
Family Medicine	Low	Broad diagnostic work, physical exams, continuity of care and highly varied patients
Emergency Medicine	Low	Unpredictable cases, physical intervention, rapid decisions and team coordination
Surgery	Low	Physical procedures, anatomy, complications and real-time decision-making
Obstetrics & Gynecology	Low	Procedures, examinations, childbirth and unpredictable emergencies
Physical Medicine & Rehabilitation	Low–Moderate	Physical assessment, functional goals and individualized rehabilitation
Internal Medicine	Low–Moderate	Complex patients and diagnostic reasoning, although many information tasks can be automated
Dermatology	Moderate	Image recognition is AI-friendly, but procedures, biopsies and clinical context remain human-intensive
Ophthalmology	Moderate	AI can screen images, while surgery and procedural care remain difficult to automate
Pathology	Moderate–High workflow impact	Digital image analysis is highly compatible with AI, but difficult diagnoses and responsibility remain with physicians
Radiology	High workflow impact	AI is exceptionally suited to image analysis, triage and measurements, but this does not mean radiologists are disappearing

Important: These categories describe relative exposure to automation of medical tasks, not a prediction that physicians in a particular specialty will lose their jobs.

What Makes a Medical Specialty Hard for AI to Replace?

Instead of asking whether AI is "smart enough" to replace a doctor, it is more useful to break the doctor's job into tasks.

A specialty tends to be harder to automate when several of these characteristics occur together:

• Physical procedures: The doctor must manipulate tissue, instruments or the patient's body.
• Unpredictability: Conditions can change quickly and require adaptation rather than a predefined workflow.
• Physical examination: Diagnosis depends partly on touch, movement, appearance and interaction with the patient.
• Human relationships: Trust, persuasion, empathy and understanding a patient's circumstances materially affect care.
• Complex multimodal judgment: The physician combines laboratory results, imaging, history, examination and subtle contextual clues.
• Accountability: Someone must ultimately take responsibility for consequential clinical decisions.
• Procedural skill: Knowing what should be done is different from physically performing it safely.

By contrast, tasks become more attractive targets for AI when the inputs and outputs are already digital and standardized. Reading images, classifying patterns, generating documentation, measuring structures and searching large amounts of medical information are obvious examples.

1. Psychiatry: One of the Hardest Specialties to Fully Automate

Psychiatry might initially seem vulnerable because generative AI can already conduct remarkably natural conversations. AI chatbots can provide information, ask questions and simulate supportive dialogue.

That does not make them psychiatrists.

A psychiatrist evaluates much more than a patient's words. Tone, behavior, history, inconsistencies, family circumstances, medication response, risk, substance use and changes over time can all matter.

Serious cases may also involve suicidal risk, psychosis, mania, substance dependence or patients who cannot accurately describe their own condition. Responsibility for these decisions is fundamentally different from operating a conversational chatbot.

The American Medical Association continues to emphasize that AI chatbots can complement healthcare information but should not replace physician guidance.

What AI will probably change: documentation, screening, symptom questionnaires, patient education, administrative work and clinical decision support.

What remains difficult to replace: therapeutic relationships, nuanced diagnosis, medication management, risk assessment and responsibility for complex psychiatric care.

2. Family Medicine: Broad, Messy and Very Human

Family medicine has an important form of protection from automation: patients rarely arrive as clean datasets.

A family physician may move from evaluating abdominal pain to managing diabetes, discussing depression, examining a rash and adjusting blood-pressure medication within a single morning.

The doctor also knows something an algorithm may not easily capture: the patient's history over years.

The American Academy of Family Physicians is actively developing AI initiatives, but its approach illustrates the likely direction of the technology. The organization describes AI as a way to reduce administrative burdens and allow family physicians to spend more time caring for patients—not as a substitute for family physicians.

Likely AI role: documentation, inbox management, chart summaries, preventive-care reminders, preliminary decision support and administrative automation.

Why physicians remain important: physical examinations, continuity, multimorbidity, ambiguous symptoms and the enormous variety of primary-care presentations.

3. Emergency Medicine: AI Doesn't Control the Emergency Room

Emergency medicine combines nearly every characteristic that makes complete automation difficult.

Patients may arrive unconscious, intoxicated, bleeding, confused or unable to provide an accurate history. Several emergencies may happen simultaneously. A patient's condition can deteriorate within minutes.

AI can become extremely valuable in this environment. It can help prioritize imaging, identify warning patterns, summarize records and support diagnostic decisions.

But deciding what to do with an unstable patient while coordinating nurses, consultants, family members, imaging, laboratory testing and procedures is a very different problem from generating a diagnosis from a dataset.

Replacement risk: relatively low.

Task-automation potential: high.

That distinction is going to become increasingly important throughout medicine.

4. Surgery: Knowing the Answer Isn't the Same as Performing the Operation

Surgery has strong protection because it exists in the physical world.

AI can analyze scans, recommend surgical plans, identify anatomy and assist robotic systems. Surgical robots can provide extraordinary precision.

But today's surgical robots generally do not independently decide that a patient needs surgery, obtain consent, manage an unexpected hemorrhage and complete an unpredictable operation without a surgical team.

Even increasingly capable robotic systems must contend with biological variability. Human bodies do not behave like identical manufactured components.

Some parts of surgery will undoubtedly become more automated. The surgeon of the future may operate with far more AI assistance than the surgeon of today.

That is not the same as eliminating surgeons.

5. Obstetrics and Gynecology

OB-GYN combines diagnosis, longitudinal care, physical examinations, procedures, surgery and unpredictable emergencies.

Childbirth is a particularly difficult environment for complete automation. Conditions can change rapidly, and physicians may have to make consequential decisions involving both mother and baby.

AI may become increasingly useful for fetal monitoring, imaging, risk prediction, documentation and clinical decision support. But those capabilities are more likely to augment obstetricians than eliminate the specialty.

6. Physical Medicine and Rehabilitation

Physical medicine and rehabilitation is another relatively resistant field because the physician is evaluating function rather than simply interpreting digital information.

Movement, pain, strength, mobility, disability, recovery goals and a patient's living environment all matter.

Wearable sensors, computer vision and AI-assisted rehabilitation could dramatically improve monitoring and treatment planning, but human assessment and individualized goals remain important.

7. Internal Medicine

Internal medicine is difficult to rank because it contains both highly automatable information work and extremely complicated human decision-making.

AI may become excellent at summarizing charts, suggesting differential diagnoses, checking drug interactions and identifying patterns across laboratory results.

But internists often care for patients with several diseases simultaneously. The technically "best" treatment for one disease may make another worse.

Choosing among competing priorities—and understanding what matters to the patient—is much harder than answering an isolated medical question.

Is Ophthalmology Safe From AI?

Ophthalmology illustrates why a specialty cannot be classified simply as safe or unsafe.

AI is well suited to analyzing standardized retinal and other ophthalmic images. Screening and detection tasks are therefore attractive targets for automation.

But ophthalmology also contains substantial procedural and surgical work.

An ophthalmologist whose work is heavily procedural may have a very different automation profile from one whose workload is dominated by screening and image interpretation.

Think about subspecialties, not just specialties. Two physicians carrying the same broad specialty label can perform very different jobs. Procedure-heavy subspecialties generally have stronger protection from complete automation than work dominated by standardized digital interpretation.

Is Dermatology Safe From AI?

Dermatology has significant exposure to computer vision because skin lesions can be photographed and analyzed by image-recognition systems.

That makes certain screening and classification tasks technically attractive for AI.

But a dermatologist's job also includes taking histories, examining the entire patient, deciding whether a lesion needs biopsy, performing procedures, interpreting pathology in context and managing chronic disease.

AI may reduce the amount of routine visual classification performed without assistance. It is much less obvious that it eliminates dermatologists.

Is Radiology at High Risk From AI?

Radiology is probably the specialty most frequently mentioned in discussions about doctors being replaced by AI—and for understandable reasons.

Medical images are digital, there are enormous datasets available for training, and many radiological tasks involve pattern recognition.

The FDA's current list of authorized AI-enabled medical devices demonstrates just how heavily medical AI development is concentrated in radiology. Numerous recently authorized systems involve radiological imaging, including image analysis, reconstruction, measurements and triage.

But the conclusion that AI therefore eliminates radiologists does not follow.

The American College of Radiology has emphasized human oversight and continuous monitoring of imaging AI. In 2026, the ACR also approved its first practice parameter specifically addressing the implementation and monitoring of imaging AI.

The more realistic scenario is that radiologists become heavy users and supervisors of AI.

Routine measurements, prioritization and some detection tasks may become increasingly automated. Radiologists may spend proportionally more time on difficult cases, integrating multiple studies, procedures, consultation and validating AI output.

Radiology may be among the specialties most changed by AI without being among the first specialties eliminated by AI. High AI adoption and high job-replacement risk are not the same thing.

Read our related guide: AI in Radiology: Pros and Cons.

What About Pathology?

Pathology faces some of the same forces as radiology as laboratories adopt digital pathology.

Once slides become high-resolution digital images, AI can help identify patterns, count cells, quantify biomarkers and flag suspicious areas.

That makes portions of pathology highly automatable.

But difficult pathology cases require integration of morphology, clinical history, molecular testing and other laboratory findings. Pathologists also carry professional responsibility for diagnoses that can determine surgery, chemotherapy and other major treatments.

The likely future is therefore substantial workflow automation rather than a pathology department with no pathologists.

Which Medical Specialties Will Be Most Affected by AI?

If "affected" means the technology will perform a meaningful portion of today's work, the specialties with standardized digital information are obvious candidates.

Radiology, pathology, dermatology and parts of ophthalmology are particularly exposed because AI can analyze images and structured data at enormous scale.

But exposure can be positive as well as disruptive.

A radiologist who can review routine examinations faster with reliable AI assistance may become more productive. A pathologist could use AI to quantify features that would otherwise require tedious manual work. An ophthalmologist could use automated screening to identify patients who actually need specialist care.

Automation can therefore increase a specialty's capacity rather than simply reduce employment.

Will AI Eventually Replace Doctors?

Current evidence does not justify saying that physicians as a profession are on the verge of disappearing.

The American Medical Association's current framework explicitly describes healthcare AI as augmented intelligence: technology designed to enhance human intelligence rather than replace it. In June 2026, the AMA adopted additional policies calling for AI to remain under physician oversight in clinical decision-making.

The AMA's AI Specialty Collaborative now brings together 21 medical specialty societies to help shape how AI is incorporated into healthcare.

That doesn't guarantee today's physician workforce will remain unchanged.

AI could increase productivity enough that some tasks require fewer physician hours. Certain services may shift toward primary care or non-physician clinicians supported by AI. Documentation and administrative staffing could shrink. Some specialties could experience changes in demand.

But that is considerably different from an autonomous AI replacing the entire physician.

For a deeper discussion, see How Long Until AI Replaces Doctors?.

Should Medical Students Choose a Specialty Based on AI Risk?

AI risk deserves consideration, but it should probably not determine your entire career.

A student entering medical school today could practice for decades. Predicting exactly what an individual specialty will look like that far into the future is impossible.

A more durable strategy is to ask:

• Do I actually enjoy this specialty?
• Does it involve work I am good at?
• How much of the job consists of standardized digital tasks?
• How much involves procedures or physical examination?
• How important are long-term patient relationships?
• Could AI make this specialty more productive rather than obsolete?
• Am I willing to become good at working with AI?

The last question may ultimately matter most.

The Safest Doctor May Be the One Who Knows How to Use AI

The competition may not ultimately be "doctor versus AI."

It may be:

a physician using AI effectively versus a physician who refuses to use it.

Doctors who learn how to verify AI output, recognize its failure modes and integrate useful tools into clinical practice may gain a substantial advantage.

The AMA's current AI evaluation framework emphasizes exactly these issues, including clinical relevance, validation, risks, effectiveness, workflow integration and ongoing monitoring.

Medicine has absorbed disruptive technologies before. Electronic health records, advanced imaging, robotic surgery and molecular diagnostics changed what physicians do without eliminating the need for physicians.

AI could be a much larger transformation, but the same principle may apply.

Bottom Line

If your definition of "safe from AI" means a specialty in which no tasks will be automated, there probably isn't one.

If it means specialties where eliminating the physician remains especially difficult, fields combining procedures, physical interaction, unpredictable situations, patient relationships and high-stakes judgment have significant advantages.

Psychiatry, family medicine, emergency medicine, surgery and OB-GYN are among the stronger examples.

Radiology, pathology, dermatology and ophthalmology may experience more direct automation of specific diagnostic tasks, but that should not automatically be interpreted as those specialties disappearing.

The safest career strategy may therefore be less about finding a specialty untouched by artificial intelligence and more about choosing a specialty you want to practice while becoming exceptionally good at using the AI tools that will inevitably become part of it.

Explore more:

• AI in Healthcare: Uses, Risks and What Comes Next
• Will AI Be Able to Diagnose Patients?
• How Long Until AI Replaces Doctors?
• What Doctor Specialties Will Get Automated?
• AI in Radiology: Pros and Cons

Frequently Asked Questions

What medical specialty is safest from AI?

There is no objectively AI-proof specialty. Psychiatry, family medicine, emergency medicine, surgery and other procedure- or relationship-intensive specialties are relatively difficult to automate completely because they require physical interaction, unpredictable decision-making and human responsibility.

Which doctor specialties are most likely to be affected by AI?

Radiology, pathology, dermatology and ophthalmology are likely to experience substantial AI-driven changes because important parts of their work involve analyzing digital images and structured data. That does not mean these physicians will necessarily be replaced.

Will AI replace radiologists?

AI is already changing radiology and many authorized medical AI systems involve imaging. A more plausible near- and medium-term future is radiologists working with increasingly capable AI systems rather than radiology operating without physicians.

Is surgery safe from AI?

Surgery is relatively resistant to full automation because it requires physical procedures and real-time responses to unexpected events. AI and robotics are nevertheless likely to automate or assist parts of surgical planning and procedures.

Is psychiatry safe from AI?

Psychiatry is relatively difficult to automate completely. AI can support screening, documentation and patient education, but complex diagnosis, therapeutic relationships, medication decisions and risk assessment continue to require substantial human judgment.

Should I avoid radiology because of AI?

AI risk alone is not a strong reason to avoid a specialty you otherwise want to practice. Radiology is likely to change substantially, but radiologists are also positioned to become some of medicine's most sophisticated users and supervisors of AI.

Will AI reduce the number of doctors needed?

It is possible that higher productivity could change physician demand in particular tasks or specialties, but healthcare demand, aging populations, regulation, access to care and the creation of new services also influence employment. There is no reliable formula for translating AI capability into a future number of physician jobs.

What skills will help doctors survive the AI transition?

Clinical judgment, communication, procedures, understanding AI limitations, recognizing incorrect outputs and knowing when not to rely on automation are likely to become increasingly valuable. Doctors who can combine medical expertise with effective AI use may have an advantage.

Career note: This article discusses technology and employment trends and is not individualized career or educational advice. AI capabilities and medical practice are changing rapidly, and no ranking can guarantee the future demand for a particular specialty.

Can AI Really Do Your Taxes? Simple Returns, Complex Cases and the Risks

AI can already help prepare a straightforward tax return, but that does not mean you should hand your financial life to a chatbot. Tax software can import forms, perform calculations, identify possible deductions and guide users through common filing situations. The danger begins when generative AI is asked to interpret complicated tax law, make assumptions about missing information or recommend aggressive positions without understanding the full facts. Simple returns are moving toward greater automation. Complicated returns still require verification, professional judgment and someone who can be held accountable when the answer is wrong.

Table of Contents

• Can AI Really Prepare Your Taxes?
• Tax Software and AI Chatbots Are Not the Same
• What Counts as a Simple Tax Return?
• What Makes a Tax Return Complicated?
• What AI Can Do Well
• Where AI Tax Preparation Can Fail
• Who Pays When AI Gets Your Taxes Wrong?
• The Tax-Data Privacy Problem
• Will AI Replace Tax Preparers?
• How to Use AI for Taxes More Safely
• What AI Tax Preparation May Look Like Next
• The Verdict
• Frequently Asked Questions

Can AI Really Prepare Your Taxes?

AI can assist with tax preparation, and software can already complete much of the mechanical work involved in many individual returns. It can import tax documents, ask interview questions, calculate totals, transfer information between forms and electronically submit a completed return.

That is different from giving a general-purpose AI chatbot your income information and asking it to determine what you owe.

Tax preparation combines several different jobs:

• Collecting complete financial information
• Classifying income and expenses correctly
• Applying current federal and state tax rules
• Identifying missing forms and inconsistencies
• Calculating taxes, deductions and credits
• Choosing between legally supportable alternatives
• Signing and filing the return
• Responding if the return is questioned later

AI is increasingly capable at the first five tasks. It is much less dependable when facts are ambiguous, the law requires judgment or the position may need to be defended during an audit.

The realistic answer: AI can help complete a tax return. It cannot guarantee that the information you supplied was complete, that its interpretation was legally correct or that the IRS will accept the position it recommended.

Tax Software and AI Chatbots Are Not the Same

Much of the confusion surrounding AI tax preparation comes from treating every form of tax technology as though it works the same way.

Type of Tool	How It Works	Primary Risk
Traditional tax software	Uses programmed tax rules, form instructions and guided questions to prepare a return	Incorrect user input, unsupported situation or misunderstood interview question
AI-enhanced tax software	Adds document recognition, natural-language explanations, recommendations and error detection	AI-generated advice may sound authoritative even when it misses an exception
General-purpose chatbot	Generates answers from broad training data and any information provided by the user	Hallucinated rules, outdated information, missing context and serious privacy exposure
Professional tax platform with AI	Helps a CPA, enrolled agent or tax attorney research, organize and review a client's information	The professional may rely too heavily on the output or fail to independently verify it

Traditional tax software is built around forms, calculations and tax rules. Generative AI produces language and recommendations. It predicts what an appropriate answer should look like, but prediction is not the same as proving that a tax position is correct.

The Taxpayer Advocate Service has warned that AI assistants may struggle to interpret complex tax laws or account for unique circumstances. It recommends that taxpayers not rely solely on AI-generated tax advice.

Do not confuse a confident explanation with a correct tax determination. A chatbot can describe a deduction clearly while overlooking an income limit, filing-status restriction, holding-period rule, state-law difference or special exception that changes the result.

What Counts as a Simple Tax Return?

A relatively simple return generally involves clearly documented income and few unusual decisions. Examples may include:

• One or two Form W-2 wage statements
• Limited bank interest reported on Form 1099-INT
• No business or rental activity
• No cryptocurrency transactions
• No foreign income or foreign accounts
• No complicated stock sales
• No multiple-state filing requirement
• Using the standard deduction
• No disputed dependent or filing-status issue

Guided software is well suited to this type of return because the taxpayer is mainly transferring information from standardized documents into standardized forms.

Even apparently simple returns can contain traps. Marriage, divorce, a new child, marketplace health insurance, unemployment benefits, college expenses, a home purchase or a move between states can introduce rules that are easy to overlook.

A short return is not automatically a simple return. A taxpayer may have only a few forms but still face a difficult question about dependency, residency, filing status, tax credits or whether income belongs on the return.

What Makes a Tax Return Complicated?

A return becomes complicated when the software must do more than transfer clearly identified numbers. Complexity grows when the correct treatment depends on facts, documentation, elections, estimates or interpretation.

Self-Employment and Gig Work

A self-employed taxpayer must identify business income, separate personal and business expenses, evaluate home-office eligibility, calculate depreciation and pay self-employment tax.

An AI assistant may identify possible deductions but cannot know whether an expense was genuinely ordinary, necessary and properly documented.

Rental Property

Rental returns can involve depreciation, repairs versus improvements, passive-activity limitations, personal-use days, security deposits and the allocation of shared expenses.

A wrong classification may affect several future returns, not only the year in which the mistake was made.

Stocks, Options and Cryptocurrency

Investment returns may involve missing cost basis, wash sales, employee stock compensation, option exercises, restricted stock, cryptocurrency exchanges and transactions spread across several platforms.

AI cannot calculate a trustworthy gain when the source records are incomplete or inconsistent.

Multiple States

Living in one state, working in another or moving during the year can create residency, allocation and tax-credit questions. State rules do not always follow federal treatment.

Foreign Income and Accounts

Foreign wages, pensions, bank accounts, investments, businesses and gifts may trigger specialized reporting requirements. Penalties for missing certain international information returns can be severe even when little or no additional tax is owed.

Business Entities

Partnerships, S corporations, C corporations and multi-member businesses require decisions about compensation, distributions, basis, ownership allocations and transactions between the owner and the business.

A general chatbot should not be treated as the final authority for these returns.

Estates, Trusts and Inheritances

The treatment of inherited property may depend on basis adjustments, valuation dates, trust terms, distributions and the type of income received.

IRS Notices, Audits and Amended Returns

Once the IRS questions a return, the issue is no longer merely data entry. The taxpayer may need to reconstruct records, interpret the notice, identify the legal issue and present evidence supporting the reported position.

Tax Situation	AI Assistance Level	Human Review
W-2 income and standard deduction	Strong	Review entries before filing
Common interest and dividend income	Strong when forms are complete	Check imported amounts and account ownership
Common education or dependent credits	Moderate	Verify eligibility rules and supporting records
Basic sole-proprietor income and expenses	Moderate	Review classifications and deductions carefully
Rental property and depreciation	Limited without complete history	Professional review strongly advisable
Stock options or complicated investments	Limited	Specialist review may be necessary
Foreign income, accounts or entities	High risk	Use a qualified international-tax professional
Partnership or corporate return	Useful as an assistant	Do not rely on unsupervised chatbot preparation
Audit, appeal or disputed tax position	Research and organization only	Qualified representation may be essential

What AI Can Do Well

Import and Organize Tax Documents

AI can extract information from W-2s, 1099s, receipts and statements. It can classify documents, identify duplicates and flag apparently missing fields.

Perform Calculations

Established tax software is generally effective at mathematical calculations when the correct information has been entered into the correct fields.

The greater risk is not arithmetic. It is whether the taxpayer or software selected the correct tax treatment before performing the calculation.

Ask Follow-Up Questions

An AI-guided interview can adapt its questions based on earlier answers. This may help taxpayers identify forms or credits they did not know existed.

Explain Tax Terms

AI can translate technical instructions into plain language and explain terms such as adjusted gross income, tax credits, basis and depreciation.

These explanations are useful for education but should be checked against current IRS instructions when they affect the actual return.

Identify Inconsistencies

AI can flag a dependent whose age appears inconsistent, expenses that differ significantly from prior years or income documents that do not match imported records.

Draft Questions for a Tax Professional

AI can help organize a situation and create a list of questions to discuss with a CPA, enrolled agent or tax attorney. This can make professional consultations more efficient.

Support Professional Review

Within a controlled tax practice, AI can summarize documents, research authorities, compare tax treatments and identify transactions requiring closer attention.

The safest role for tax AI is assistant, not decision-maker. Use it to organize, calculate, explain and flag issues. Do not assume it can independently determine every relevant fact or defend the return later.

Where AI Tax Preparation Can Fail

It Can Invent Tax Rules

Generative AI can produce a nonexistent deduction, misstate an income limit or cite an authority that does not support its conclusion. This behavior is known as an AI hallucination.

Because tax explanations often contain technical language and numbers, an invented answer can look convincing enough to escape casual review.

It May Use Outdated Information

Tax laws, thresholds, forms and filing procedures change. An answer that was correct for one filing year may be wrong for the next.

The model may also mix federal rules with a state rule or apply a new provision to a year before it became effective.

It Does Not Know What You Forgot to Mention

AI only knows what it can access. If you omit a cash payment, foreign account, cryptocurrency wallet, prior depreciation schedule or important life event, it may prepare a logically consistent return from incomplete facts.

It Can Misunderstand Ambiguous Facts

Whether a worker is an employee or independent contractor, whether an activity is a business or hobby, and whether a person qualifies as a dependent can require a detailed factual analysis.

A small factual difference may change the legal result.

It May Recommend the Largest Refund Instead of the Most Defensible Return

Users naturally prefer an answer that reduces tax or increases a refund. An overly agreeable AI system may reinforce the interpretation the user wants instead of challenging it.

This creates particular risk when the suggested deduction or credit depends on facts that have not been verified.

It Cannot Inspect Original Evidence

A chatbot may accept a summarized description without recognizing that receipts, mileage records, contracts or ownership documents do not support it.

It Cannot Promise the IRS Will Agree

Tax law contains uncertain and disputed areas. A valid analysis may require comparing authorities, documenting assumptions and understanding the taxpayer's tolerance for audit risk.

Documented warning: The Taxpayer Advocate Service reported on an informal review in which tax-company chatbots initially gave inaccurate or irrelevant answers to as many as half of 16 complex tax questions. That was a small test, not a universal error rate, but it demonstrates why polished chatbot answers should not be treated as binding tax advice.

Who Pays When AI Gets Your Taxes Wrong?

The uncomfortable answer is usually the taxpayer.

The IRS states that taxpayers are ultimately accountable for the accuracy of every item reported on their returns. This remains true when a paid professional prepares the return.

A reputable paid preparer must generally:

• Sign the return
• Include a valid preparer tax identification number
• Exercise applicable due diligence
• Provide the taxpayer with a copy
• Ask reasonable questions when information appears incomplete or inconsistent

A general-purpose AI chatbot does none of these things. It does not sign the return, hold a professional license or accept representation responsibilities.

Possible Costs of an Incorrect AI-Prepared Return

• Additional tax
• Interest on the unpaid amount
• Accuracy-related or other penalties
• Delayed refunds
• Loss or repayment of credits
• Cost of filing an amended return
• Professional fees to correct the problem
• Time spent responding to notices or an examination

Some commercial software offers calculation guarantees, but those guarantees vary. They may not cover incorrect facts supplied by the user, unsupported deductions, misunderstood interview questions or advice generated outside the protected filing product.

The liability problem will slow fully autonomous tax AI. Producing an answer is easy compared with deciding who is financially and professionally responsible when that answer creates an audit, penalty or missed reporting obligation.

The Tax-Data Privacy Problem

A complete tax return can contain nearly everything an identity thief needs:

• Social Security numbers
• Birth dates
• Home addresses
• Employer information
• Bank account and routing numbers
• Investment account details
• Income and business records
• Information about children and other dependents

Uploading these documents to an unapproved public chatbot can create serious privacy and security risks. Do not assume that every AI service provides the protections required of professional tax-preparation systems.

Questions to Ask Before Uploading Tax Information

• Is the service specifically designed for tax preparation?
• Is the data encrypted during transmission and storage?
• Will submitted information be used to train an AI model?
• Can employees or contractors view the data?
• Is information shared with third parties?
• Can the data be permanently deleted?
• What happens if the company suffers a breach?
• Does the service explain where data is stored?

Tax professionals have legal and regulatory obligations to safeguard client information. Federal rules also restrict how return preparers may use or disclose tax-return information.

Never paste an unredacted tax return, W-2, Social Security number, bank account number or identity document into a general public chatbot. For general questions, remove names, account numbers, addresses and every other identifying detail.

Will AI Replace Tax Preparers?

AI will reduce the amount of manual tax preparation, particularly for straightforward returns. It may also reduce demand for preparers whose main service is transferring numbers from familiar documents into standard forms.

It is less likely to replace professionals handling:

• Tax planning before a transaction occurs
• Business structures and owner compensation
• Multi-state and international tax
• Estates and trusts
• Partnership and corporate returns
• Tax audits, appeals and collections
• Disputed classifications or valuations
• Representation before the IRS
• Situations with incomplete or conflicting records

The IRS notes that attorneys, CPAs and enrolled agents can represent taxpayers before the agency in audits, collections and appeals. A chatbot does not possess those representation rights.

The tax preparer's job is therefore likely to shift from entering data toward reviewing automated work, identifying missing facts, evaluating risk and defending conclusions.

This is part of the broader change covered in our guide to AI, accountants and the future of finance jobs.

Tax Work	Likely Direction
Entering standard tax forms	Increasingly automated
Routine calculations	Already largely automated
Explaining common tax terms	AI-assisted
Detecting missing documents	Increasingly AI-assisted
Complex tax research	AI-assisted with professional verification
Choosing a defensible position in a grey area	Human professional remains responsible
Client representation during an audit	Requires an authorized person
Accepting legal and professional accountability	Human or regulated firm remains necessary

How to Use AI for Taxes More Safely

1. Use Tax-Specific Software

Use an established tax-preparation product rather than asking a general chatbot to create a finished return from raw documents.

2. Verify the Filing Year

Confirm that every answer, threshold, form and instruction applies to the tax year being filed—not merely the current calendar year.

3. Compare the Answer With an Official Source

Check important conclusions against current IRS forms, instructions, publications or other authoritative guidance.

4. Do Not Upload Sensitive Documents to Public AI

Redact names, Social Security numbers, addresses, account information and employer identifiers before using AI for general educational questions.

5. Ask What Facts Could Change the Answer

Instead of asking only, “Can I deduct this?” ask the AI to list every eligibility requirement, exception and missing fact that could change its conclusion.

6. Require Source Identification

Ask for the relevant form instructions, IRS publication, regulation or code section. Then open and read the cited material because AI can generate incorrect citations.

7. Review the Entire Return

Check names, Social Security numbers, filing status, dependents, income, bank information, credits and deductions before authorizing electronic filing.

8. Escalate Complex Situations

Use a qualified professional when the return involves a business entity, foreign assets, substantial investments, unusual compensation, rental depreciation, an IRS dispute or another high-risk issue.

9. Keep Supporting Records

An AI explanation does not prove a deduction. Retain the documents, receipts, logs, calculations and other evidence supporting the amounts reported.

10. Choose Someone Who Can Help Later

When hiring a preparer, consider whether that person will remain available if the IRS asks how the return was prepared.

What AI Tax Preparation May Look Like Next

Future tax systems are likely to become more automated, but the change will probably occur in stages rather than through one chatbot suddenly replacing every tax professional.

Automatic Document Collection

With permission, systems may collect wage statements, investment records, accounting data and prior-year information directly from approved sources.

Continuous Tax Estimates

Individuals and businesses may receive updated tax projections throughout the year instead of discovering the result during filing season.

AI-Generated Draft Returns

Software may prepare a complete draft, list assumptions and request supporting documentation for uncertain items.

Risk Scoring and Audit Warnings

Systems may compare the return with prior years, information reports and common examination issues to identify entries that deserve additional review.

Professional Review by Exception

Instead of preparing every line manually, tax professionals may review the small percentage of transactions that software flags as unusual, unsupported or legally ambiguous.

Limited Autonomous Filing

For highly standardized returns, taxpayers may eventually approve a nearly complete return produced from verified data. Complicated returns will remain harder to automate because the system must identify uncertain facts and choose positions that someone may later need to defend.

The likely model: Software prepares the draft, AI explains and checks it, the taxpayer confirms the facts, and a qualified professional reviews high-risk cases. That is less dramatic than a robot replacing every tax preparer—but far more realistic.

The Verdict

AI can genuinely help prepare taxes. For a taxpayer with complete wage documents, limited investment income and no unusual circumstances, modern software can already perform most of the mechanical work.

But complicated taxation is not merely a calculation problem. It is a fact-finding, classification, documentation and legal-judgment problem.

AI does not automatically know what the taxpayer forgot to disclose. It may not recognize that a small factual distinction changes the rule. It can confidently invent an authority, apply an outdated threshold or recommend a favorable position without asking whether the taxpayer can prove it.

The honest conclusion: AI will increasingly handle simple tax preparation and assist professionals with complicated returns. It should not be trusted as the sole decision-maker for complex tax issues until it can reliably identify uncertainty, verify every material fact and accept meaningful responsibility for the consequences of an incorrect return.

Until then, the safest approach is to use AI for organization, explanations and preliminary review while relying on tax-specific software, official instructions and qualified professionals for the final return.

Frequently Asked Questions

Can ChatGPT prepare my tax return?

ChatGPT can explain tax concepts, organize information and help identify questions, but a general chatbot should not be used as the sole preparer of a federal or state tax return. It may use outdated information, misunderstand your facts or generate an incorrect rule or citation.

Can AI file taxes directly with the IRS?

Approved tax software can transmit electronic returns after the taxpayer reviews and signs them. A general AI chatbot does not independently sign and file a return on the taxpayer's behalf.

What tax returns can AI handle most easily?

AI-assisted tax software is best suited to returns with complete standardized forms, wage income, limited interest or dividends, no complicated business activity and few unusual deductions or credits.

What tax situations should not rely on AI alone?

Use professional review for partnerships, corporations, foreign income, multi-state issues, rental depreciation, stock options, significant cryptocurrency activity, estates, trusts, audits and other situations requiring interpretation or specialized reporting.

Who is responsible if AI makes a tax mistake?

The taxpayer is ultimately accountable for the information reported on the return. A paid preparer may also have professional and legal responsibilities, but a general chatbot does not accept those obligations.

Is it safe to upload tax documents to an AI chatbot?

Do not upload unredacted tax documents to a general public chatbot. Tax records contain Social Security numbers, addresses, income information and bank details that could cause serious harm if exposed or misused.

Will AI replace CPAs and tax preparers?

AI will automate more data entry, calculations and standard-return preparation. Professionals will remain important for tax planning, complicated returns, disputed issues, representation and decisions requiring accountable judgment.

How can I check AI-generated tax advice?

Identify the filing year, locate the relevant IRS form instructions or publication, verify any cited authority and consult a qualified professional when the answer depends on complicated or disputed facts.

Sources and Methodology

This article distinguishes established tax-preparation software from generative AI chatbots. It does not assume that every automated calculation is artificial intelligence or that every tax question carries the same level of risk.

• Taxpayer Advocate Service: Is AI-Generated Tax Advice Making the Grade?
• IRS: Warning About AI-Generated Tax Answers and Misleading Advice
• IRS Topic 254: How to Choose a Tax Return Preparer
• IRS: Guided Tax Software and Electronic Filing Options
• IRS: Rules Governing the Use and Disclosure of Tax-Return Information
• IRS Publication 4557: Safeguarding Taxpayer Data
• IRS: Tax-Professional Credentials and Representation Rights

Continue Reading:

• Will AI Replace Accountants? Jobs at Risk and Growing
• What Is a Hallucination in AI?
• AI Job Replacement Risk Calculator
• Jobs AI Will Replace or Transform
• AI Ethics: Core Principles and Real Risks

Is ChatGPT Getting Worse? What the Data and Users Show

ChatGPT has not become uniformly worse, but it has changed enough that many longtime users are experiencing real differences. Newer models perform better on several reasoning, coding and professional benchmarks, yet some users find them more concise, less conversational, more restricted or less consistent with detailed instructions. The evidence points to a mixed conclusion: ChatGPT is becoming more capable overall, but not every update improves every task or preserves the qualities individual users valued most.

Table of Contents

• Is ChatGPT Actually Getting Worse?
• Why ChatGPT May Feel Worse
• What the Research and Benchmarks Show
• Common ChatGPT Complaints Examined
• Which Tasks Have Improved or Declined?
• How to Get Better Answers From ChatGPT
• When to Use Another AI Tool
• The Verdict
• Frequently Asked Questions

Is ChatGPT Actually Getting Worse?

ChatGPT is not getting worse in one simple, measurable direction. It is improving rapidly in some areas while changing or becoming less satisfying in others.

OpenAI released the GPT-5.6 model family on July 9, 2026, reporting substantial improvements in coding, professional knowledge work, scientific research, computer use and tool-based tasks. OpenAI's GPT-5.6 system card also reports slightly fewer factual errors than GPT-5.5 when tested on conversations that users had previously flagged for factual problems.

However, stronger benchmark scores do not guarantee that every user will prefer the new model. A person who mainly values warmth, conversational writing, long explanations or creative brainstorming may judge an update very differently from someone using ChatGPT for software development, data analysis or technical research.

The practical answer: ChatGPT is generally more capable than earlier versions, but model replacements, shorter default answers, changing personalities, stronger safeguards and inconsistent performance across tasks can make it feel worse for a particular user or workflow.

Why ChatGPT May Feel Worse

The Model You Liked May No Longer Be Available

ChatGPT is a service rather than one permanent model. OpenAI regularly introduces new models, changes defaults and retires older versions.

GPT-4o, GPT-4.1, GPT-4.1 mini and OpenAI o4-mini were retired from ChatGPT on February 13, 2026. OpenAI acknowledged that a subset of paying users preferred GPT-4o for its conversational warmth and creative ideation. The company said that feedback influenced personality and customization improvements in later GPT-5 models.

This helps explain why someone can reasonably say ChatGPT became worse even while newer models score higher on technical evaluations. The user's preferred experience may have depended on a particular model's tone, pacing, creativity or willingness to explore an idea.

Newer Models May Be More Concise

OpenAI described GPT-5.5 as providing smarter and more concise answers. Concision can be valuable when users want a fast answer, but it may feel like a downgrade when they expect complete code, a detailed article, a comprehensive analysis or step-by-step instructions.

A response that is technically correct but omits examples, background, edge cases or implementation details may score well in an evaluation while still disappointing the person who requested it.

Different Plans and Settings Can Produce Different Results

ChatGPT users may encounter different GPT-5.6 models, reasoning levels and usage limits depending on their subscription and selected settings. GPT-5.6 includes Sol, Terra and Luna tiers, and higher reasoning settings allow the model to spend more effort on difficult tasks.

This means two users can enter similar prompts and receive noticeably different levels of depth, reasoning or accuracy. A user may also receive a different experience after reaching a plan limit or changing a model setting.

Long Conversations Can Become Less Reliable

A long chat may contain outdated instructions, abandoned ideas, contradictory preferences and irrelevant details. As the conversation grows, the model must determine which information still matters.

When ChatGPT starts repeating itself, overlooking recent instructions or continuing an earlier approach that you no longer want, starting a fresh conversation with a clean summary can produce better results than continuing the old thread.

Safety Improvements Can Create Friction

AI companies continuously adjust safeguards in response to misuse, new risks and regulatory pressure. These safeguards are important, but they can occasionally interfere with legitimate research, fiction, cybersecurity, medical education or historical analysis.

A refusal is not necessarily evidence that the model has become less intelligent. It may reflect a policy or risk-classification change. From the user's perspective, however, the result can still feel less useful.

What the Research and Benchmarks Show

ChatGPT's Behavior Can Change Between Updates

A widely discussed Stanford and UC Berkeley study compared versions of GPT-3.5 and GPT-4 released only a few months apart. The researchers found substantial changes across mathematical problems, sensitive questions, code generation, instruction following and multi-step knowledge tasks.

Some results became worse while others improved. The study therefore did not prove that ChatGPT had suffered a universal decline. Its most important finding was that the behavior of the same commercial AI service can change significantly over a relatively short period.

Important context: The famous prime-number result from this study measured one narrow task using a specific prompt and grading method. It should not be interpreted as evidence that GPT-4 lost nearly all of its general intelligence. It does show why AI services require continuing evaluation instead of assuming every update improves every capability.

OpenAI Has Reversed a Bad Model Update

In April 2025, OpenAI rolled back a GPT-4o update after it made the model excessively agreeable and flattering—a behavior known as sycophancy. OpenAI said the model was validating users in ways that could reinforce anger, impulsive decisions or harmful beliefs.

This incident is strong evidence that model updates can introduce noticeable behavioral regressions. It also shows that user reports can identify problems that conventional benchmark scores fail to capture.

The GPT-5.5 Hallucination Benchmark Needs Context

Artificial Analysis reported that GPT-5.5 with its highest reasoning setting achieved 57% accuracy on the AA-Omniscience knowledge benchmark but recorded an 86% hallucination rate.

That does not mean 86% of all ChatGPT responses were false. AA-Omniscience deliberately asks thousands of difficult questions across many subjects and rewards models for declining to answer when they lack reliable knowledge. Its hallucination rate reflects how often the model attempted an answer and was wrong rather than recognizing uncertainty.

The result exposed an important weakness: a model can possess more factual knowledge and answer more questions correctly while also being too willing to guess when it should say, “I don't know.”

GPT-5.6 Shows Further Improvements—but Not Perfection

OpenAI's GPT-5.6 system card reports that GPT-5.6 Sol made slightly fewer factual errors than GPT-5.5 and was substantially less likely to repeat errors that users had previously reported.

OpenAI also cautioned that these tests use conversations selected because users had already flagged factual problems. They are intentionally difficult cases and do not represent the average ChatGPT conversation.

The larger lesson is that no single benchmark can determine whether ChatGPT is “better.” A model can improve in coding, research and factuality while changing in tone, creativity, length or willingness to answer.

Common ChatGPT Complaints Examined

Complaint	What the Evidence Suggests	Conclusion
Answers are shorter and less detailed	Newer models have been promoted as more concise and token-efficient.	Often valid, especially when the prompt does not specify the required depth.
The personality feels colder or more corporate	OpenAI acknowledged that some users preferred GPT-4o's warmth and conversational style.	Valid as a user-experience complaint, even if capability benchmarks improve.
ChatGPT ignores formatting instructions	Research has documented changes in instruction following between model versions.	Can be valid, but clearer formatting requirements and examples often help.
ChatGPT invents facts or sources	Hallucination remains a documented limitation, particularly when the model is uncertain.	Valid. Important factual claims should be independently checked.
ChatGPT agrees with everything I say	OpenAI publicly rolled back a model update because of excessive agreement and flattery.	A documented risk known as sycophancy.
Coding quality is getting worse	Current GPT-5.6 evaluations show broad coding improvements over earlier OpenAI models.	Not supported as a universal decline, but results vary by language, repository and prompt.
It refuses harmless questions	Safeguards and risk classifications change over time and can produce false positives.	Possible, especially in sensitive or dual-use subject areas.
It was smarter before	Older models may have matched a user's preferred task, style or workflow better.	Sometimes true for a particular use case, but not necessarily for overall capability.

Which Tasks Have Improved or Declined?

Writing and Editing

Writing quality is difficult to measure because preferences differ. Newer models can follow complex briefs, analyze large documents and produce polished business material, but some users find their default tone too compressed or standardized.

For better writing results, provide a sample of the desired voice, specify the audience and state exactly how detailed the final version should be.

Coding

Current OpenAI evaluations show major improvements in software engineering, terminal use, tool coordination and work across large codebases. Coding is therefore one of the weakest areas for a claim that ChatGPT has universally deteriorated.

However, coding failures remain possible. ChatGPT may omit dependencies, invent methods, misunderstand a repository or claim that incomplete code is production-ready. Generated code still requires testing and review.

Factual Research

ChatGPT can synthesize information rapidly, particularly when it has access to web search and supporting documents. It should not be treated as an authoritative source by itself.

The risk is highest when a question is obscure, recent, ambiguous or outside the model's reliable knowledge. Learn more in our guide to hallucinations in AI.

Creative Brainstorming

Creative quality may feel more variable because users often become attached to the style of a particular model. The retirement of GPT-4o demonstrated that users may prefer an older model's warmth and imaginative behavior even after more technically capable models become available.

Complex Reasoning and Professional Work

This is where the strongest measurable improvements are occurring. Newer models can spend more effort examining a problem, coordinate tools, work across documents and revise their own output.

The trade-off is that advanced reasoning modes may take longer, consume more of a user's plan allowance or produce an answer that feels overly analytical for a simple request.

How to Get Better Answers From ChatGPT

1. Specify the Required Depth

Do not ask only for an “article,” “analysis” or “code.” State the expected length, sections, examples, edge cases and level of completeness.

2. Create a Clear Output Contract

Tell ChatGPT exactly what the response must contain and what it must avoid. For example: “Provide complete working code, include error handling, do not use placeholders and explain how to install each dependency.”

3. Separate Facts From Assumptions

Ask the model to label confirmed facts, reasonable inferences and uncertain claims separately. This makes unsupported guesses easier to identify.

4. Ask It to Admit Uncertainty

Include an instruction such as: “Do not guess. Clearly state when information cannot be verified.” This will not eliminate hallucinations, but it can reduce pressure to produce an answer at any cost.

5. Use Web Search for Current Information

Prices, laws, product specifications, political positions, software documentation and current events can change rapidly. Request current sources and open the most important ones yourself.

6. Break Large Projects Into Stages

Use separate stages for research, outline, first draft, fact-checking and final editing. Asking for everything in one prompt increases the chance that the model will skip requirements.

7. Start a New Chat When Context Becomes Confused

Copy the important decisions into a clean summary and begin a new conversation. This removes old instructions and abandoned approaches that may be affecting the answer.

8. Request a Self-Check

After receiving an answer, ask ChatGPT to compare it against your original requirements, identify unsupported claims and correct any omissions before producing the final version.

A useful follow-up prompt: “Review your answer as a skeptical editor. Identify any factual claims that need verification, any instructions you failed to follow and any sections that are too vague. Then provide a corrected final version.”

When to Use Another AI Tool

No AI assistant is strongest at every task. Instead of asking which service is universally best, choose tools according to the work you are doing.

ChatGPT Remains Strong For

• Complex reasoning and multi-stage analysis
• Coding and tool-based workflows
• Document, spreadsheet and presentation work
• Image generation and multimodal tasks
• Projects that benefit from integrations and saved context

Consider a Second Tool For

• Cross-checking factual or controversial claims
• Comparing different writing styles
• Research that requires visible source citations
• Tasks tightly integrated with another company's ecosystem
• Testing whether a refusal or poor answer is model-specific

Claude, Gemini, Perplexity and other AI systems may produce better results for particular prompts. You do not necessarily need several paid subscriptions. Free plans can be useful for comparison and verification. See our guide to the top AI tools you can use for free.

High-stakes warning: Do not rely exclusively on ChatGPT or any competing chatbot for medical diagnoses, legal decisions, financial transactions, safety-critical engineering or other decisions where an incorrect answer could cause serious harm. Use qualified professionals and authoritative sources.

The Verdict

ChatGPT has not simply become less intelligent. Current models are substantially more capable in many technical and professional tasks than the versions available a few years ago.

At the same time, user frustration is not imaginary. Models are replaced, personalities change, answers may become shorter, safeguards can interfere with legitimate requests and improvements on standardized benchmarks may not benefit the task an individual user performs every day.

The most accurate conclusion is that ChatGPT has become more capable but also more complex and less consistent as a single, familiar product experience. The model that excels at coding or research may not be the model a user prefers for creative writing, conversation or detailed instruction following.

Judge ChatGPT by the work you need completed rather than by its newest model name. Use clear instructions, select an appropriate reasoning level, verify important facts and compare another tool when the output does not meet your needs.

Frequently Asked Questions

Is ChatGPT really getting worse?

Not across every task. Newer models generally perform better on complex reasoning, coding, tool use and professional benchmarks. However, model changes can affect tone, response length, creativity, instruction following and willingness to answer. A user may therefore experience a real decline in a specific workflow even while overall capability improves.

Why does ChatGPT feel less helpful than before?

Possible reasons include the retirement of a preferred model, shorter default responses, changing safety rules, different model routing, accumulated context in a long conversation or a newer personality that does not match your preferences.

Did OpenAI admit that a ChatGPT update was bad?

Yes. OpenAI rolled back an April 2025 GPT-4o update because it made the model excessively flattering and agreeable. The company acknowledged that the behavior could validate harmful beliefs or impulsive decisions.

Why did OpenAI retire GPT-4o?

OpenAI said usage had largely shifted to newer GPT-5 models and that later models incorporated improvements based on feedback from people who preferred GPT-4o's warmth and creative style. GPT-4o was retired from ChatGPT on February 13, 2026, although its API availability was initially unaffected.

Does ChatGPT hallucinate 86% of the time?

No. The 86% number came from GPT-5.5's performance on the specialized AA-Omniscience benchmark. It measured incorrect attempts on difficult knowledge questions where a model could have declined to answer. It does not mean that 86% of ordinary ChatGPT responses are false.

How can I make ChatGPT give longer answers?

Specify the required length, sections, examples and level of detail. Tell it not to summarize, not to use placeholders and not to stop until every requirement is addressed. Using a higher reasoning setting may also help with complex requests.

Is ChatGPT Plus still worth paying for?

It can be worthwhile for users who regularly need higher limits, advanced models, file analysis, research, coding, images or other integrated tools. A casual user who mainly asks simple questions may find that the free versions of several AI tools are sufficient.

What is the best alternative to ChatGPT?

There is no single best alternative for every task. Claude is often considered for writing and document work, Perplexity for source-led research, and Gemini for workflows connected to Google services. Testing the same prompt in more than one tool is the most reliable way to determine which one fits your needs.

Sources and Methodology

This article distinguishes official company evaluations, independent benchmarks, academic research and subjective user experiences. No single source is treated as proof that ChatGPT has universally improved or declined.

• OpenAI: GPT-5.6 Release and Evaluations
• OpenAI: GPT-5.6 System Card
• OpenAI: Retirement of GPT-4o and Older ChatGPT Models
• OpenAI: Sycophancy in GPT-4o
• Stanford and UC Berkeley: How Is ChatGPT's Behavior Changing Over Time?
• Artificial Analysis: Independent GPT-5.5 Benchmark Results

Continue Reading:

• What Is a Hallucination in AI?
• Top AI Tools You Can Use for Free
• Basics of Artificial Intelligence
• Has AGI Already Arrived?
• AI Ethics: Core Principles and Real Risks

This version keeps the existing URL and image, removes the unsupported market-share, cancellation and QuitGPT figures, and does not include FAQ schema.