Back in 2023, we ran a small experiment. We asked ChatGPT to write an article about its own weaknesses and published it without changing a single word.

The text was smooth and well-structured, but it said very little. It read like a school essay written by someone who skimmed the textbook the night before.

A lot has changed since then. ChatGPT can now search the web, read your files, analyze images, and even complete tasks for you. So are the old problems gone?

Some of them, yes. Others remain, and a few new ones have appeared along the way.

In this guide, we'll walk you through the real limitations and ethical considerations of using ChatGPT today, along with practical tips for using it well.

What ChatGPT Does Well and Where It Still Struggles

Let's start with what works, because it does a lot.

At its core, ChatGPT is a natural language processing tool. Put simply, it's very good at working with words, and it has improved a lot over the past few years.

Here's where it's most useful:

• Writing and editing. Drafts, rewrites, emails, and product descriptions in minutes.
• Summarizing. Long reports, meeting notes, and research papers, shortened to the main points.
• Explaining things. Complicated topics broken into simple steps.
• Brainstorming. A long list of ideas to start from when you're stuck.
• Coding help. Writing, reviewing, and debugging code.
• Working with files and images. Spreadsheets, PDFs, screenshots, and charts.

The writing quality is genuinely good. ChatGPT produces human-like text that's often hard to tell apart from human writing.

Remember the old complaint that it couldn't handle sarcasm and irony? That problem has mostly gone away. Modern models pick up on jokes, tone, and subtext in most everyday conversations. They can still miss a niche cultural reference, but it rarely causes real trouble.

Outdated knowledge is less of an issue, too. Since October 2024, ChatGPT has had built-in web search, so it can pull in recent information instead of stopping at its training date.

The remaining problems are harder to spot. Answers look polished and confident, which makes mistakes easy to miss. Let's go through them one by one.

6 Limitations of ChatGPT You Should Know

You can keep using ChatGPT with all of these in mind. It just helps to know where its weak spots are.

1. It Still Makes Things Up

The most serious problem is that ChatGPT sometimes gives answers that sound right but are completely wrong. In the AI world, this is called a "hallucination."

OpenAI itself published research explaining why language models hallucinate. In short, models are trained and tested in ways that reward guessing over saying "I'm not sure."

Newer models handle this better. In one of OpenAI's own tests, an older model got 75% of tricky factual questions wrong. A newer one brought that down to 26%, mostly by admitting more often that it didn't know the answer.

That's a big improvement, though errors still happen.

The risk is highest with specific details: names, dates, numbers, quotes, legal references, and citations. Web search helps, but it doesn't fully solve the problem. ChatGPT can still misread a source or attach a real link to a claim the source never made.

What to do about it: Treat every specific fact as a draft until you've checked it. Ask for sources and open them.

2. Limited Understanding of Context

ChatGPT knows a little about almost everything, but it knows nothing about you.

It doesn't know your clients, your budget, your internal processes, or why your team made a certain decision two years ago. So it fills the gaps with assumptions. The answer may look reasonable on paper but still not fit your situation.

Usually, the fix is simple: give it the context and background information it's missing. Who is this for? What's the goal? What have you already tried? What are the constraints?

The more specialized your field (healthcare, law, finance, complex software), the bigger this gap gets, and the more careful you need to be.

What to do about it: Spend 30 seconds explaining the situation before you ask. The answer will be noticeably better.

3. A Lack of Common Sense in Unexpected Places

Today's reasoning models can solve hard math problems and write working code. A few minutes later, they can trip over something a ten-year-old would get right.

This is one of the odd things about AI. A lack of common sense shows up at random: a calculation off by one, a date that doesn't add up, or a plan that ignores an obvious real-world obstacle.

Keep in mind that broad general knowledge and good judgment are different things. ChatGPT has read more than any person ever could, yet it has never had to deal with the consequences of a bad decision.

What to do about it: Do a quick sanity check on anything important. Ask yourself whether it actually makes sense in the real world.

4. It Tends to Tell You What You Want to Hear

Ask ChatGPT whether your business idea is good, and it'll likely say yes.

Language models tend to agree with the person they're talking to. Researchers call this "sycophancy." In April 2025, OpenAI even had to roll back a ChatGPT update because the model had become too flattering and agreeable.

That particular update is gone, but the general tendency remains. A supportive partner is nice for brainstorming. For business decisions, you need someone who will point out the problems.

What to do about it: Ask it to argue against you. For example: "What are the three strongest reasons this plan could fail?"

5. It Struggles With Complex Questions and Long Tasks

Simple questions usually get solid answers. Complex questions with many moving parts are harder.

When a task has many steps, small errors add up. One wrong assumption early on can quietly break everything after it, and the final result still looks finished. That's what makes it risky.

Consistency is another issue. Ask the same question twice, and you might get two different answers. That's fine for creative work but a problem for anything that needs to be reliable.

What to do about it: Break big tasks into smaller steps, and check each one before moving on.

6. New Security Risks When It Acts for You

ChatGPT can now browse websites, fill in forms, and work with connected apps. That's useful, and it also gives attackers a new way in.

The main threat is called prompt injection. OpenAI describes it as a type of social engineering attack aimed at AI. Someone hides instructions inside a web page, email, or document. When the AI reads it, it may follow those instructions instead of yours, for example, by leaking data or taking an action you never asked for.

OpenAI openly calls defending against this a challenge for the whole AI industry. Nobody has fully solved it yet.

What to do about it: Give AI agents only the access they truly need, and require your approval before they send, pay for, or delete anything.

Ethical Considerations of Using ChatGPT

A few years ago, AI ethics was mostly a topic for conferences and research papers. Today, it affects ordinary work.

As Artificial Intelligence becomes part of daily business, these questions get very practical. They touch your customers, your data, and your reputation. Here's what to keep in mind.

Bias in Training Data

ChatGPT learned from a huge amount of human-written text, and people have biases.

That's how bias in training data ends up in the answers. Developers have added many safeguards, but bias can still slip through in subtle ways: assumptions about gender or nationality, one point of view treated as the "default," or weaker results in languages other than English.

This matters most when AI is involved in decisions about people, such as hiring, lending, education, or customer support.

Privacy and Security Concerns

A simple rule: whatever you type into a chatbot leaves your hands.

Depending on your plan and settings, your conversations may be stored and used to improve future models. Regulators are paying attention. In December 2024, Italy's data protection authority fined OpenAI €15 million for using personal data for training and for lacking proper age checks. OpenAI said it would appeal.

Courts can get involved as well. In 2025, as part of a copyright lawsuit, a US court ordered OpenAI to keep ChatGPT conversations that would normally have been deleted. The court largely lifted the requirement a few months later, but it showed that a deleted chat isn't always gone right away.

For businesses, the privacy and security concerns are very concrete. Don't paste client data, passwords, contracts, or personal information into consumer AI tools. Use business plans with proper data controls, and write down what your team can and can't share.

Copyright: Who Owns What?

AI models learned from huge amounts of content, including articles, books, code, and images. Many of the people who created that content never agreed to it.

That's why some of the biggest legal fights in tech are about AI and copyright. For example, The New York Times sued OpenAI and Microsoft in December 2023. As of autumn 2026, the case is still in court, with no trial date set.

There's also a more immediate question for you. ChatGPT can sometimes reproduce chunks of existing text or code. If you publish AI-assisted content or ship AI-written code, you're responsible for ensuring it doesn't copy someone else's work.

Misuse and Abuse of the Technology

The same features that make ChatGPT helpful also make it attractive to scammers and other bad actors.

Think phishing emails, fake reviews, deepfakes, and misinformation produced at scale. AI companies block many of these uses, and new laws require clear labels on deepfakes. Still, the misuse and abuse of the technology keep changing, and no safeguard catches everything.

For you, this means two things: check what you read online more carefully, and be open about how you use AI yourself.

Job Displacement and the Human Touch

In 2023, job displacement was mostly a prediction. Today, many companies are reshaping roles in support, marketing, content, and software development around AI.

In practice, people who use AI well are gaining an edge over those who don't. Entry-level roles face the biggest risk because juniors used to learn by doing exactly the tasks AI now automates.

There's also a quieter cost: the loss of human interaction. When every support chat, email, and follow-up is automated, customers notice. Automation saves time, but people still build trust with people.

Emotional Reliance

Many people now use ChatGPT for more than work. They ask it for advice, support, and sometimes just company.

A chatbot that's always available, endlessly patient, and inclined to agree can feel comforting. It can also lead to unhealthy reliance, especially for vulnerable users and teenagers. OpenAI introduced measures such as parental controls in September 2025, but this is still one of the most debated issues around AI chatbots.

Environmental Cost

Every AI answer runs on a data center somewhere, and data centers use a lot of electricity.

According to the International Energy Agency, electricity demand from data centers worldwide is set to more than double by 2030, with AI as the main driver. One query uses very little. Billions of them add up.

There's no need to feel guilty about asking ChatGPT a question. Still, using AI where it adds real value, rather than everywhere by default, is part of using it responsibly.

Responsibility of Developers, Users, and Society

Getting this right is a shared job.

• Developers are responsible for safer models, transparency, and protection against misuse.
• Users and businesses are responsible for checking outputs, protecting data, and being upfront about AI use.
• Society and regulators set the rules everyone follows.

Those rules already exist. In the EU, AI Act obligations for providers of general-purpose AI models have applied since August 2025. Since August 2026, transparency rules have also required companies to tell people when they're interacting with AI. Deepfakes and certain AI-generated content must be labeled as well. Some of the strictest requirements for high-risk systems were moved to December 2027, but the overall direction is clear.

Using ChatGPT for Content Creation: What Google Actually Says

Our original experiment focused on one use case: content creation. So it makes sense to come back to it.

Many people still believe Google punishes anything written by AI. That's not quite how it works.

Google's guidance on AI-generated content says it looks at the quality of content, regardless of how it was produced. What Google does punish is spam. In March 2024, it introduced a scaled content abuse policy: publishing lots of pages mainly to manipulate rankings counts as spam, whether those pages were made by AI, by people, or by both.

Here's what that means in practice:

• Edit before you publish. Generic AI text published at scale is what gets sites into trouble.
• Add something new. Our 2023 experiment showed how AI text can be smooth and still say nothing a reader couldn't guess.
• Use your experience. Real examples, your own data, and expert opinions are things AI can't make up for you.
• Let AI help with the process. Research, outlines, first drafts, and editing work well, as long as a person owns the final version.

The Technology Side: When ChatGPT Isn't Enough

ChatGPT is a general-purpose tool built to be decent at everything, for everyone. Sometimes that's exactly what you need. In other cases, a tool built for your business does the job better.

Many of the limitations above, like missing context, made-up facts, and risky permissions, can be reduced with the right setup. For example:

• AI connected to your own data. With an approach called RAG (retrieval-augmented generation), the AI answers based on your documents, knowledge base, or product data, in addition to its general training.
• AI agents with clear boundaries. Agents that handle specific tasks in your systems, with only the permissions they need and human approval where it matters.
• Chatbots built for your customers. Assistants that know your products, follow your rules, and hand the conversation to a person at the right moment. If you're curious what that looks like in practice, check out these real-life chatbot examples.

This is the kind of work we do at Incora. Our team builds custom AI solutions, from RAG systems that answer based on your own data to AI agents that automate real workflows. We design each one around your processes, security needs, and the limitations covered in this article.

How to Use ChatGPT Responsibly: A Quick Checklist

Responsible use and development of AI can be fairly simple. These eight habits cover most of the risks we've discussed:

Conclusion

Three years after our first experiment, ChatGPT has grown from an impressive demo into an everyday work tool. Many of the problems we wrote about in 2023 are fixed. The ones that remain are less obvious, so they deserve your attention.

Knowing the limitations and ethical considerations of using ChatGPT helps you decide when to trust it, when to double-check it, and when to bring in a person.

Future research and development will keep moving these boundaries. Models will likely become more accurate, more transparent, and better at saying what they don't know. Until then, let AI handle the heavy lifting, and keep the final decisions with your team.