The commercial release of ChatGPT in November 2022 introduced individuals to generative artificial intelligence (AI). Users were impressed by how quickly they received answers after typing questions into the chatbot interface — lengthy, seemingly factual responses appeared in seconds. However, those more familiar with this type of AI know to exercise caution, as the chatbots can regularly produce misinformation presented as facts.
How can you fact-check an AI tool’s answers, and why should the associated steps become essential to your research process?
1. Know Why Wrong Answers Occur
Generative AI tools work by predicting the next word in a text string. It does this by relying on patterns within its training data. The material used to make these tools work comes from numerous sources, ranging from Wikipedia articles to Reddit threads. One of the main issues is that the AI can’t assess the veracity of its sources. This becomes problematic given the frequency with which people purposefully post sarcastic or humorously false information online.
Most researchers within the AI community refer to the issue of AI getting things wrong as “hallucinations.” Some psychologists have pushed back on that description, asserting that it suggests a conscious perception, which AI cannot currently achieve. They propose calling these instances of wrong information “confabulations,” instead.
2. Learn the Main Types of AI Inaccuracies
Those familiar with untrustworthy AI content generally say it falls into three main categories.
Factual Errors
Factual errors occur when the generated AI content contains aspects such as historical misrepresentations, incorrect dates or scientific falsehoods. You may have experienced these problems when asking chatbots questions related to mathematics, spelling words backwards or listing the number of countries beginning with a particular letter.
Spotting the inaccuracies in this category can be especially challenging, as the AI output may mix accurate and misleading information, presenting both within a single sentence or paragraph.
Fabricated Content
Some AI answers contain wholly made-up material. It happens because AI developers usually benchmark a model’s performance by rewarding confident guesses and penalizing uncertainty. That means you are highly unlikely to experience a chatbot admitting it does not know the answer to your question.
The professionals who create AI chatbot tools also prioritize engagement, encouraging people to interact with the products for as long as possible. Experts clarify that correcting inaccurate AI answers would undermine the business model by prompting people to seek answers elsewhere. It makes sense that individuals would use these products less often after realizing they cannot trust them.
Nonsensical Answers
AI chatbots do not understand the answers they produce. This limitation explains why the material may initially seem reliable but fall apart under scrutiny. You may find that the content makes little or no sense, although it frequently contains words and sentence structures to make it sound convincing.
People using AI Overviews have experienced this issue by asking the tool to explain made-up idioms and posting the results on social media. They didn’t always get responses, but frequently received some that sounded plausible. These interactions and their varying success rates explain another generative AI problem — people can receive different outputs to the same general questions based on what their prompts contain.
3. Use Lateral Reading Techniques
Lateral reading helps you fact-check AI because it involves assessing a source’s credibility by comparing it against other options. Avoid taking the AI’s words at face value. Instead, open another browser tab and go to a search engine, library database or another source that enables you to see the sources of facts.
A downside of many AI chatbots is that they produce information without showing supporting links. This formatting has reduced organic website traffic because many people read the AI content without looking deeper. However, AI tools can facilitate your lateral reading approach if they include features that can search the web in real time. You’ll then receive links to supplement the chatbot’s content, allowing you to compare those sources with what the tool said.
4. Find the Primary Sources
Many online articles cite secondary rather than primary sources. This practice occurs when a tech blog’s reporter uses a study as a source by linking to a publisher that covered the research, rather than to the entity that conducted it, for example. One of the best ways to fact-check AI is to find the original mentions of the information in question and compare them to details found elsewhere.
This step becomes especially important when you consider how organizations are currently deploying AI. One survey found that about half of respondents say they’ve already started using AI or will do so this year. However, out of those, only one-third have an AI strategy, while just 7% have AI-related KPIs. This gap between rapid adoption and actual AI governance means many organizations are using AI-generated information without a strong plan in place to verify its results, making fact-checking all the more crucial.
The rising use of paywalls in the media industry may make this suggestion more challenging to implement, however — particularly if you lack the necessary subscriptions to see full articles. People who use AI chatbots at work should consider asking their managers for media subscription budgets. Company leaders who care about high quality and accuracy should be more likely to allow lower-level team members to purchase media access for their fact-checking activities.
5. Be Cautious of News From AI Chatbots
Many companies developing AI chatbots incorporate news aggregators into them, positioning these features as the fastest way to access daily headlines. You may also think they will expedite your fact-checking process by making it easier to find reliable sources.
That is a short-sighted conclusion, according to a 2025 study of nearly two dozen public service media organizations from 18 countries. Participants were professional journalists who evaluated over 3,000 responses from four AI chatbots on several factors, including accuracy.
One of the main takeaways was that these tools regularly misrepresent the news, regardless of the AI tool used or the territory and language associated with the content. Although 45 of the AI answers contained at least one significant issue, Gemini performed the worst, with 76% of the answers containing issues. The researchers attributed that outcome largely to its poor sourcing.
6. Research a Source’s Author
Many of the most effective fact-checking techniques of the AI era became well-established long before now. The practice of learning about a source’s author is a good example. Perusing their previous work can reveal useful characteristics, such as whether they are biased or balanced and if they have the expertise required to position them as a worthy resource to cite.
The PROVEN acronym can help you remember what to assess when considering an author’s trustworthiness:
- Purpose: What need does this source fill, and who is the audience?
- Relevance: How closely does the source match your needs?
- Objectivity: Does the author present their information thoroughly?
- Verifiability: How easily can you check the author’s sources?
- Expertise: What makes the author qualified to cover this topic?
- Newness: Does the author use current, reputable sources?
Just as AI chatbots can fabricate entire responses, they may create fake journalists, too. In one compelling example, an editor received a pitch from a supposed journalist who wished to cover how a former mining town had transformed into a training ground for individuals investigating deaths. The editor responded that he could not find supporting information and asked the journalist how she’d heard about it.
As the situation progressed, the editor became suspicious and accused the journalist of fabricating her stories, an allegation to which she did not reply. Numerous outlets eventually removed articles previously published under the journalist’s byline, which were reportedly generated by AI.
Preserve Your Reputation With This AI Fact-Checking Framework
Besides fabricating paragraphs of text, AI chatbots can create fake sources and provide faulty links without admitting they are incorrect. You cannot automatically trust your chosen tool. Fact-checking the technology with this framework strengthens your work by significantly reducing the chances of it containing an incorrect citation or fact. The more frequently you use this cautious approach, the more efficient your workflows should become as you study the trustworthiness of AI-generated responses.
Eleanor Hecks is a small business writer and researcher, sharing her insights as Editor-in-Chief of Designerly Magazine. She is particularly passionate about helping businesses turn complex information about the rapidly evolving world of AI into practical business insights.
