News About AI Chatbot Errors, Hallucinations and Misinformation Outnumbers News About AI Accuracy 3.4 to 1

Summary

Picture a physical therapy clinic owner who hears from a patient that ChatGPT said the clinic does not take her insurance, which is wrong. He searches the news for how often AI gets things wrong and finds a steady stream of stories about chatbots making things up. He wonders whether anyone writes about when they get it right.

We counted news articles about AI chatbots, AI assistants and AI search whose URLs use error words (hallucination, inaccurate, misinformation, fabricated, made up, errors, mistakes) or accuracy words (accurate, accuracy, reliable, trustworthy, factual). Between January 2022 and September 2026, 3,362 used error words and 979 used accuracy words.

Error stories outnumber accuracy stories 3.4 to 1. This measures what the news chooses to write about, not how often AI is actually wrong. But it shapes what customers expect when they read an AI answer about your business.

What we measured

We took articles whose slug names ChatGPT, OpenAI, Gemini, Copilot, Claude, Perplexity, Grok, DeepSeek, chatbots, AI assistants, AI Overviews, AI Mode, AI search, LLMs or AI answers, and removed aviation and password stories that matched by accident. We then flagged error words and accuracy words in the slug. Negated accuracy phrases such as not reliable or less accurate count as errors. Articles with both kinds of words (18) are left out of both counts.

We also tagged topics by slug words: health, legal, money and markets, and elections and politics. A spot check of 30 error articles found most were about chatbot mistakes, false answers or misinformation; a spot check of 40 accuracy articles found most were about chatbot accuracy, with some about medical AI models and some skeptical in tone.

Error stories outnumber accuracy stories every year

Articles about AI chatbots and AI search by the words in their URL:

2023 was the year of hallucination: 375 slugs used the word, as chatbots reached the mainstream and early mistakes made headlines. In 2024, error coverage shifted toward elections and misinformation. By 2026 the gap had narrowed to 2.5 to 1. Some of the accuracy articles that year cover new model releases announced with claims of improved accuracy.

Health is the one topic where accuracy keeps pace

Error and accuracy articles by topic, 2022 to September 2026:

Medical researchers publish many studies testing whether chatbots answer health questions correctly, and those studies often report decent accuracy, which balances the health coverage. Legal coverage is the opposite: 226 error articles, many about lawyers filing AI-made citations or privacy complaints over false answers, against just 6 about accuracy.

Error coverage has turned more negative

Average GDELT tone of error and accuracy articles by year:

In 2023 error stories were only mildly negative, often written with curiosity about a new technology. By 2025 they averaged -2.26, and 415 of 716 scored below -2. Accuracy stories stayed positive throughout.

What this means

Customers have read years of stories about AI getting things wrong, so many will double-check what an AI answer says. That works in your favor only if the details they check, such as your website, Google Business Profile and directory listings, are correct and consistent. When AI tools find conflicting facts about a business, they can repeat the wrong one. Cleaning up conflicting listings is unglamorous, but it is often the first fix an AI SEO agency makes.

Ask the major AI tools about your hours, services, prices and location, and note anything wrong. Then fix the source pages they cite. Our free AI visibility audit finds those errors and traces where they come from, which is the part an AEO agency should never skip.

Limits

This counts news coverage, not error rates. It says nothing about how often any AI tool is actually right or wrong.

Matching uses URL slugs and a word list. Some articles use these words in other senses, and some accuracy articles are skeptical in tone.

Topic tags come from slug words and miss articles that do not name the topic in the URL.

Non-English coverage is undercounted because the word lists are in English.

Frequently asked questions

Related studies

• How the news covers AI searchOnly 0.3% of AI Search News Mentions Small Businesses
• How the news covers AI searchAds in AI Answers: News Coverage Up 841% in 2026
• How the news covers AI searchGEO, AEO and AI SEO News: 183 Articles a Month in 2026
• How the news covers AI searchLocal Business AI Search News: From 6 in 2024 to 157 in 2026