How to Fix What AI Says About Your Brand: Wrong and Outdated Descriptions

Short answer

AI gets brand facts wrong for three reasons: a model's knowledge lags months behind, the pages it retrieves are outdated or contradict each other, and models guess when facts are thin. Fix the sources first. Update your own pages, correct third-party profiles and listings, make your facts identical everywhere, and request recrawls. Use each engine's feedback button too, but treat it as a nudge, not a fix. Then re-ask the questions weekly, because answers vary.

Key takeaways

Wrong answers have three common causes: model knowledge that lags, stale or conflicting sources the engine retrieves, and guessing when a fact appears rarely.
Fix the sources. Google says knowledge panel descriptions "can't be edited" and that updating the source usually changes the panel too.
Every major engine has a feedback route, but none promises a correction. OpenAI's correction route is for personal data about individuals, not brands.
In a 2025 study of AI answers about news, 45% of responses had at least one significant issue, and outdated information was among the most common accuracy problems.

Why does AI get brand facts wrong or out of date?

Cause	How it shows up	Evidence
Training lag	The model describes an old pricing model, a retired feature or a former CEO	Claude Haiku 4.5 has a reliable knowledge cutoff of February 2025 and was released in October 2025. Anthropic says Claude may not have been trained on the most up-to-date information
Stale or conflicting sources	The answer repeats an old review, directory entry or press article	OpenAI says search results and citations "can be incomplete, outdated, or incorrect." The EBU and BBC found that sometimes the sources themselves were out of date
Guessing	A plausible but invented detail, such as a wrong founding year or location	OpenAI's research says models guess because training and evaluation reward guessing over admitting uncertainty
Mixed-up entities	Facts from another company with a similar name	Google's knowledge panel help lists "the content of the knowledge panel is mixed with another of the same or similar name" as a known problem

The OpenAI paper adds a point that matters for smaller brands: the hallucination rate after pretraining "should be at least the fraction of training facts that appear once." Facts that appear rarely are the likeliest to be wrong. That is our inference for brands, not a measured result, but it fits the advice to repeat your facts consistently in more places.

Errors are common across the board. In a 2025 study by 22 public service media organizations, which reviewed 2,709 answers about news from ChatGPT, Copilot, Gemini and Perplexity, 45% had at least one significant issue, 31% had significant sourcing problems (the biggest single cause), and outdated information was among the most common accuracy problems. The study covered news, not brands, but the same failure modes, such as outdated sources, can show up in answers about a company.

How do you find out what AI says about you?

Write 10 to 15 questions about your brand: "What is [brand]?", "How much does [brand] cost?", "Who is [brand] for?", "Does [brand] integrate with [tool]?", "[Brand] alternatives".
Ask each on every engine you care about, logged out, more than once, because answers vary (see why AI answers change).
Record every claim and the sources cited. Gemini and ChatGPT show sources when they used the web, though Google notes that not all Gemini responses include them.
Compare with the truth. List the claim, what's correct, and which source the error came from.

A claim in one run isn't a pattern. Fix what appears repeatedly.

How do you fix it?

Trace the claim to a source. If an answer says you cost $49 a month, find the page that says so. It is often an old blog post, a directory or a review profile.
Fix your own pages. Keep one current page for each core fact, such as pricing, features and who you serve, with a visible updated date. Redirect or remove outdated pages instead of leaving them live.
Correct third-party sources. Update review profiles, directories and your LinkedIn page, and ask publishers to correct articles. If a Wikipedia or Wikidata entry is wrong, follow their conflict-of-interest rules: disclose your connection and propose changes on the talk page rather than editing directly. See entity SEO for AI.
Make facts identical everywhere. Name, description, pricing, location and plan names should match across your site and every profile.
Get recrawled. Ask Google to recrawl key URLs. Google says crawling "can take anywhere from a few days to a few weeks," and repeated requests for the same URL won't speed it up. IndexNow can notify participating engines such as Bing. OpenAI says about 24 hours for its systems to adjust to a robots.txt change, which only matters if you were blocking its crawler.
Publish the corrective answer. Write a clear page for the exact question, such as "Does [brand] support [feature]?", with the answer in the first two sentences and a date.
Report the error in the engine (below).
Re-check weekly. Search-enabled answers change as pages are re-read. A model's own memory changes only with a new release.

How do you report an error in each engine?

Engine	Route, per its own help pages
Google AI Overviews	Click the thumbs down at the bottom of the overview, choose Report a problem, pick a category, add detail and submit
Google AI Mode	Use the thumbs down at the bottom of the response and add details
Gemini	Click Bad response below the answer and add a reason, or use Settings & help, then Send feedback. For legal reasons, use More, then Report legal issue
ChatGPT	Use the Thumbs Down button. To correct inaccurate information about a person, submit a request at privacy.openai.com
Claude	Use the thumbs down button or the feedback address on Anthropic's help page
Perplexity	Use the flag icon below the answer, or open a support ticket, per its help center

Google says "AI Overviews can and will make mistakes" and that feedback and human review help it improve quality, which is not a promise to fix your answer. Gemini attaches the conversation to your feedback.

What can't you fix directly?

Brand descriptions. We found no AI engine that documents a dedicated way for a company to correct facts in its answers, and the feedback buttons are general. The nearest route is Google's knowledge panel, where a verified representative can suggest edits. OpenAI says it will "consider" correction requests about individuals "based on applicable law and the technical capabilities of our models." Google's knowledge panel help says descriptions "come from various data sources and can't be edited," and that to update one you contact the source. Claude's help center offers a removal route for site owners, not a correction route: removing content from your site is described as the best way to keep it out of Claude's web answers. For harmful or false claims, use the engines' legal reporting routes and get advice, as we aren't lawyers.

How do you know the fix worked?

Re-ask the same questions on the same engines, weekly, and count how often the correct fact appears across many answers. A single correct answer doesn't mean the problem is gone, and a single wrong one doesn't mean it persists. Citepoint tracks answers on a schedule across six engines, which makes it easier to see when a description changes, though it can't correct what an engine says. The free scan shows who AI recommends for three buyer questions in your category.

Frequently asked questions

How long until a fix shows up in AI answers?

For engines that search the web, it can show once the corrected page is recrawled, which Google says can take a few days to a few weeks, and once the engine chooses to cite it. Facts held in a model's own memory change only with a later release.

Can I pay to correct how AI describes my brand?

No engine documents a paid route for corrections. The workable routes are fixing the sources the engine reads and using its feedback tools.

Why does one engine have my facts right and another wrong?

Each engine uses a different model, search index and set of sources. Google says AI Mode and AI Overviews may use different models and techniques, so even Google's two features can differ. Measure and fix each engine separately.

Will the thumbs-down button change the answer?

It may help the engine over time, but none of the help pages we read promise a specific correction. Google says it uses feedback and human review to evaluate and improve quality. Use it, and fix the sources as well.

Should I block AI crawlers to stop wrong answers?

Usually not. Blocking a search crawler keeps your corrected pages out too. OpenAI says sites opted out of OAI-SearchBot won't be shown in ChatGPT search answers, which removes your chance to be described accurately.