Sooner or later it happens to every brand that pays attention to AI answers. Someone on the team asks an assistant a simple question about the company and gets back something wrong: a product you retired two years ago, a price that was never real, a founder who never worked there, or a claim that you don’t serve a market you’ve been in for a decade. The temptation is to treat it as a glitch and move on. That’s a mistake. Wrong answers repeat, and every buyer who asks the same question gets the same bad information.
The good news is that most AI errors about a business have traceable causes, and you can work on them the same way you’d work any operational incident: detect, diagnose, fix the source, and verify. This is the playbook we’d hand to an operator who just found their first one.
Step 1: Capture the error properly
Before anyone starts editing pages, write down exactly what happened. Record the assistant or search product, the date, the exact prompt, the full answer, and any sources it cited. Take a screenshot. Then run the same question with a few natural variations, because the error may only appear on certain phrasings.
This matters for two reasons. First, AI answers change, and you need a baseline to know whether your fix worked. Second, the wording of the wrong answer is often your best clue to where it came from. If the assistant repeats a specific phrase, search for that phrase. It frequently leads straight to the page that seeded it.
Step 2: Classify the error
Not every mistake has the same root cause. In practice, most fall into a handful of buckets:
- Stale facts. The information was true once. An old pricing page, an archived press release, or an outdated directory profile is still live and still being read.
- Conflicting facts. Your own properties disagree with each other. The website says one thing, the LinkedIn page another, a partner listing a third. The model picks one, or blends them.
- Entity confusion. Your brand shares a name or a similar name with another company, product, or person, and the assistant is merging the two.
- Third-party error. A review site, article, or forum thread says something false about you, and it’s being treated as a source.
- Gap filling. There is simply no clear, crawlable statement of the fact anywhere, so the model guesses.
Each bucket has a different fix, so it’s worth taking ten minutes to decide which one you’re in before you act.
Step 3: Fix the source, not the symptom
For stale and conflicting facts
Start with what you control. Update or redirect outdated pages instead of leaving them orphaned. Make sure the current fact is stated plainly, in text, on the page that should own it: pricing on the pricing page, leadership on the about or team page, service areas on the locations page. Then sweep your own footprint: social profiles, app store listings, marketplace pages, directory entries, and your schema markup. One inconsistent profile can keep an old fact alive long after the website is fixed.
For entity confusion
Make it easy for machines to tell you apart. Use your full legal or distinctive name in key places, describe what you do and where you operate in the first lines of your about page, and use Organization markup with sameAs links pointing to your official profiles. If the confusion is with a similarly named product, a short, factual clarification on your site, such as what you are and what you are not, can help both people and machines.
For third-party errors
Contact the publisher with a polite, specific correction request and supporting evidence. Many editors and review platforms will fix a factual error when it’s clearly documented. Where you can respond publicly, such as on review sites or in forum threads, do so factually and without arguing. A calm, correct reply on the same page gives future readers and crawlers the right information alongside the wrong one.
For gaps
Publish the missing fact. If assistants keep guessing whether you offer a free trial, integrate with a certain tool, or serve a certain region, answer it directly on a crawlable page, ideally in a sentence that could stand on its own if quoted.
Step 4: Use the feedback channels, but don’t rely on them
Most AI products offer some way to flag a bad answer, such as a thumbs-down or a report option. Use it, and note that you did. But treat it as a supplement. Feedback buttons don’t change the underlying sources, and the sources are what determine whether the error comes back next month or shows up in a different assistant entirely.
Step 5: Verify and keep watching
Re-run your saved prompts after your fixes have had time to be recrawled. Answers that rely on live retrieval may update relatively quickly; answers drawn from a model’s training data can take much longer, sometimes until the next model update. Keep the incident open until you see the corrected answer consistently, and add the prompt to a recurring check so you’d catch a regression.
Make it someone’s job
The biggest failure we see isn’t technical. It’s that nobody owns the problem. Marketing assumes the web team will handle it, the web team assumes it’s a PR issue, and the wrong answer sits there. Assign an owner, give them a simple incident template, and agree on who can approve changes to which properties. If you’d rather have outside help building that process, a dedicated generative engine optimization team can run the audits and monitoring alongside you.
AI assistants will keep getting things wrong sometimes. The brands that come out ahead are the ones that notice quickly, fix the source, and make the right answer the easiest one for a machine to find.