TL;DR
- AI engines do not keep a master file of your business — they synthesize answers by pattern-matching across every mention they have seen, so outdated sources often win by volume.
- Not-cited and cited-but-wrong are different diagnoses — the first is an access or identity problem, the second is a cleanup problem across your site and third-party pages.
- About 90% of wrong AI answers trace back to inconsistencies inside your own website, which is why the fix order is own site first, then Google Business Profile, then directories, then press.
- Live-search engines like Perplexity reflect changes in days. Pattern-based engines like ChatGPT take weeks or months, with one practitioner reporting an 8-week turnaround.
- Pick the single worst wrong fact, correct every page on your own site that mentions it, update Google Business Profile or LinkedIn, and set a monthly calendar reminder — corrections do not stay fixed on their own.
Last week I wrote about the four-surface check. It is the way to find out whether AI engines know about your business, and what they are saying.
Some readers will have found good news. The AI knows them. It mentions them by name. It gets the basics right.
Some will have found a different result. The AI mentions them, but says the wrong thing. Wrong address. Wrong service.
A product line they dropped three years ago. An old phone number. Sometimes a confused blend of the business and a competitor down the street.
That is not nothing. That is actually a different problem, with a different fix. This article is about that fix.
Think of your business as a file inside the AI’s mind. Not one clean page — a thick file, built up from every mention of you the engine has ever seen. An old press release. A new homepage.
A directory entry from 2022. A review site you forgot exists. A customer’s LinkedIn post.
The AI does not curate the file. It reads everything in it, and tells the customer what most of the records say.
When most of the file is correct, the answer is correct. When too many old photos are still sitting in the file, the answer is wrong.
You cannot reach into the file. But you can change what goes into it next.
Why did the AI get it wrong in the first place?
AI engines don’t keep a master file of facts about your business. They synthesize answers from every mention of you they have ever seen. Old reviews, dead press pages, directory entries from three years ago, your current site. When the old mentions outnumber the correct ones, the wrong answer wins by volume.
The fix isn’t one edit. It’s making the correct version louder than the outdated one, across every place the AI is listening.
ZipTie put it plainly in March. AI models cannot tell "official" from "frequently mentioned." Say twelve old comparison articles list your price as $49. One correct page on your site says $79.
The AI does not know which to trust. It averages. It guesses. It often picks the loud wrong answer.
This is why a single-page edit rarely works. Your site is one voice. The web is a chorus. You need to change enough of the chorus to tip the average.
Is it "not cited at all" or "cited with wrong facts"? These need different fixes.
Not-cited means the engines cannot find you or cannot recognize you as an entity. Cited-but-wrong means they found you, but the file they built on you contains outdated or contradictory records.
The first problem is about access and identity. Crawler permission, structured site markup, entity signals.
The second is a cleanup problem across your site, your directory listings, and the third-party pages that still describe the old version of you.
These are not the same project. If an AI has never heard of you, adding more correct content helps. If an AI has heard of you wrongly, you need to find the wrong sources and either correct them, drown them out, or both.
Most of this article is about the second case. The cited-but-wrong case. It is the more common one for businesses that have been around for more than a year.
Start where you have the most control — your own site
About 90% of wrong AI answers trace back to inconsistencies inside your own website. An old pricing table in a 2021 blog post. A team page listing someone who left. A service description you revised on the homepage but not in the footer.
Clean your own site first. It is free, it is fast, and it removes half the contradictions the AI is hearing. In a 2026 analysis of where wrong answers come from, Mentio traced the large majority back to inconsistencies on a business’s own pages.
This is the unglamorous part, and the most useful. Open your site and read it like a stranger would. Look for pages you have not touched in two years.
Look for services you dropped. Look for prices that changed. Look for bios of people who are not there anymore.
Then fix them. Delete, rewrite, or redirect — whichever makes sense.
While you are there, add one small thing. A page called "About our business" or "Company facts." Put the correct, current facts on it, in plain sentences.
Your full legal name. Your address. What you do. Who you serve.
When you started. Your current pricing structure, even if general.
This page becomes your one source of truth. The AI learns to lean on it. Mentio’s April guide calls this the single highest-leverage page a small business can publish for AI visibility. The shape that makes that page work is what About bio AI engines trust walks through.
Fix the echoes — directories, profiles, and third-party pages
After your own site, the echoes matter. Google Business Profile first. Then business-profile platforms like LinkedIn and Crunchbase.
Then industry directories. Then review platforms. Press coverage last, because it is hardest to change.
A consistent story across five or six places is what the AI starts to trust.
A single correct page on your site surrounded by twelve outdated external mentions is not enough.
Sight AI’s April article puts the priority order this way — high-authority press sources, then business-profile platforms, then industry directories, then review sites, then Wikipedia. Their phrase is worth keeping: one bad source can poison the well.
Work down the list. Update Google Business Profile today. Update LinkedIn and Crunchbase this week.
Pick three directories that actually drive traffic to you and fix those next. Do not try to fix everything at once.
Press coverage is the hardest one. If an old article has the wrong facts, your options are small. Email the author, ask for a correction, move on if they say no.
You cannot rewrite someone else’s archive. You can only add enough new signal to outweigh it.
Why does it take weeks, not hours, to show up?
Live-search engines like Perplexity can reflect your changes in days. Pattern-based engines like ChatGPT often take weeks or months. They built their picture of you from many sources and need enough fresh signal to outweigh the old.
One business reported an eight-week turnaround after a coordinated fix across site markup, LinkedIn, Crunchbase, and press. That is a normal pace.
Faster is possible. Faster is not typical.
The uneven timing confuses people. A business owner makes a change on Monday, checks Perplexity on Thursday, sees the right answer, and thinks the whole job is done. Then they check ChatGPT on the weekend and see the same wrong answer from before.
Both results are real. Perplexity did a fresh search after your change. ChatGPT is still quoting the older pattern it learned before.
Give it time. Give it more signal. Check again in a month.
What do I do when the wrong answer comes back?
It will come back. One practitioner got a clean fix. Four months later, a new article with old information re-ingested and dragged the wrong answer back into ChatGPT.
This is not failure. It is how the tools work.
Add a calendar reminder to re-check your answers every month.
Keep your facts page up to date.
When something old re-surfaces, the cleanup is faster the second time because you already know the levers.
The practitioner’s story, logged in the Am I Cited discussion thread in January, is worth remembering. The fix worked. The fix held for four months.
Then a new blog post on a different site repeated the old information, and the AI noticed the new post and pulled it back in. None of that meant the original fix was wrong. It meant the work was not finished.
This is the ongoing cost of AI visibility. Not expensive in hours, but not a one-time project either. Think of it the way you think of cleaning a windshield.
You do it this week. You will do it again next month.
Other questions worth answering
How do you trace which source is feeding a bad description into ChatGPT?
Per ZipTie’s March 2026 guide, ask the engine directly. ChatGPT, Claude, and Perplexity will name their sources when you ask which URLs they consulted. Open each one and read what it says about you. The wrong record usually lives in a small handful of places, not many.
Does pressing your story onto news outlets actually help engines update what they say?
Less than you would guess. Per Mentio’s April 2026 guide, press releases account for just 0.04% of citations across the major engines, roughly one in 2,500. Spend the same effort on Google Business Profile, LinkedIn, or Crunchbase. Press coverage matters when it lands, but the return is thin.
How should you measure whether a cleanup worked, given that engines reply unevenly run to run?
Run the same prompt five times per engine and track frequency, not snapshots. Three numbers matter. How often the engine returns the right answer, how often it returns the wrong one, and how often it changes.
SparkToro’s 2026 research across 2,961 runs found fewer than one in 100 prompt-pairs produced matching brand lists. The honest yardstick is ‘3 of 5 right’ before, then ‘5 of 5 right’ after.
How much does schema markup actually move the needle versus plain text on your domain?
Per Mentio’s April 2026 study, between 65 and 71 percent of pages quoted by ChatGPT and Google AI carry some structured markup. That is correlation, not proof. The honest read is that schema is cheap, low-risk, and worth adding once your facts text is consistent. Skip it until the words on your homepage are right.
What happens when a competitor’s name keeps blending with yours in engine responses?
The mechanism is volume confusion. When you and a competitor share a city and a service category, engines treat the overlap as evidence rather than ambiguity. The remedy is louder identity markers. As of 2026, that means making your legal name, address, and service phrasing identical across Google Business Profile, LinkedIn, and Crunchbase.
Which wrong fact should you correct first?
Pick the single worst wrong fact an AI engine is stating about you. Find every page on your own site that mentions that fact — current and old. Correct them all.
Then update the one external profile that matters most: Google Business Profile for local businesses, LinkedIn or Crunchbase for B2B.
That is about two hours of work. It will not fix everything. It will start the clock.
The old photos are in the file. They do not leave on their own. But every new correct record you add moves the average.
Over a few weeks, the AI’s picture of you will shift. Not because you rewrote the file, but because the new records are louder than the old ones.
You did not build this problem. You did inherit it. And you are the only person in a position to start cleaning it up. Nobody else has the motivation to keep the record of your business accurate.
If the audit turned up a wrong answer you do not know how to trace, you can contact me. A fact you cannot find on any of your own pages. A claim that seems to come from nowhere. Sometimes the source is a site you forgot about.
Sometimes it is one third-party page quietly feeding three different engines. I will help you find where the wrong record lives and what kind of fix it needs. No pitch.