AI brand monitoring: how to see what AI says about your business

How is AI brand monitoring different from classic brand monitoring?

AI brand monitoring means asking AI assistants about your business by name and reading what they say. The point is what is said, not how often. A business can be named and still be described wrongly: old prices, a closed location, a competitor that has since merged, a product it no longer sells. A customer who reads that answer may never reach your site to learn otherwise.

Classic brand monitoring has something to watch. A web alert looks for your name in pages that are already published. Social listening does the same with public posts. An AI answer is written for one person at the moment they ask. ChatGPT, Gemini, Claude and Perplexity do not publish a feed of what they tell people, so there is nothing to subscribe to. You have to ask.

Site owners do get some counts. Google Search Console’s Generative AI performance report, open to all websites since August 31, 2026, shows impressions of links to your site in AI Overviews and AI Mode. Bing Webmaster Tools’ AI Performance report shows citations of your pages in Copilot and Bing’s AI summaries, from sampled data. As their help pages describe them, neither shows the words an answer used about you.

Counting is its own job: buyer questions without your name, asked every week and read as a share. Our guide to AI visibility tracking covers that kind of AI brand tracking. This guide covers reading the answers and correcting what is wrong.

Which questions should you ask about your own business?

Write the truth down first. One page is enough: your exact business name and any past names, every open location with its address and hours, your phone number, current prices, what you sell today, what you stopped selling, and who owns the business. You will check every answer against this sheet.

Then ask these six questions, with your real names in the brackets. They are brand questions: your name is in them, so they show how you are described, not whether a buyer would find you.

If you have stopped selling something, add a seventh: “Does [business] still offer [product]?”

Ask in a private window, logged out where the assistant allows it, with web search on where there is a choice. Keep the wording the same from round to round, and save each answer with its date and links. Six questions across ChatGPT, Gemini, Claude and Perplexity make 24 answers, few enough to read every word.

This round is a sample, not a log. ChatGPT brand mentions in other people’s chats are private to those people, and the same goes for the other three assistants.

What should you look for in each answer?

Accuracy. Mark every fact you can check: name, address, hours, price, product, owner. Compare each with your sheet and write right, wrong or out of date. Out of date deserves its own mark, because it points to a source that was once true.

Tone. Read the adjectives and the hedges. “Some customers complain about” and “mixed reviews” are claims, so ask what they rest on. Also check whether the description matches what you do now, or what you did years ago.

Comparison. Look at who is placed next to you. A much larger firm or a different kind of business suggests you are filed in the wrong group. A competitor that has closed or merged suggests the answer rests on an old page.

Source. Open every link, and note who owns the page and when it was last updated. OpenAI’s help page “Searching the web with ChatGPT” says answers that use web search may include citations you can open. Anthropic’s help page, dated March 16, 2026, says to review cited sources, because the original page may hold context that Claude’s summary leaves out. Our guide to AI citations goes deeper into reading them.

Watch for answers with no link. Anthropic names outdated training data among the causes of wrong answers, so a claim with no source may come from the model’s memory rather than from a page you can fix. If you ask an assistant where a fact came from, treat the reply as a lead and open the page yourself: the same Anthropic page warns of quotes that sound credible but are not grounded in fact.

How do you trace a wrong answer back to its source?

Work through these in order until you find the page.

• Open the cited page. If it says the wrong thing, that page is your target. If it says the right thing, the assistant misread it or mixed sources, so go to the assistant’s feedback channel.
• Search your own site for the old fact. Old blog posts, PDFs and forgotten location pages are still published.
• Search the web for the wrong fact itself. Put the old price, old address or old product name in quotes with your business name. Google’s help page says quotes ask for an exact match.
• Nothing found? The claim may come from memory. Report it, and state the right fact plainly on your site.

Two searches to copy:

site:yourdomain.com "the old price or old address"
"Your Business Name" "the old price or old address" -site:yourdomain.com

An invented example shows why the order matters. A bakery closed its second shop two years ago. Asked for hours and address, an assistant lists both shops and cites a directory listing the owner never claimed. The bakery’s site and its Google Business Profile are already right, so rewriting the site would change nothing. The fix is that one listing.

Keep one row per wrong fact in a sheet: assistant, question, what it said, the source it leaned on, the fix, the date fixed and the date you checked again.

Where does each wrong answer get fixed?

Fix the source first. A report to an assistant is a request. A corrected page is a fact that can be read the next time.

For a mix-up with a similar name, Organization schema is how you state your name, address, phone and official profiles in a form machines read.

On reviews, Google says you can report any review, but only those that violate its policies are eligible for removal, and it asks you not to report one just because you disagree with it. Yelp says it will not remove a critical review unless it clearly violates its Content Guidelines.

What can you report to ChatGPT, Gemini, Claude and Perplexity?

We read each company’s pages on November 4, 2026. None of them promises that a report will change an answer. Buttons and labels move.

The third column is written for personal data. OpenAI’s help article on removing personal data from ChatGPT responses asks you to explain why the information is wrong and give reliable evidence, may ask for proof of identity, and says OpenAI “cannot independently investigate disputed facts”. If the wrong fact concerns a named person, such as the owner, these routes may apply to that person. We found no page from OpenAI or Google that gives a business a route to correct a plain fact about itself, such as a price.

The nearest thing is Google’s Report Content on Google tool. It lists “Report content that defames you or your business/organization” and asks whether a request relates to content generated by AI within a Google product, naming AI Overviews and AI Mode. That route is for content that may break a law, not for an out-of-date price. A claim of defamation is a question for a lawyer.

Whichever button you use, include the exact question, the answer, the date, the right fact and a link to the page that proves it. Ask from a chat with nothing private in it: Google says the conversation is added to your feedback, and Anthropic says it stores the entire related conversation for up to 5 years.

Can you edit Wikipedia or Wikidata about your own company?

We read these rules on November 4, 2026.

Wikipedia. The English Wikipedia’s conflict of interest guideline counts “an owner, employee, contractor, investor or other stakeholder” as having a financial conflict, and says such editors are “strongly discouraged from editing affected articles directly”. The route it gives is the article’s talk page: start a new section, put the {{Edit COI}} template at the top, describe the change and cite a reliable source. Its FAQ for organizations allows you to correct minor “hard fact” errors, such as an incorrect date, once you have disclosed your conflict. If you are paid for the work, which includes employees, its paid-contribution disclosure policy requires you to state your employer, client and affiliation.

The same FAQ says “nobody owns any article in Wikipedia”, and the guideline says editors “may decline to act” on a request.

Wikidata. Its page on autobiography says: “In general on Wikidata it is allowed to edit the data item about yourself, (or someone you know or represent).” The page speaks of people and does not name companies. It requires disclosure if you edit for reward, and says not to remove sourced statements: raise a wrong one at Project chat.

None of the pages we read from the four AI companies says how much an answer relies on either site. Correct a wrong fact there because it is wrong, not as a way to be named.

How long does a fix take to show up in AI answers?

No AI company publishes this. None of the pages we read from the four AI companies says how soon a corrected page reaches an answer, so any number you see elsewhere is someone’s observation, not a rule. Only the steps before that point have some published timing.

After a fix, write the date in your sheet and ask the same question on a schedule you can keep.

• Still wrong, and it cites the page you fixed. It may be working from an older copy. Wait and ask again.
• Still wrong, and it cites another page. You have another page to fix.
• Still wrong, with no source. It may be the model’s memory. Report it and keep your own page right.
• Right once. Ask again before you trust it, because the next answer can differ.

How do you track brand mentions on the open web for free?

Published pages are what an assistant reads when it searches, so it helps to catch a wrong fact when it first appears. We checked these four on November 4, 2026.

• Google Alerts. At google.com/alerts, enter your business name in quotes and open the options: how often, which sources (news, blogs, web, video, books, discussions, finance), language and region. Google’s help page says you get emails “whenever we find matching search results”.
• Talkwalker Alerts. Its own page describes a free service covering news, blogs, forums, websites and X, delivered by email, RSS or Slack.
• A monthly search by hand. Search for “Your Business Name” -site:yourdomain.com, using the quotes, minus sign and site: search that Google’s help page documents.
• The Links report in Google Search Console. It shows the sites that link to you most and the pages they link to.

What these miss:

• Every AI answer. None of them reads what an assistant said.
• Anything private. Closed groups, direct messages and pages behind a login are outside what a public search can see.
• Mentions without a link, in the Links report, which covers backlinks only.
• Whatever falls outside an alert. Google’s help page does not list what alerts cover, so a quiet inbox does not prove nobody mentioned you.

You would need a paid tool when the asking itself becomes the job: many locations, brands or languages, or answers that must be collected and stored on a schedule. AI brand monitoring tools automate that part, and choosing one is a separate question with its own guide. Before you pay, ask whether the tool checks answers against your facts or only records whether you were named. In the second case, the reading is still yours.

What the MWY checker does and does not read about your brand

The free AI visibility checker works on the part you control: your own site. It reads public pages as raw HTML without running JavaScript, which is how outside tests have mostly found AI bots reading pages, and scores 24 items. Several bear directly on whether your facts can be read: business details as structured data, one way of writing the business name, address and phone as text, public prices, and published and updated dates. The free result shows your score, the three groups and the 5 items that matter most.

The checker does not tell you whether those facts are true. It can see that a price is written on a page. It cannot know that the price is current. It does not read your Google Business Profile, a directory, a review site, Wikipedia or Wikidata, and the free check asks no assistant anything.

Pro opens soon. It adds 20 buyer questions a week, asked to ChatGPT, Gemini, Claude and Perplexity through their APIs with web search on, and it stores each answer word for word with the sources it cites. You can edit or replace any of the questions. The report shows which answers name your site, which other businesses they name and which pages they cite. It does not mark a fact as right or wrong: that reading stays with you.

MWY does not correct an answer or file a report for you, and it does not promise that an assistant will name your business or fix a wrong fact. Nobody outside the AI companies controls that.