How to monitor what AI says about your brand.

More people now ask an AI assistant about a company before they visit its website. Customers ask whether you are any good, journalists ask what you are known for, candidates ask what it’s like to work for you. Monitoring your brand in AI means knowing what those answers say, whether they are true, and where they come from.

What to monitor

Five things matter most:

1. Whether you are mentioned at all. When someone asks for the best options in your category, are you on the list? Being left out is easy to miss, because nothing tells you it happened.
2. How you are described. The words an assistant reaches for, the strengths it names and the caveats it adds. This is your reputation, summarised by a machine.
3. Whether the facts are right. Founders, owners, locations, prices, product names, dates. Out-of-date facts are common, and they spread.
4. Who else is named. Which competitors appear beside you, and whether they are recommended ahead of you.
5. Which sources the answer cites. The pages behind an answer tell you what shaped it, and where to start if it’s wrong.

Which questions to ask

Ask the questions real people ask, in their words. A good starting set covers four groups:

• Reputation: “Is Lope trustworthy?”, “What are common complaints about Lope?”, “What do customers say about Lope?”
• Facts: “Who owns Lope?”, “Where is Lope based?”, “How much does the Lope Daily cost?”
• Comparison: “What are the best running shoe brands?”, “Lope or New Balance for wide feet?”
• News and people: “Has Lope been in the news lately?”, “What is it like to work at Lope?”

Ask each one in the assistants your audience uses, with web search on where there’s a choice. Our free check writes eight of these for you and opens each in ChatGPT, Perplexity, Claude or Copilot.

Spotting errors

Read each answer as a sceptical editor would. Most mistakes fall into a few types:

• Out of date: an old price, a former chief executive, a discontinued product.
• Mixed up: facts about a company with a similar name, or a parent company’s news told as yours.
• Unsourced numbers: a market share or customer count that traces back to nothing. Of the statistics AI quoted that Trakkr traced back to their source, fewer than half reached one that says the same thing.
• Unfair framing: an old complaint presented as current, or one bad review treated as the consensus.

Note the sources cited beside each error. If the same page keeps appearing, that page is usually the cause.

Correcting errors

You can’t edit an AI answer, but you can fix what it reads.

• Start with your own pages. Put the facts people ask about in plain sentences: who you are, who owns you, where you are, what you sell and what it costs. Date anything that changes.
• Update your profiles. LinkedIn, app stores, review sites and business listings are read often and fall out of date quietly.
• Correct public references properly. On Wikipedia, suggest changes on the article’s talk page and say who you represent rather than editing your own entry.
• Answer where the conversation happens. If a forum thread drives an unfair answer, a clear, honest reply from your team is often read as well.
• Use the feedback buttons. Flag a wrong answer in the assistant itself. It rarely fixes things for everyone, but it costs nothing.

Then ask the same questions again in a few weeks. Answers that draw on live web search can change within days; what a model learned in training changes much more slowly.

The sources AI trusts

AI answers about brands lean heavily on a small set of sites: encyclopedias, forums, video, professional networks and review sites show up in answers about almost every brand. The data shows which sites appear for the most brands.

Your own site is a smaller part of the picture than most teams expect. In Trakkr’s citation study, a brand’s own pages were a small minority of what AI cited about it. Getting the facts right on the sites AI already reads is usually the fastest way to change what it says.

When to use a tool

Checking by hand works for a handful of questions once a month. It stops working when you need to:

• See changes as they happen, such as after a launch, a price change or a news story.
• Cover many questions across products, markets and languages.
• Compare with competitors, so you know whether a drop is yours alone.
• Show the picture to others, with history and reports a leadership team can read.

That is what AI brand monitoring tools do: they ask your questions on a schedule, store every answer, and show what changed and which sources moved it. Our ranking compares them on published criteria, with prices.