How to monitor your brand in ChatGPT

To monitor your brand in ChatGPT, ask it the questions your customers actually ask, in fresh chats, on a repeating schedule, and record every answer against a list of facts you have verified. That is the whole method. The rest of this page is about doing it in a way that produces evidence instead of anecdotes.

Ask buyer questions, not vanity questions

"What is [your brand]?" is the question founders ask. Buyers ask harder ones: "What does [brand] cost?", "Is [brand] better than [competitor]?", "What are the best tools for [your category]?", "Is [brand] legitimate?". The commercial damage from a wrong answer concentrates in exactly these questions, because they arrive at the moment someone is deciding.

Write down 5 to 10 of them once, and use the same list every time. Consistency is what turns spot checks into a trend line.

Understand the two modes you are checking

ChatGPT answers from two different places, and a monitoring routine has to cover both.

Without search, answers come from training data: a compressed memory of the web as it existed before the model's cutoff. If your pricing changed after that cutoff, the model does not know, and it will state the old figure with full confidence.

With search enabled, ChatGPT retrieves live pages through a web index before answering. This mode can be more current, but it inherits the index's blind spots: pages that render only through JavaScript, pages the index has not crawled recently, and stale snapshots of pages that have since changed. We have watched an index serve a months-old version of a homepage long after a full redesign; search-mode answers built on that snapshot described a company that no longer existed in that form.

Check your questions in both modes. They fail differently.

Use fresh chats, and repeat each question

Two rules keep the data honest. First, every check happens in a new chat, because earlier turns in a conversation steer later answers. Second, ask each question more than once across your checks. Answers vary between runs; a brand that appears in three of five runs has a presence problem that a single lucky check would hide.

Record like you might need to prove it

A screenshot in a folder beats a memory. For each check, capture the date, the exact question, the full response, and which claims in it are right or wrong against your verified facts. The wrong claims are the work queue: each one traces to some source the answer was built from, and that source is what you can actually fix. When a wrong answer costs you a deal, a dated record is the difference between an anecdote and a case.

Where manual checking stops working

The method above is free and genuinely works. It also has a ceiling: five questions, two modes, and enough repetition for the variance to mean something is dozens of chats per week, every week, forever, with a hand-built archive. And it still only covers ChatGPT.

This is the job AIVIS automates: it runs your buyer questions against ChatGPT, Claude, Gemini, and Perplexity on a schedule, verifies every claim in every answer against facts you control, assigns each response a verdict, and preserves the record with timestamps. The checking is the easy half; the verification against your facts is the half a manual routine usually skips.

Learn more

• ChatGPT says wrong things about my brand: what to do when a check finds a problem
• When ChatGPT gets your brand wrong: the complete fix playbook
• What is AI visibility?: the concept behind the checks
• AIVIS pricing: automated verification from $39 a month