What does ChatGPT say about your company? How to check

The useful version is asking the questions your customers ask, across ChatGPT, Claude, Gemini, Perplexity and Google’s AI Overviews, and writing down whether you’re named, what’s said, and which pages are linked. Do that on a few different days and you’ll know where you stand before a prospect does.

Can I see what ChatGPT says about my company?

Use three types of check: ask about your company directly, ask buyer questions without naming your brand, and repeat those questions over time. Each reveals a different part of your visibility.

A branded question can reveal factual errors or confused positioning. Unbranded recommendation and comparison questions show whether your company appears before a buyer has named it. You need both views.

Which questions should I ask?

Ask the questions a customer would ask before they’ve heard of you. Write 15 to 20 of them in plain language, mostly recommendation and problem questions, with two or three about your company by name so you can see how you’re described.

Three types cover it. Recommendation questions: “who’s a good commercial insurance broker for a small manufacturer in the Midlands”. Problem questions: “what should I look for in a payroll provider”, “how do I cut energy costs in a hotel”. Comparison questions: “how do X and Y differ for a company our size?”. Then a couple of branded ones, “what does [company] specialise in” and “is [company] any good”, so you can check what’s said and whether it’s accurate.

Write them the way a customer would, not the way your website does. If your site says “integrated facilities solutions” and your customers say “someone to look after the building”, ask about someone to look after the building. How to create tracked prompts goes into this in detail.

How do I check ChatGPT, Claude, Gemini, Perplexity and Google?

Run each question in a fresh conversation in each assistant, with web search on where it’s an option, from the country you sell in. The fresh conversation matters: assistants carry context from earlier messages, so a run of questions in one chat will colour each answer with the last.

For ChatGPT, use a temporary chat so memory and past conversations don’t shape the answer. Claude and Gemini both search the web when the question calls for it; check the answer shows sources. Record any sources Perplexity displays and note when no citation is available. For Google, run the question as a normal search and read the AI Overview if one appears, then run it again in the AI Mode tab, because the two don’t always agree. If you’re checking from outside the market you sell in, use a VPN or ask a colleague there, because location changes the sources.

Read past the first paragraph. Assistants often name a business in a list further down, or link to a page without naming the company in the text. Both count, in different ways. Mentions and citations in AI search explains why.

Why do the answers change every time I ask?

Because each answer is generated fresh, often from a fresh search, and the assistant doesn’t pick the same sources or the same wording twice. Models are updated regularly, and the web they read changes daily. One answer is a sample, not a verdict.

The practical fix is repetition. Ask each question on three or four different days and count how many times you’re named out of how many tries. A company named in three of four runs is visible for that question. A company named once in four is on the edge. That count is far more reliable than any single screenshot, which is the whole point of tracking the same questions over time rather than checking once. How tracked prompts work covers the method.

What should I write down?

Eight things per answer, in a spreadsheet: the question, the assistant, the date, whether you were named, where in the answer you appeared, what was said about you in one line, which pages were linked, and which other businesses were named. After a week the patterns are obvious.