How to Find Out What ChatGPT Says About Your Brand

Can you just ask ChatGPT yourself?

Yes, and you should, but treat it as a first look rather than a measurement. One answer, on one day, in one account, tells you what was possible, not what buyers typically see.

Asking ChatGPT directly is the fastest way to get a feel for the problem. Open a new chat and ask the questions a buyer would ask: "What are the best tools for X?", "What is [your brand] known for?", "What are the alternatives to [competitor]?" Read the answers carefully. Are you mentioned? Where in the list? How are you described? Which competitors appear next to you?

The limits show up quickly, though:

• Your account is not neutral. Memory, custom instructions and previous chats can shape the answer you get. A buyer's ChatGPT is not your ChatGPT.
• Answers vary between runs. Ask the same question twice and you may get a different list or a different order.
• Phrasing matters. "Best CRM for startups" and "Which CRM should a 10-person startup use?" can produce different recommendations.
• It doesn't scale. Ten questions across six assistants, repeated weekly, is hundreds of answers. Nobody does that by hand for long.

So do the manual check, save what you find, and use it to decide which questions matter most. Then move to a method that gives you comparable data over time. If you want a quick starting point without doing it by hand, the free AI visibility report runs a set of questions for you.

Why do answers differ between AI assistants?

Because each assistant uses a different model, a different way of finding sources and different rules for when to search the web. The same question can therefore produce a different shortlist in ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews and Mistral.

Some assistants answer mainly from what the model learned during training. Others search the web in real time and cite the pages they used. Many do both, depending on the question. A question about a fast-moving category is more likely to trigger a live search; a general question may be answered from the model's existing knowledge.

That has practical consequences for brands:

• Different sources, different winners. An assistant that leans on review sites will favour brands with strong reviews. One that leans on recent articles will favour brands with recent press.
• Different citation habits. Some assistants show their sources clearly; others mention brands without linking anywhere. That changes how you can trace why you were or weren't included.
• Different audiences. Your buyers may prefer one assistant over another depending on their role, market and habits.

This is why checking only ChatGPT gives you a partial picture. A brand can look strong in one assistant and invisible in another. Measuring across several engines shows where the gaps really are, and often points to the source behind the gap. Our article on why AI recommends a competitor instead of you goes deeper into the source side.

Which questions should you track?

Track the questions real buyers ask right before they choose: category questions, comparison questions, problem questions and alternatives to named competitors. Keep the list stable so you can compare results over time.

A good question set usually mixes four types:

• Category questions. "What are the best [category] tools for [audience]?" These show who the assistant considers the leaders.
• Problem questions. "How do I [solve the problem you solve]?" These show whether you are named when the buyer doesn't know your category yet.
• Comparison questions. "[Your brand] vs [competitor]: which is better for [use case]?" These show how the assistant frames you against specific rivals.
• Alternative questions. "What are the alternatives to [competitor]?" These are often asked by buyers who are ready to switch.

Write them the way people talk, with context and constraints, not as short keywords. Include the markets and languages you sell in, because answers can change with both. Separate branded questions, where your name is in the question, from unbranded ones; unbranded questions are where new demand is won or lost.

Start with twenty to fifty questions that map to real revenue rather than hundreds of loosely related ones. You can always add more once you see which questions move. If you sell to several segments, group questions by segment so you can see where you are strong and where you are missing.

How do you measure AI visibility over time?

Ask the same questions on a schedule, across several assistants, and store every answer. Then calculate the share of answers that mention you, compare it with competitors and watch how it changes.

Three measurements cover most of what you need:

• Visibility. The share of AI answers to your tracked questions that mention your brand. This is your headline number.
• Share of voice. How your mentions compare with competitors across the same answers. It shows whether you are gaining or losing ground relative to the market, not just in absolute terms. See share of voice for how Refora calculates it.
• Sentiment. How the assistant describes you when it does mention you: positively, neutrally or negatively, and with which attributes. A mention that frames you as "outdated" is a different problem from no mention at all. See sentiment analysis.

Alongside these, keep the sources. When an answer cites pages, record them. Over time you will see which domains shape your category, which of your own pages get cited, and where competitors are being cited and you are not.

Finally, treat changes as experiments. When you publish a new page, earn a review or get covered in an article, note the date and watch the same questions afterwards. That turns AI visibility from a one-off curiosity into a metric you can actually manage.

Frequently asked questions