You don't need software to get a first read on how AI talks about your brand. You need five questions, three tabs, and a notes page. It takes about 30 minutes.
The five questions split into two kinds. Two don't mention you, so they show whether the AI brings you up on its own. Three do mention you, so they show what it says once it knows who you are. Our own grader uses the same split.
The five questions
Swap the brackets for your own category, buyer and competitor.
Questions that don't name you
- "What is the best [category] for [type of buyer]?" Example: "What is the best invoicing tool for a five-person agency?"
- "What are the top [category] options for [specific need or budget]?" Example: "What are the top invoicing tools for teams that bill in several currencies?"
Questions that name you
- "What is [Brand] and who is it for?"
- "Is [Brand] worth it compared with [competitor]?"
- "How much does [Brand] cost and what is included?"
How to run the test
- Use a fresh chat for each question. Turn off memory or personalization if the tool has it, so earlier chats don't tilt the answer.
- Use the same wording on all three engines. ChatGPT, Claude, and Perplexity answer differently, and that difference is the point.
- Ask each question more than once. Answers vary from run to run. Two or three tries per question gives you a pattern instead of an anecdote.
- Write down the date. Models and search indexes change, so a result is only true for a moment.
What to write down
For every answer, record five things:
- Were you mentioned?
- If so, where in the answer: first, middle, or an afterthought?
- What did it say about you, in its own words?
- Was anything wrong or out of date?
- Which competitors did it name, and which sources did it cite?
How to read what you find
- Missing from questions 1 and 2: a visibility problem. The AI doesn't connect you with your category.
- Mentioned, but low in the list or described generically: a position problem. You are in the room but not the pick.
- Wrong or outdated facts in questions 3 to 5: an accuracy problem. We cover how to fix those in What to do when AI gets your brand wrong.
Keep it repeatable
Save the exact wording. Small changes to a question can change the answer, so if you rephrase every month you can't tell what moved. Run the same five questions once a month and keep the notes in one place.
Prefer not to do this by hand? The free AI Visibility Score asks ChatGPT, Claude, and Perplexity a version of these questions and scores the answers out of 100.