How to run a 20-query AI search audit when you're doing this for the first time

What does ChatGPT say when someone asks for a solution in your category? You probably don't know. You might have searched your brand name once, seen it show up, and stopped. That is not an audit. That is checking whether you exist.

An audit tells you where you win, where you lose, and what to fix first. The shape I walk founders through in consultancy sessions is a 20-query test across three AI engines. You can finish it in one sitting. You will have a score at the end, a gap analysis, and enough evidence to decide whether this matters for your business.

Here is how to run it.

Pick the 20 queries that define your category position

The audit starts with the query list, not the engines. Most founders start backward, testing their brand name in ChatGPT and calling it done. That tells you nothing about whether buyers find you when they need you.

You need four query types, five queries each.

Category queries. These are the "what is the best X for Y" questions. Someone is looking for a solution and has not decided on a vendor yet. If you sell B2B SaaS for inventory management in Southeast Asia, the query is "best inventory software for ecommerce brands in Indonesia." If you run a premium co-working space in Jakarta, the query is "co-working space Jakarta CBD with meeting rooms." Write five of these. They should feel like the questions you hear in discovery calls or see in support tickets when someone is evaluating you against competitors.

Problem queries. These are the pain points your product solves, phrased as questions. "How do I track stock across three warehouses?" or "Why is my inventory count always wrong at month-end?" Write five. If you have recorded sales calls or a support ticket archive, pull the exact language buyers use. Do not make these up from a keyword tool. Real buyer language beats search volume.

Comparison queries. These name you and a competitor. "ProductA vs ProductB for small teams" or "ProductA vs ProductB pricing." Pick your top three competitors and write one or two queries per competitor. If no one is searching your brand name with "vs" yet, that is fine. The query still tells you whether the AI engine knows enough about your positioning to compare you accurately.

Brand queries. These are your name plus a qualifier. "ProductA review," "ProductA pricing," "ProductA for startups," "is ProductA good for remote teams." Write five. These should reflect what someone types when they have heard of you and are deciding whether to continue.

You now have 20 queries. Write them in a spreadsheet. One query per row. Three columns next to each query for ChatGPT, Perplexity, and Gemini.

Test the queries and score what you find

Open ChatGPT, Perplexity, and Gemini in three browser tabs. Paste the first query into ChatGPT. Read the answer. Record what happened in the spreadsheet.

You are scoring four things for each answer.

Mention. Did the engine name your brand anywhere in the answer? Yes or no. If yes, note whether it was in the prose or only in a citation at the end. ChatGPT cited 9.5 domains per answer and Gemini 3.5 in one 232,838-answer study from June to September 2026, so most queries will return a list. You are looking for whether you made the list at all.

Position. If the engine mentioned you, where? First, second, third, or buried in a longer list? If the answer is a narrative without a numbered list, note whether you were named in the opening paragraph or only later. Position matters. The first name in an AI answer gets the click. The fifth name gets skipped.

Context. What did the engine say about you? Copy the exact sentence into your spreadsheet. You are checking whether the description is accurate, whether it positions you the way you position yourself, and whether the language matches your messaging. If the engine says "ProductA is good for large enterprises" and you sell to startups, that is a signal. If it says "ProductA focuses on pricing transparency" and you have never said that anywhere, you have a content gap or a competitor is dominating that narrative.

Citation. Did the engine cite a source when it mentioned you? If yes, what was the source? Your own site, a review site, a competitor's comparison page, Reddit, a news article? Write it down. Only 11% of domains cited by ChatGPT also appear in Perplexity, so you will see different sources per engine. You want to know which pages the engines trust when they talk about you.

Repeat for Perplexity. Repeat for Gemini. Do this for all 20 queries.

This will take you 60 to 90 minutes if you are doing it for the first time. Do not rush. You are building a baseline. Every audit I run after this one will be faster because you will know what to look for.

What the scores tell you and what to fix first

When you finish testing, you will have a grid. Twenty queries, three engines, 60 data points. Now you score it.

Count how many times you were mentioned per engine. That is your mention rate. If ChatGPT mentioned you in 8 out of 20 queries, your mention rate is 40%. If Perplexity mentioned you in 3 out of 20, your mention rate is 15%. Write these down. You now have a number. The number will go up or stay flat over the next 90 days depending on what you fix.

Count how many times you were in the top three per engine. That is your position rate. If you were mentioned 8 times but only ranked first or second twice, your position rate is 25% of mentions or 10% of total queries. This tells you whether you are winning the queries where you show up or just making the list.

Count how many queries returned zero mentions across all three engines. These are your blind spots. If "best inventory software for ecommerce brands in Indonesia" returned zero mentions on all three engines, you have a content gap. Either you have never written about that question, or the content you wrote is not structured in a way the engines can extract and cite.

Count how many citations pointed to your own site versus third-party sites. If 12 mentions cited your site and 8 cited a competitor's blog or a review site, you have some authority but you do not control the narrative everywhere. If zero mentions cited your site, you have a technical problem or a content problem. The engines can see you but they do not trust your content enough to cite it.

Now you know what to fix.

If your mention rate is below 30% across all engines, you have a content gap. The engines do not know enough about you to name you when someone asks a relevant question. Your first fix is to write answer-shaped content for the queries where you got zero mentions. I cover this in how to write a content brief that gets cited in AI answers.

If your mention rate is above 30% but your position rate is below 20%, you are getting mentioned but not recommended. The engines know you exist but they rank competitors higher. Your first fix is competitive. Find the queries where a competitor ranks first, check what source the engine cited, and write a better answer to the same question. Better means more specific, more recent, or more clearly structured. Perplexity cited content updated within 30 days at an 82% rate, so recency is a lever you can pull.

If your citation rate skews toward third-party sites, you have an authority gap. The engines mention you but they trust someone else's description more than yours. Your first fix is to add structured data to your own pages and make sure your service pages answer the exact questions buyers ask. I walk through the minimum schema setup in the minimum schema setup for a startup that wants AI citations.

If Gemini mentions you but ChatGPT and Perplexity do not, you have strong SEO foundations but weak answer-engine content. Gemini pulls heavily from Google's index. ChatGPT and Perplexity prioritize different signals. Your first fix is to add FAQ schema, inline citations to primary sources, and explicit question-answer pairs in your content.

If ChatGPT mentions you but Perplexity does not, your content is probably older than six months or it lacks the live-web signals Perplexity prioritizes. Perplexity favors recent content and forum-style sources. Your first fix is to refresh your top pages with 2026 dates in the title or heading and consider whether you have a presence in relevant Reddit threads, Quora answers, or industry forums that Perplexity might cite.

What you have now and what happens next

You have a scored baseline. You know your mention rate, your position rate, your citation mix, and your blind spots. You know which engines mention you and which ignore you. You have 20 queries you can re-test in 30 days to see whether the fixes worked.

This audit is not a dashboard. It is a diagnostic. The work happens after the audit, not during it. But you cannot fix what you have not measured, and the 20-query test gives you enough signal to know whether AI search matters for your business and where to start.

If you want help running this audit or interpreting what you find, book a 30-minute call and I will walk you through it. If you want to train your team to run this on their own every quarter, the training offering includes a live cohort version of this exact audit with your actual queries.

Run the audit this week.