To audit brand visibility on LLMs yourself, ask ten fixed buyer questions in ChatGPT, Perplexity, Gemini and Google's AI answers, then score every answer on four things: were you named, where, was your site cited, and was it accurate. With three AI tools, one person can do this in about an hour with a spreadsheet and no paid tool.
This page gives you the ten prompts, the scoring sheet and the rules for reading it. It also shows what happened when we asked five AI tools how to run this audit, so you can see what they recommend and how much they disagree. Visiby sells an AI visibility platform and a managed service, so we say where a tool helps and where it does not.
01 — Four checksWhat does a DIY AI brand audit check?
It checks four things in every answer an AI tool gives to a question your buyers ask. We define them once here and use them the same way on the rest of the page.
| What you check | What it means | Why it matters |
|---|---|---|
| Named | Your brand appears in the answer text (a mention). If the answer presents you as a pick, that is a recommendation | You cannot be chosen if you are not named |
| Position | Where you appear: in the first three names, later, or not at all | Buyers read the top of a list first |
| Cited | Your own page is linked as a source (a citation) | Shows which page the tool relied on, and sends clicks |
| Accurate | What the answer says about you is right: product, audience, price, location | A confident wrong answer can cost you a sale |
These four do not move together. An answer can name you without linking to you, or link your page without naming you. Keep them in separate columns and never add them into one score you cannot unpick. For how each number is usually calculated, see AI share of voice and citation rate.
02 — Before you startWhat do you need before you start?
You need an hour, a spreadsheet, a short list of competitors and access to the AI tools your buyers use. Copy our scoring sheet or make one with these columns: date, AI tool, model or mode if shown, signed in or not, country, prompt, prompt type, answered or not tested, named, position, cited, accurate, competitors named, pages cited, wrong claims, action.
Set the conditions before you open any AI tool, and keep them the same for the whole run:
- Use a private browser window and a new chat for every question, so one answer does not shape the next.
- Sign out where the tool allows it. If a tool will not answer signed out, sign in and write that down.
- Use the same country, language and device for every answer.
- Turn web search on in ChatGPT if it offers the choice. OpenAI says "Responses that use web search may include citations" (OpenAI Help), and you need those links to score the cited column.
- Write down your correct facts first: what you sell, who it is for, your prices and where you work. You will score accuracy against this list.
A one-hour plan:
| Minutes | Step |
|---|---|
| 0 to 10 | Fill in your ten prompts and your fact list |
| 10 to 40 | Ask each prompt in each tool and paste the answers into the sheet |
| 40 to 50 | Score named, position, cited and accurate for each answer |
| 50 to 60 | Sort by the weakest scores and pick the first two fixes |
If one hour is too short for five tools, run three: ChatGPT, Perplexity and Google. Then add Gemini and Google AI Mode next month.
03 — The 10 promptsWhich 10 prompts should you ask?
Ask seven questions that do not contain your brand name and three that do. Report the two groups separately. A question with your name in it will almost always get an answer about you, so mixing them into one rate makes you look more visible than you are.
Fill in the brackets with your own category, buyer and competitors, then never change the wording between runs.
| No. | Type | Prompt |
|---|---|---|
| 1 | Discovery | What are the best [category] options for [type of buyer]? |
| 2 | Discovery | Which [category] companies would you recommend in [country or city]? |
| 3 | Discovery | How do I solve [problem your product solves]? |
| 4 | Discovery | What are the best alternatives to [main competitor]? |
| 5 | Discovery | [Competitor A] vs [Competitor B]: which is better for [use case]? |
| 6 | Discovery | What should I look for when choosing a [category] provider? |
| 7 | Discovery | Which [category] tools are best for [industry or team size]? |
| 8 | Brand | What is [your brand] and who is it for? |
| 9 | Brand | How much does [your brand] cost and what is included? |
| 10 | Brand | What are the downsides of [your brand]? |
Prompts 1 to 7 tell you whether AI tools put you in front of new buyers. Prompts 8 to 10 tell you what they say once a buyer already knows your name, which is where wrong prices and old product names show up.
For more prompts to choose from, see AEO prompts. If you only need to check ChatGPT, monitoring ChatGPT brand mentions has a free method for that one tool; this page covers five.
04 — Running each toolHow do you run the prompts in each AI tool?
Run the same ten prompts in each tool and note one thing that differs per tool. We saw these differences on 4 November 2026.
| AI tool | What to note |
|---|---|
| ChatGPT | Answered signed out. Web search was not always used, so some answers had no links at all |
| Perplexity | Signed out, it asked us to sign up and gave no answer. We ran it on a free account instead |
| Gemini | Answered signed out. The model name (Gemini 3.5 Flash-Lite on our run) is shown on the page, so write it down |
| Google AI Overviews | Not every search shows one. One of our five questions showed no AI Overview, so record "no AI Overview", not "not named" |
| Google AI Mode | Answered signed out. Most answers listed source cards; one listed none |
Two more rules from the tools' own help pages. Google says "AI Mode and AI Overviews may use different models and techniques, so the set of responses and links they show will vary" (Google Search Central), so score them as two tools, not one. For Gemini, Google says its source links "may include content that is related to parts of Gemini's response" (Gemini Apps Help). That means a link under a Gemini answer is not proof the page was used to write it, so mark it as "link shown" in your notes.
A question that gets no answer or an error is not tested. Never score it as "not named", or your rate will look worse than it is. If a tool only answers after you sign in, keep those answers in their own group and say so.
05 — ScoringHow do you score each answer?
Give each answer 0, 1 or 2 points on each of the four checks. Leave accurate blank when you are not named, because there is nothing to check.
| Check | 0 | 1 | 2 |
|---|---|---|---|
| Named | Not named | Named in passing or in a long list | Recommended as a pick |
| Position | Not named | Named, but not in the first three | In the first three names |
| Cited | None of your pages linked | A page about you on another site is linked | One of your own pages is linked |
| Accurate | A key fact is wrong: price, product, audience or location | Mostly right, but missing or out of date | Right on every fact you checked |
Then turn the points into four rates for the discovery prompts, and the same four for the brand prompts:
- Mention rate: answers where named is 1 or 2, divided by answered runs
- Top-three rate: answers where position is 2, divided by answered runs
- Own-site citation rate: answers where cited is 2, divided by answered runs
- Accuracy rate: answers where accurate is 2, divided by answers that named you
Always show the denominator. For example, "named in 6 of 28 answered runs, 7 runs not tested" tells a manager far more than "21%". With 28 answers, one answer moves a rate by more than three points, so do not read a small change as a trend. For how many runs a rate needs before you trust it, see are AI visibility tools accurate.
06 — AI checkWhat did AI tools say when we asked?
We asked five AI tools the five questions behind this page and logged every answer. The table shows what appeared. It shows what the tools said, not why they said it.
Method: five questions in five AI tools on 4 November 2026, one new chat per question, English, run from India with US settings on Google. ChatGPT, Gemini (3.5 Flash-Lite), Google AI Overviews and Google AI Mode were signed out. Perplexity would not answer signed out, so those five answers come from a signed-in free account.
| AI tool | Answered | Visiby named |
|---|---|---|
| ChatGPT (signed out) | 5 of 5 | 0 |
| Gemini 3.5 Flash-Lite (signed out) | 5 of 5 | 0 |
| Google AI Mode (signed out) | 5 of 5 | 0 |
| Google AI Overviews (signed out) | 4 of 5, one search showed no AI Overview | 0 |
| Perplexity (signed in, free account) | 5 of 5 | 0 |
| Total | 24 of 25 planned | 0 of 24 |
All five tools agreed on the basics. All five answers to "What should I score in an AI brand audit?" said to score all four checks rather than pick one. All four answers to "How do I audit my brand's visibility myself?" said to use a fresh or private session.
They did not agree on the numbers:
| Question | What the five tools suggested |
|---|---|
| How many prompts for a full audit? | 20 to 30 (ChatGPT, Perplexity) or 20 to 50 and 30 to 50 (Google AI Mode, Gemini) |
| How many prompts in one hour? | 5 to 10 (Google AI Overviews), 8 to 12 (Perplexity), 10 (ChatGPT) |
| How to weight the four checks? | Equal weights (ChatGPT); accuracy and mentions 25% each, position and citations 20% each, sentiment 10% (Perplexity); accuracy 35%, mentions and position 40% together, citations 25% (Google AI Mode); no weights, but accuracy first (Gemini) |
| Which free tools or templates? | 26 different tools and templates across the five answers. The two most named, both free one-off checkers, were each in 3 of 5 answers. None was in all five. |
So no AI tool gives you a standard you can copy. Pick one method, keep it fixed, and compare yourself with your own last run.
Two answers, from Perplexity and Gemini, also suggested a quick check worth adding: take the AI tool's description of your brand and swap in a competitor's name. If the sentence still reads true, the description is too generic to win you a buyer.
07 — Fix firstWhat should you fix first?
Fix wrong facts first, then gaps in who names you, then citations. A tool that names you with the wrong price does more harm than one that leaves you out.
- Wrong facts. For each wrong claim, write the prompt, the tool, the wrong text and the correct fact. Update the page on your site that should state that fact, and check the pages the answer cited. If an old directory listing or review page has your old price, ask for it to be corrected.
- Not named on discovery prompts. Look at which competitors were named instead and which pages were cited for them. Those are the pages the tools cited for that question, so start your reading there.
- Named but not cited. Check that a clear page on your site answers that question in plain words, can be crawled, and states the facts the answer used.
- Generic descriptions. If the competitor-swap test passes, make your difference plain on your homepage and product pages in one sentence a buyer would repeat.
Then run the same ten prompts again next month and compare the rates, prompt by prompt. If you want to go deeper on reading the results, see how to read an AI visibility report.
08 — Tool or teamTool or team?
A one-hour audit tells you whether you have a problem. It does not fix it, and it does not watch for changes week to week.
If you want the work done for you, the Visiby Growth Engine tracks your buyer questions every week in five AI tools (ChatGPT, Perplexity, Google AI Overviews, Google AI Mode and Claude), then publishes the pages and fixes and reports what changed. It starts at $1,000 a month, billed monthly, with no contract. It does not track Gemini, and it does not promise rankings.
If you want to see where you stand first, run the free check. It shows what ChatGPT says about your brand for three real buyer questions, takes about three minutes, and gives you a baseline to compare with your own audit.
09 — FAQFrequently asked questions
10 — SourcesSources
- OpenAI Help Center, Searching the web with ChatGPT, checked 4 Nov 2026
- Google Search Central, AI features and your website, updated 10 December 2025, checked 4 Nov 2026
- Gemini Apps Help, View related sources from Gemini Apps, checked 4 Nov 2026
- Visiby AI check, 25 planned runs of five questions in ChatGPT, Perplexity, Gemini, Google AI Overviews and Google AI Mode, 4 Nov 2026
Arun Pandit is the founder of Visiby, an AI-visibility tracker by FNA Technology that measures how often ChatGPT, Perplexity, and Google AI Overviews cite a brand. He writes about generative engine optimization from the data Visiby collects across the brands it tracks. View full profile →