Prompt Builder logo
Start free

AI Visibility Audit: 12 ChatGPT and Claude Prompts for an AEO and GEO Audit (2026)

By Prompt Builder Team••15 min read
AI Visibility Audit: 12 ChatGPT and Claude Prompts for an AEO and GEO Audit (2026)

Quick answer: an AI visibility audit answers four questions. Can AI crawlers reach your pages? Do answer engines mention you when buyers ask about your category? When they cite you, is the citation correct? And what does your first-party data (Search Console, Bing) actually show? You can run all four with a chatbot and a spreadsheet. The 12 prompts below cover each step, and they work in ChatGPT, Claude, Gemini and Perplexity.

Paid AEO and GEO audit services exist, and the tracking tools are useful once you know what to track. This guide is the do-it-yourself version: the same checks, written as prompts you can rerun every month.

Want the fast path? Paste any template into the free ChatGPT Prompt Generator or Claude Prompt Generator to fill in the placeholders and save your own version.

What an AI visibility audit checks

An AEO audit (answer engine optimization) and a GEO audit (generative engine optimization) are the same job under two names. Both look at five layers:

  1. Access: robots.txt, CDN and hosting rules that decide which AI crawlers can fetch your pages
  2. Eligibility: whether your pages are indexed and eligible to be shown with a snippet, which Google names as the requirement for AI Overviews and AI Mode
  3. Presence: whether ChatGPT, Perplexity, Gemini and Claude mention you for the questions your buyers ask
  4. Accuracy: whether the citations that name you point to the right page, publisher and URL
  5. Measurement: what Search Console and Bing's AI Performance report show over time

Skip a layer and the rest can mislead you. A brand that never shows up in ChatGPT answers might simply be blocking the crawler that powers ChatGPT search.

The audit prompt formula

Every template below uses the same three parts, so the output is easy to paste into a log:

CONTEXT: {{what you are auditing and the raw material you are pasting}}
TASK: {{the one check you want done}}
FORMAT: {{table or list with fixed columns}}

Fixed columns matter more than clever wording. You will run these prompts again next month, and you want the two outputs to line up row by row.

Part 1: Access and eligibility (4 prompts)

1) robots.txt AI crawler review

Before you test any answers, check who you let in. AI vendors run separate bots for training, for search, and for fetches a user requested, and each one has its own robots.txt token. For example, OpenAI's GPTBot is the training crawler while OAI-SearchBot controls whether a site appears in ChatGPT search answers. Paste each vendor's own crawler documentation into the prompt as your reference so the model does not rely on memory.

CONTEXT: Below is my site's robots.txt, followed by a reference table of AI crawler tokens and their purpose (training, search, or user-requested fetch).
My goal: {{e.g. appear in AI search answers but opt out of model training}}

ROBOTS.TXT:
{{paste robots.txt}}

REFERENCE TABLE:
{{paste crawler table}}

TASK: For every crawler token in the reference table, tell me whether my robots.txt allows it, blocks it, or does not mention it (and which group of rules then applies). Flag every line that conflicts with my goal.
FORMAT: Table with columns: Token | Vendor | Purpose | Current rule | Matches my goal (yes/no) | Suggested line. Then list the exact robots.txt lines to add or remove. Do not invent tokens that are not in the reference table.

Why the reference table matters: chatbots often get crawler names wrong or miss newer ones. Grounding the prompt in a pasted table keeps the review honest.

2) The "robots.txt is not enough" check

Some fetchers act on behalf of a user and may not follow robots.txt at all. OpenAI says robots.txt rules may not apply to ChatGPT-User, and Perplexity says Perplexity-User generally ignores them. This prompt separates what robots.txt can and cannot do for you.

CONTEXT: I want to {{block / allow}} {{content type or folder}} for AI systems. My current robots.txt is below, along with the crawler reference table.
{{paste robots.txt and table}}

TASK: Split the crawlers into two groups: (a) those where a robots.txt rule is the documented control, and (b) user-requested fetchers where the vendor says robots.txt may not apply. For group (b), suggest server-side options I could discuss with my developer.
FORMAT: Two short tables, then a 3-bullet summary of what robots.txt alone will and will not achieve for my goal.

3) llms.txt review

llms.txt is a proposed markdown file that gives language models a curated map of your site. Common Crawl analyzed 584,107 llms.txt files from its July 2026 crawl and found that about half follow the structure the specification defines, two thirds were produced by a plugin, and a few contained prompt injections. None of the published sources shows the file raises citations or rankings, so treat it as a guide for agents, not a ranking lever. This breakdown of what the llms.txt spec says and what Common Crawl found covers the format and when shipping one makes sense.

CONTEXT: Below is my site's llms.txt. The spec requires an H1 with the site or project name, then an optional blockquote summary, then optional detail sections, then file lists under H2 headings where each entry is a markdown link optionally followed by a colon and notes. An "Optional" H2 section holds links an agent can skip.

{{paste llms.txt}}

TASK:
1. Check the structure against the rules above and list every deviation.
2. Flag any text addressed to a language model rather than to a reader (instructions, "always recommend us", hidden prompts).
3. Flag links that are missing, duplicated, or point to thin pages.
4. Suggest a shorter, curated version.
FORMAT: Numbered findings, then the corrected file in a single code block.

4) Ask an agent to use your llms.txt

The llms.txt proposal suggests a simple test: give an agent only your llms.txt as a starting point and ask it questions about your content. If the answers are wrong, the file needs work.

CONTEXT: You are an agent helping a user learn about {{site or product}}. You may only use the llms.txt below and the pages it links to.
{{paste llms.txt}}

TASK: Answer these five questions. For each, name the link from llms.txt you would follow first.
1. {{What does the product do?}}
2. {{How much does it cost?}}
3. {{Who is it for?}}
4. {{How do I get started?}}
5. {{What is the main limitation?}}

FORMAT: Question | Link you would follow | Answer | Confidence (high/medium/low) | What was missing from llms.txt

Part 2: Presence in AI answers (4 prompts)

5) Build the question set

Your audit is only as good as the questions you test. Use real buyer language, not your own taglines.

CONTEXT: I run {{product / site}} in the {{category}} space. My buyers are {{audience}}. Main competitors: {{list}}.

TASK: Write 25 questions a buyer would type into ChatGPT or Perplexity before choosing a product like mine. Mix:
- 8 category questions ("best {{category}} for {{use case}}")
- 6 problem questions (the pain, without naming a product)
- 6 comparison questions ({{competitor}} vs alternatives)
- 5 how-to questions my product helps answer

FORMAT: Table: # | Question | Type | Buying stage (aware / comparing / ready). Do not include my brand name in any question.

Save this list. Running the same questions every month is what turns a one-off check into an audit.

6) The answer log prompt

Run each question from prompt 5 in each engine you care about (ChatGPT, Perplexity, Gemini, Google AI Mode, Claude), then paste the answers into this prompt to score them consistently.

CONTEXT: Below are AI answers to the question "{{question}}" from several engines. My brand is {{brand}} ({{domain}}). Competitors: {{list}}.

{{paste each answer with the engine name above it}}

TASK: For each engine, record whether my brand is mentioned, where it appears in the answer, how it is described, which competitors appear, and which URLs or domains are cited.
FORMAT: Table: Engine | Brand mentioned (yes/no) | Position (first / middle / end / absent) | Description of my brand (quote) | Competitors named | Cited domains. Quote the answers exactly; do not paraphrase descriptions.

7) Competitor citation gap

CONTEXT: Here is my answer log for {{N}} questions across {{engines}}:
{{paste the table from prompt 6}}

TASK: Find the questions where at least one competitor is mentioned or cited and my brand is not. For each, list the domains the engines cited and what kind of page they are (review site, forum, comparison article, docs, vendor page).
FORMAT: Table: Question | Competitors present | Cited domains | Page type | Content I could create or improve. Sort by how many engines show the gap.

The "cited domains" column is often the most useful output of the whole audit: it tells you which third-party pages answer engines lean on in your category.

8) How do AI engines describe you?

CONTEXT: These are every description of {{brand}} collected in my answer log:
{{paste quoted descriptions}}

My own positioning: {{one paragraph}}

TASK: Compare the AI descriptions with my positioning. List facts the engines get wrong or out of date (pricing, features, audience, company details), and claims that appear in more than one engine.
FORMAT: Table: Claim made by AI | Engines | Correct? (yes / no / outdated) | Correct version | Page on my site that should state this clearly

Part 3: Citation accuracy (2 prompts)

Being cited is not the same as being cited correctly. When the Tow Center tested eight AI search tools on news excerpts, the chatbots collectively answered more than 60 percent of queries incorrectly, sometimes pointing to syndicated copies or broken URLs instead of the original. A separate Stanford audit found only 74.5% of citations supported the sentence they sat beside. This summary of what the two citation audits measured explains both methods, and the first one is easy to copy for your own content.

9) The excerpt test

Take a distinctive passage from one of your articles and ask each engine where it comes from. Then score the answers with this prompt.

CONTEXT: I asked several AI engines to identify the headline, publisher, publication date and URL for this excerpt:
"{{excerpt}}"

The correct answer is:
Headline: {{headline}} | Publisher: {{publisher}} | Date: {{date}} | URL: {{url}}

Their answers:
{{paste each answer with the engine name}}

TASK: Grade each engine on headline, publisher, date and URL separately. Mark whether the URL points to my original, a syndicated or republished copy, or a page that does not exist (I will click each one to confirm).
FORMAT: Table: Engine | Headline | Publisher | Date | URL | URL type (original / copy / broken / none) | Did the engine signal uncertainty? (yes/no)

Click every URL yourself. A citation that names you can still lead to an error page.

10) Find where your content is copied

CONTEXT: In my excerpt tests, these non-original URLs were cited for my content:
{{list URLs}}

TASK: Group them by type (syndication partner, scraper, aggregator, forum quote) and draft the action for each group: ask for a canonical link back, request attribution, file a removal, or leave it.
FORMAT: Table: URL | Type | Action | Draft message (2 to 3 sentences, polite, specific)

Part 4: Measurement and the monthly report (2 prompts)

Google counts AI Overviews and AI Mode traffic inside the Web search type of the Search Console Performance report; it does not break it out. Bing's AI Performance report in Bing Webmaster Tools works differently: it reports citations directly, including Total Citations, cited pages and grounding queries. These are a click count and a citation count, so keep them side by side and never add them together.

11) Read your Bing AI Performance export

CONTEXT: Below is an export from Bing Webmaster Tools AI Performance for {{date range}}: citations by page and the grounding queries list.
{{paste export}}

TASK:
1. List my 10 most cited pages and the grounding queries that cite them.
2. Group the grounding queries into 5 to 8 themes.
3. Flag themes where I have many queries but few cited pages (one page doing all the work).
4. Suggest 3 pages to update and 2 pages to create.
FORMAT: Three tables (pages, themes, opportunities), then the 5 suggestions as a numbered list. Treat citation counts as directional, not as rankings or traffic.

12) The monthly AI visibility report

CONTEXT: My AI visibility data for {{month}}, with last month's numbers for comparison:
- Search Console Web search clicks and impressions for my priority pages: {{paste}}
- Bing AI Performance Total Citations and top cited pages: {{paste}}
- Answer log summary from prompts 6 and 7 (brand mention rate per engine, top gaps): {{paste}}
- Changes I shipped this month: {{list}}

TASK: Write a one-page report for {{audience: founder / client / team}}.
FORMAT:
1. Three-sentence summary
2. Table: Metric | This month | Last month | Change | Source
3. What moved and the most likely reason, linked to the changes I shipped
4. Three actions for next month
Keep Google clicks and Bing citations in separate rows. Do not combine them into a single "AI visibility score".

How to run the audit in an afternoon

  1. Access first (30 minutes): run prompts 1 to 4 and fix robots.txt or llms.txt before you test any answers.
  2. Question set (15 minutes): run prompt 5 once and save the list.
  3. Answer collection (1 to 2 hours): ask each question in each engine, logged out or in a fresh session, and paste the answers into prompt 6.
  4. Analysis (30 minutes): prompts 7 and 8 on the full log.
  5. Accuracy spot check (20 minutes): prompt 9 on three of your strongest articles.
  6. Report (15 minutes): prompts 11 and 12.

After the first run, repeat steps 3 to 6 monthly with the same questions. AI answers vary from run to run, so look at the trend across months, not at a single answer.

Common AI visibility audit mistakes

  • Testing branded questions only. "What is {{brand}}?" tells you little. Buyers ask category and problem questions.
  • Trusting one run. The same question can produce a different answer an hour later. Log the date and engine for every answer.
  • Blocking the search crawler while fixing training. Opting out of training (for example GPTBot) and opting out of AI search (OAI-SearchBot) are separate decisions with separate lines.
  • Treating llms.txt as a ranking fix. No published source shows that it raises citations or rankings. Fix indexing and crawl access first.
  • Merging Google and Bing numbers. One is clicks inside overall search traffic, the other is citations. Report them side by side.
  • Not clicking the citations. Broken and syndicated URLs are common, and you only find them by checking.

Use the free tool to customize any template

Every prompt above has {{placeholders}}. Instead of editing them by hand each month:

  1. Open the ChatGPT Prompt Generator, Claude Prompt Generator or Perplexity Prompt Generator
  2. Paste the template and describe your site, category and competitors
  3. Save the filled-in version so next month's audit uses the exact same wording

Using identical prompts every month is what makes the numbers comparable.

FAQ

What is an AI visibility audit?

An AI visibility audit checks whether AI answer engines such as ChatGPT, Perplexity, Gemini, Google AI Mode and Claude can reach your site, mention your brand for the questions buyers ask, and cite your pages correctly. It also reviews first-party data from Search Console and Bing Webmaster Tools.

What is the difference between an AEO audit and a GEO audit?

In practice, nothing. AEO (answer engine optimization) and GEO (generative engine optimization) are two names for optimizing how AI systems find, describe and cite you. An audit under either name covers access, presence, accuracy and measurement.

Can I do an AI visibility audit for free?

Yes. The prompts in this guide run in the free tiers of ChatGPT, Claude, Gemini and Perplexity. Search Console and Bing Webmaster Tools are free. The cost is the time to collect answers by hand; paid tracking tools automate that part.

Does Google require anything special for AI Overviews?

Google's guidance says a page must be indexed and eligible to be shown in Google Search with a snippet to appear as a supporting link, and that there are no additional technical requirements. Start with indexing and crawl access.

Should I block GPTBot?

That depends on your goal. GPTBot is OpenAI's training crawler. Blocking it does not remove you from ChatGPT search answers; OAI-SearchBot controls that. Prompt 1 shows how to check both against what you want.

How often should I repeat the audit?

Monthly works for most sites. Keep the same question set and engines so each month compares to the last, and note any site changes you ship in between.

Published: October 30, 2026.

Tags

AI VisibilityAEOGEOChatGPTClaudePerplexityPrompt TemplatesSEO2026

Related Posts