An AI visibility audit answers three questions: for the prompts your buyers ask, how often do ChatGPT, Perplexity and Google AI Overviews name your brand; which pages do they cite instead; and what would have to change for them to cite you. A useful audit ends with a list of specific pages to create or get onto, not a score.
You can run a basic version yourself in a few days with a spreadsheet and a free tracker. Pay for one when you need hundreds of prompts, several engines, or someone to turn the findings into a placement plan. Prices are on the audit pricing page.
What a good audit includes
| Part | What it answers | What you should receive |
|---|---|---|
| 1. Prompt set | Which questions matter for your category | 30–200 prompts grouped by intent: "best X for Y", "X vs Y", "how to", "alternatives to", problem questions |
| 2. Baseline across engines | Where you appear today | For each prompt and engine: named or not, position in the answer, cited or not, competitors named |
| 3. Cited-source map | Which pages the assistants rely on | The URLs and domains cited for your prompts, ranked by how often they appear |
| 4. Gap list | Why the answers go to competitors | Prompts where competitors are named and you are not, with the sources that name them |
| 5. Technical access | Can AI crawlers read your site | robots.txt and firewall check for OAI-SearchBot, PerplexityBot, Googlebot; pages that need JavaScript to show text |
| 6. Content check | Do your own pages answer the prompts | Which prompts have no matching page on your site; pages that are vague about price, use case or comparisons |
| 7. Entity and review check | Is your brand described consistently | Name, category and facts on your site, review sites, Wikipedia/Wikidata if applicable, directories |
| 8. Plan | What to do first | A ranked list: pages to create, third-party pages to get onto, fixes, and how progress will be measured |
The cited-source map (part 3) is the part most audits skip and the one that matters most. In the AI answers we track, assistants cite guides and blog posts most (34% of 12,785 citations in October 2026), then platform documentation (15%) and "best" lists (14%). Which of those pages mention you decides most of the outcome. The breakdown is on the methodology page.
What an agency should deliver
- The raw data, not only charts: prompts, engines, dates, answers and cited URLs, in a file you keep.
- The method: how many times each prompt was asked, on which days, logged in or not, which country. AI answers vary between runs; a single run per prompt is not a baseline.
- A named list of third-party pages to get onto, with why each matters (how often it is cited for your prompts).
- A measurement plan: the same prompt set re-run monthly, so the audit becomes a baseline.
Red flags:
- A single "AI visibility score" with no prompt-level data behind it.
- Screenshots of AI answers without dates or prompts.
- Promised or guaranteed citations. No vendor controls what an assistant cites.
- "Add llms.txt and schema" as the main recommendation. Neither decides which sources an assistant cites; see how to get cited by AI.
DIY checklist: a basic audit in five steps
- Write 30–50 prompts your buyers would ask. Mix category questions ("best expense software for startups"), comparisons ("Ramp vs Brex"), alternatives ("alternatives to Expensify") and problems ("how to stop employees losing receipts"). Use your sales calls and support tickets for wording.
- Run each prompt in ChatGPT (with search), Perplexity and Google (AI Overview or AI Mode), logged out where possible, at least twice on different days. Record: are you named, where, which competitors, which URLs are cited. A free tracker or checker saves time here; see free and paid AI visibility tools.
- Count the cited domains and URLs across all answers. The top 20 URLs are your target list.
- Check each target page: are you on it? Is a competitor? Is it updated regularly? Who publishes it — a media site, a vendor list, a review site, a community thread?
- Check your own site: robots.txt allows OAI-SearchBot and PerplexityBot; key pages show their text without JavaScript; there is a page that answers each prompt type directly (pricing, comparisons, use cases).
Then decide: fix your own pages, get onto the cited third-party pages, or both. If the target list is long, that is where citation services come in.
How long it takes and what it costs
A DIY audit of 30–50 prompts takes two to four working days by hand, less with a tracker. Paid audits vary from fixed one-off packages to the first month of a retainer; the published prices we found are on the audit pricing page, and providers are compared on the citation services page.
FAQ
What does an AI visibility audit include?
A prompt set for your category, a baseline of where you appear in ChatGPT, Perplexity and Google AI answers, a map of the pages those assistants cite, a gap list against competitors, a technical access check, and a ranked plan.
What should I expect an agency to deliver from an AI visibility audit?
The raw prompt-level data with dates and cited URLs, the method used, a named list of third-party pages to get onto, and a measurement plan to repeat the same prompts monthly.
Can I run an AI visibility audit myself?
Yes. Write 30–50 buyer prompts, run each in ChatGPT, Perplexity and Google at least twice, record mentions and cited URLs, and count the most-cited pages. A free tracker or checker speeds up the recording.
How many prompts does an audit need?
Enough to cover each type of buyer question in your category. 30–50 is a workable start for one product line; larger audits use 100–200 prompts across several engines.
Is an AI visibility score enough?
No. A score summarises the data but does not say which pages to change or get onto. Ask for the prompt-level answers and cited URLs behind it.
How is an audit different from a tracking tool?
A tool records where you appear; an audit explains why and says what to do. Many audits use a tool for the recording part. See tools vs audits vs services.