Lumiono

AI visibility audit: what gets tested, and how the score is built

An AI visibility audit answers two questions: can the assistants read you correctly, and do they choose you? The first is technical and fixable in a week. The second is decided by pages you do not own. Here is exactly what we test, how the number is produced, and the half you can do yourself for free.

What an audit is not

It is not a Google ranking report and it is not a mystery score out of 100 from a dashboard. A useful audit produces three artefacts: a pass/fail technical sheet, a prompt matrix showing which brands each model names, and a list of the specific third-party pages that would change the answer.

The 12 checks

#CheckPass condition
1AI crawler accessGPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, Google-Extended allowed in robots.txt
2Server-rendered contentServices, prices and locations visible in raw HTML (no JS needed)
3llms.txtPresent, factual, no adjectives, links to every money page
4Entity schemaOne Organization/LocalBusiness block, sameAs filled, values match visible text
5Service schemaService + Offer with price or price range on each service page
6NAP consistencyIdentical name/address/phone on site, GBP, Apple Maps, top 5 directories
7Answer-shaped pagesEvery money page has question-style H2s with a direct answer underneath
8Sitemap hygieneReal lastmod, no redirects or 404s, under 10k URLs, submitted to Google and Bing
9FreshnessPrices and dated claims are current
10Competitor coverageModel names at least one competitor for the target prompts (proves the category is nameable)
11Brand mention baselineShare of answers naming the brand, measured, not guessed
12Corroboration surfaceAt least 3 independent pages stating the same facts

How the prompt matrix is built

Twenty questions a real buyer in the category would ask an assistant, written the way they speak: category, city, urgency, budget, qualification. Each question is sent to two or three models with the same settings, one clean run each, answers capped so the model cannot pad. Every answer is then searched for the brand name and for each competitor name.

The output is not a vibe. It is a table: 40 answers, brand named in 0, competitors named in 11, and the three answer shapes that produced those 11. That table is the work order — it shows which pages and which third-party mentions actually move the answer.

Reading the score

Two numbers matter: coverage (share of answers naming you) and accuracy (of the answers that name you, how many state your price, area and hours correctly). A brand named in 40% of answers with the wrong phone number is worse off than a brand named in 10% with clean facts — the first sends buyers to a competitor's line.

What you can run yourself, free

What we charge

The audit is included in the $1,500 launch: technical fixes, the llms.txt and schema build, answer-shaped pages, the baseline matrix on day three, and 30 days of tracking. Ongoing tracking is $500/month and reports coverage and accuracy against the same fixed question set, so month two is comparable to month one. If a competitor is named and you are not, the report names the pages responsible.

Questions

How many prompts are needed for a trustworthy audit?

Twenty questions with two or three models, run cleanly, gives 40–60 data points — enough to see which brands are consistently named. Fewer than ten questions mostly measures luck.

Which models should be tested?

At minimum ChatGPT and Perplexity, because they behave differently: Perplexity attaches source links and moves first, ChatGPT names brands without links and moves slower. Add Gemini if Google AI Overviews matter for the category.

Is an AI visibility audit a one-time job?

No. Model answers drift with every retrieval update and every competitor publishing a page. Monthly re-runs of the same fixed question set are what make the trend readable.

Can you guarantee a percentage of answers?

No, and be careful with anyone who does. What can be guaranteed is the work: technical readiness, fact sheets, answer pages and third-party placements, each verified in writing.

What does the free scan cover?

Checks 1–6 on one domain: crawler access, raw HTML content, llms.txt, entity schema, service schema and NAP consistency, with the failing lines quoted.

Related

How to get cited by ChatGPT · What the launch includes · Pricing · Our measured baseline


Get the free AI-readiness audit

Send your domain and service area — you get the exact scope, price and delivery date in writing. Launch is $1,500 one-time, tracking is $500/month.

Run the free audit See pricing