How we audit AI visibility: the 42-point rubric, section by section

An AI visibility audit checks two things: whether generative engines can crawl your site and understand what it is about, and whether they cite you when someone asks about your category. We run one for free against a 42-point rubric grouped into five sections, and a person writes it, not a scanner. It takes 48 business hours, there is no mandatory call, and what you get back is a score out of 42 with the failures ordered by how expensive they are to fix. The full rubric is below, section by section, in case you would rather run it yourself.

What does an AI visibility audit actually check?

It does not check rankings. A generative engine does not hand you a position in a list — it composes an answer and cites three to five sources. Either you are among those sources or you do not exist for that question.

That splits the audit in half:

• Can they reach you? Every engine crawls with its own agent and documents how to allow or block it: OpenAI publishes theirs, Perplexity documents theirs, and Anthropic explains its own. An inherited robots.txt that excludes them is a common, silent failure.
• Can they use you? Even when they crawl you, the engine lifts blocks — a definition, a list, an answer to a question — not whole pages. If your page has no liftable blocks, there is nothing to cite.

The context that makes this matter: in the first four months of 2026, 68.01% of US Google searches ended without a single click, up from 60.45% in 2024. The answer is increasingly consumed where it is generated.

Why 42 points instead of an automated score?

Because a single number is not actionable. Each of the 42 points is scored pass or fail, so the output tells you what to fix rather than how badly you are doing.

And because some things a scan cannot judge:

• Whether your FAQ block answers the question a buyer actually asks, or one the marketing team invented.
• Whether your Spanish reads native or reads translated.
• Whether the business listed in three directories is the same business, with the same name and the same address.

That is why the diagnosis takes 48 business hours instead of five minutes: someone opens your site, scores it point by point, and writes up what they found.

What do the 15 technical-foundation points check?

It is the heaviest section, because if an engine cannot crawl, parse or verify your site, nothing else matters. We check:

• An llms.txt at your domain root, declaring what the site is about, its priority pages and its canonical positioning.
• Valid JSON-LD markup, starting with Organization.
• A sitemap and robots.txt that agree with each other and with each engine's agent, documented per crawler.
• Ownership verification in Search Console and Bing Webmaster Tools.
• Internationalization: per-language routes and correct hreflang.
• Core Web Vitals, because a page that does not load does not get cited either.

What do the 10 content-structure points check?

Here we measure whether your content is written in liftable blocks. The ten points cover question-shaped H2s, FAQ blocks with FAQPage markup, numbered lists, definition callouts, author bylines and freshness signals.

The point most often failed is the first one. A page whose H2 reads "Our services" answers nothing; one that reads "What does a GEO audit include?" does, and that is the shape an engine can quote. Google documents how the FAQ block should be implemented so it is machine-readable.

What do the 8 entity-consistency points check?

An engine needs to be confident that every mention refers to one business. We cross-check company name, founder name, address, URL format, LinkedIn profile, directory listings, Wikidata entry and Google Business Profile.

It is the dullest section and the one most often broken. If your directory listing says one city and your site says another, the engine holds two partial entities instead of one solid one — and neither carries enough confidence to name you.

What do the 5 citation-measurement points check?

Without measurement there is no diagnosis, only opinion. The five points are:

1. A written list of target queries — the real questions your buyer would ask.
2. A monthly audit method that can be repeated identically.
3. A scoring rubric: mentioned, cited with a link, recommended.
4. A place where the history is kept.
5. A benchmark against the providers who currently do appear in those answers.

Without this you cannot tell "the work is compounding" apart from "we published more".

What do the 4 distribution points check?

Much of what an engine knows about you does not live on your site. We check presence in forums and communities, listings in sector directories, citations in third-party publications, and knowledge-graph entries.

It is the shortest section and often the highest-leverage one: in competitive categories, answers to "who is the best X agency?" are assembled almost entirely from third-party roundups and directories, not from the provider's own site.

What can you audit yourself today?

In an afternoon, with no paid tools:

• Open yourdomain.com/llms.txt. If it 404s, that is your first lost point.
• Open yourdomain.com/robots.txt and look for the agents each engine documents. Confirm you are not blocking them by accident.
• Take your five most important pages and count how many H2s are question-shaped.
• Search your business name in three directories and compare name, address and URL character by character.
• Ask an engine the five queries your best customer would ask, and note which sources it cites.

The full 42-point checklist is downloadable and is the same one we use. A careful pass takes you three to four hours.

What do you get, and how long does it take?

The free diagnosis returns four things:

• Your readiness score out of 42, section by section.
• The concrete failures we found, ordered by cost to fix.
• What you can fix yourself and what needs development help.
• A proposal with scope and cost, if you want us to do it.

Free, within 48 business hours, in Spanish or English, with no mandatory call.