The AI Visibility Guide

Why discovery changed

For two decades, being found meant ranking in a list of links. The searcher did the evaluation: scan ten results, open a few, decide. Your job was to be one of the ten.

Assistants collapse that. Someone asks for the best tool for a job and gets two or three names with reasons. There is no page two, and no opportunity for the reader to discover you further down. You are either in the answer or absent from it.

That shifts what wins. A ranked list rewards relevance to a query. A recommendation rewards describability — whether a model can state what you are without hedging.

The signals that matter

They fall into four groups, which is exactly how our score is structured. Access: can a crawler fetch and parse the page at all. Clarity: does the copy state the category, the audience, and the outcome. Answerability: is the content shaped so a passage can be lifted and used. Proof: is there evidence a system can check.

The weighting is not equal. Clarity and answerability carry 30% each in our methodology because they determine whether you can be described and quoted, which is upstream of everything else. Access is 20% — necessary but usually already satisfied. Proof is 20%, weighted lower because it takes longest to build.

Auditing your own site

You can do a useful manual pass in fifteen minutes. Fetch your homepage without JavaScript and confirm the headline and description are in the raw HTML. Read the first screen and ask whether a stranger could name your category from it. Search your page source for application/ld+json to see whether structured data exists. Check whether any page answers a specific buyer question completely.

The automated version runs 32 such checks and reports which passed — see the methodology for the full list, or run it directly on your site.

Fixing what you find

Order matters more than effort here. Unblock access first. Then write the category sentence and propagate it to the title tag, meta description, and llms.txt. Then publish an FAQ block with FAQPage schema — the single fastest way to become quotable. Then a comparison page naming real alternatives.

Proof comes last in sequence and takes longest. Start with one named customer story; it outperforms any amount of anonymous praise. The Evidence Ladder explains how the tiers are weighted.

Measuring progress

Track two things separately. Structural signals — reproducible, fast-moving, fully in your control. And actual mentions — noisy, slower, but the outcome you care about.

For the second, fix a set of realistic category prompts and re-run them monthly across engines, recording whether you are named. Use buyer phrasing, not your brand name: asking "what is [product]" tests recall, not recommendation.

Common mistakes

Writing more content instead of clearer content. Volume does not fix an unclear category statement. Blocking every AI crawler indiscriminately, which removes you from live answers while only intending to opt out of training. Gating the material that would prove your expertise. And treating an AI visibility score as a growth metric — it measures discoverability, which is necessary but not sufficient.