Methodology

Two kinds of checks

The audit checks your site for AI assistants in general: whether AI crawlers may fetch it (robots.txt), whether the content is in the HTML they read, llms.txt, structured data and answer-ready content. These checks apply to every assistant.

The citation tests ask two engines only: Claude and Perplexity. Beakon does not query ChatGPT, Gemini or other assistants.

How a citation test is scored

You choose the queries: typed by you, or generated from your site and reviewed by you before any test. Each query is asked to each engine several times, because AI answers change from one run to the next: 2 runs per query and engine for a test, 3 for weekly tracking, 1 for the free mini test, which asks Perplexity only.

On the Claude side, a Claude model with web search answers, at most two searches per answer. On the Perplexity side, the fast preset of Perplexity answers, with web search. We do not name the models here because the providers update them: the exact model of each answer is kept with the result.

A result reads "cited 2 times out of 3": the number of answers that mention your site, out of the answers received. A failed run is not counted.

Brand named, or site cited as a source

Two counters are kept apart. "Brand named": the answer's text names your brand, your domain or one of the other names you entered (a product line, a sub-brand). "Site cited as a source": one of the sources the answer cites is a page of your site.

An assistant often names a brand without linking to it, and sometimes uses a page as a source without naming the brand. Both matter, for different reasons: the first is what the user reads, the second is what sends visitors.

Brand names that are common words

Some brands are also ordinary words (Orange, Slack, Booking). For them, detection is strict: the word counts only written as a name, with a capital letter, outside the start of a sentence, not followed by a generic noun ("booking sites") and not inside a title. The full domain always counts.

The known limit: a real mention at the very start of a sentence ("Orange is the safest choice") can be missed. We prefer missing a mention to inventing one. Adding a more specific alias, such as a product or sub-brand name, helps.

AI visibility on the raw HTML, SEO on the rendered page

Many AI crawlers do not run JavaScript: they read the HTML your server sends. The AI visibility score is therefore measured on that raw HTML. A page whose text only appears once JavaScript runs is reported, and an empty home page caps the AI score.

Classic SEO is measured on the rendered page (with the deep audit), as Google reads it.

This rule is scoring method 2. When your history spans two methods, the chart and the comparisons show a method change, never a drop of your site.

What the results depend on

Assistants name well-known brands far more often than niche ones. A low rate for a young brand says as much about its notoriety as about its site.

Answers vary between runs, between days and between engines. With a few runs per query, a single result is a sample, not a measure: read the trend over several weeks and across queries rather than one number.