How BeacX audits are checked

Every BeacX audit is made in two passes. First the engine reads your public pages the way Google and AI answer engines do, and scores them. Then a person opens the live page behind each finding, confirms it, and removes anything that does not hold up. Only what survives the second pass reaches you.

What the engine reads

BeacX is our own audit engine, version 6.7.0 on 20 October 2026. It reads your sitemap and follows your links, one or two pages at a time, within the rules of your robots.txt. Pages built with JavaScript are rendered the way a browser renders them, so a page that looks empty to a simple crawler is not reported as empty.

For each page it records the technical basics (status, redirects, canonicals, indexing rules, headers, speed signals), the content (titles, headings, passages, images, links), the structured data, the author and date signals, and the language. From that it builds the workbook, the verdict and the structured data files.

How the readiness score works

The AI search readiness score is out of 100 and averages four layers, each weighted the same:

1. Crawler access: can AI and search bots fetch the pages? Robots rules per bot, with search and training bots told apart [2], noindex [1], pages that only appear with JavaScript.
2. Entity clarity: can engines tell what the site is? Structured data [3], the declared business type, a consistent name, address and phone.
3. Answer extractability: can an answer be lifted cleanly? A direct answer at the top, passages short enough to quote, question headings, llms.txt.
4. Authorship and trust: can they see who wrote it, and when? Named authors, update dates, credentials, and on health sites a separate medical reviewer.

The weights are equal on purpose. A weighted score invites the question “why is schema worth 30 and trust worth 15”, and nobody can answer that honestly. Equal weights can be explained in one sentence.

Every point in the score depends only on your own pages. That is why the Fix Pack can refund you if the score does not move: we only guarantee what we and your developer control.

Readiness and visibility are measured apart

Whether engines can read you and whether they cite you are different questions. The readiness score answers the first from the crawl. The second is measured by asking Google AI Overviews and ChatGPT search a fixed, agreed set of questions and reading the answers. The two are never averaged together, and engines that answer from memory are never scored on citations, because they cannot cite anyone.

Rules that keep false alarms out

These are written into the engine, so the person checking starts from a clean report:

• A page we could not read, because of a bot-blocker, a timeout or a file too large, is marked “not checked”. It is never called broken.
• A layer we could not measure drops out of the score and is named. It is never counted as zero.
• On a health site, the medical-reviewer check only counts where it applies. A dental practice is never marked down for something that does not apply to it.
• A site with nothing worth fixing is told so. The report never invents a problem to open a conversation.
• Every modelled number, such as extra clicks or points from a fix, is labelled modelled.
• Fields only you know, such as an address in a structured data file, are marked TO FILL IN, never guessed.

The person’s check

Before a report is sent, a person goes through it finding by finding:

1. Opens the live page behind each finding and confirms what the engine saw.
2. Removes findings that do not hold up, and anything that would mislead.
3. Reads the structured data files and corrects any type that does not fit the business. On one of our own sites the engine suggested a local business block, with opening hours, for an online card game; that is the kind of line the check removes.
4. Puts the fixes in order of value and writes the time each one takes.

If a finding we checked still turns out to be wrong, tell us. We correct the report and refund that part of the fee.

How we test the engine

The engine runs 1,174 automated tests before every release. In October 2026 we ran it on 28 real websites (clinics, law firms, local services, online stores, software companies and large publishers, in English, Arabic and Turkish) next to a leading desktop crawler and a leading audit suite, and checked every finding by hand against the live page. Each wrong claim we found became a fix and a test. We will publish accuracy figures once a round on sites the engine has never seen is complete.

What we never claim

We never promise rankings or AI citations. We never report a number we did not measure. We never send a finding nobody has checked.

See four real reports Prices The SEO audit The AI visibility audit Start with the free check