Semakod Audit

The brief

There are plenty of site audit tools and most have one of two problems. Either they dump a hundred technical lines with no order of importance, or — more and more often lately — they let a language model write the recommendations, which sound convincing but say something different on the second run.

On top of that came a new question older tools do not answer at all: how does AI see the site? Does it have an llms.txt, does robots.txt let AI assistants' search bots through, is the content shaped so a direct answer can be pulled out of it? Today this decides visibility much like classic SEO does, yet almost nobody measures it.

The solution

An audit built as a sequence of concrete measurements, not generated text.

Every check has a status and a weight, and every recommendation is written in advance — it is picked based on the measured value. Nothing in the report is generated, so two runs on the same site give the same answer, and every number in the score can be traced back to the check it came from.

An audit runs in four steps. The HTML is fetched and the server response time measured along the way — no browser is launched, so it takes a fraction of a second. In parallel robots.txt, llms.txt and the sitemap are read and up to fifteen internal links are verified; if a source fails, the audit finishes without it. Meanwhile PageSpeed Insights measures mobile and desktop, each in its own request with retries on failure. Finally the score is calculated from the individual statuses and weights.

The score is made up of six areas, each with its own weight:

• Speed (25%) — Core Web Vitals from PageSpeed Insights, measured separately for mobile and desktop
• SEO (25%) — indexability, title, H1, canonical, Open Graph, robots.txt, sitemap
• AEO (12%) — FAQ and HowTo schema, question headings, direct answers, paragraph length
• GEO (13%) — llms.txt, rules for AI bots, entity profiles, verifiable authorship
• Correctness (13%) — viewport, favicon, custom 404, HSTS, broken links
• Accessibility (12%) — landmarks, form labels, accessible names, tab order

How it’s built

Next.js 14 with the App Router and TypeScript, styled with Tailwind. The fetched HTML is parsed by cheerio — a parser, not a browser. Core Web Vitals come from the PageSpeed Insights API, because Google's own numbers are the only ones that match what Google actually ranks by.

What is more interesting is what is not there: no database, no sign-in, no headless browser. The audit is computed afresh on every request and stored nowhere. Everything else follows from that — the tool works without registration, keeps not a single row about other people's sites, and fetching a page takes a fraction of a second instead of the several seconds it would take to start Chromium.

There is one trade-off, and it is stated openly on the site too: sites that render their content only with JavaScript are not fully visible to a static fetch. It is a deliberate exchange — launching a browser would cover that case, but it would be paid for with infrastructure, slower runs and the need to charge for the tool or limit it.

Outcome

A report you can start working from right away: fatal findings at the top, which take priority over everything else (noindex or a block in robots.txt means the site simply is not in search), and below them the numbers broken down to individual checks.

The AEO and GEO sections answer a question clients ask more and more often — whether an assistant will cite them in its answer, or skip them.

The tool is free and needs no sign-up — just paste an address.

50+

concrete checks across six areas

6

weighted areas in the overall score

2 min

from entering an address to the report

0

data stored — the audit is kept nowhere