How does AICOT score a website?
AICOT opens the pages of a site in a real browser, runs a fixed list of checks on them, and turns the failed checks into a score from 0 to 100 for each of seven areas and for the site as a whole. This page describes exactly how.
Scoring version 4 · Last updated
What is measured?
Each audit covers seven areas. The overall score is their weighted average.
| Area | What it covers | Weight |
|---|---|---|
| SEO | Titles, descriptions, headings, structured data, thin and duplicate content. | 20% |
| Content | The text of up to five pages, reviewed by an AI model for clarity, direct answers, concrete facts and trust signals. | 15% |
| Technical | HTTPS, robots.txt, sitemap, broken links, redirects, indexing rules and security headers. | 15% |
| Performance | Loading speed and Core Web Vitals, measured with Lighthouse on a simulated mobile phone. | 15% |
| AI visibility | Whether AI assistants can read and quote the site, and whether they name or cite it when asked customer questions. | 15% |
| Accessibility | WCAG 2.2 level AA rules, tested with axe-core. | 10% |
| Conversion | A call to action on the first screen, a way to get in touch, and forms short enough to finish. | 10% |
How is an area's score calculated?
An area starts at 100 and loses points for each failed check. A check costs more the more severe it is and the larger the share of pages it fails on:
score = 100 × (1 − Σ weight × share of pages failing ÷ Σ weight of all the area's checks)
The weight of a check comes from its severity: critical 10, high 5, medium 2, low 1. A critical problem on every page therefore costs ten times as much as a low one.
Which areas are scored differently?
- Performance is 60% the average Lighthouse performance score of the sampled pages (the home page and one page from each of two other sections) and 40% the performance checks.
- Accessibility rules come from axe-core, an open-ended list, so the weighted failures are measured against a fixed budget of 30 points instead of a list of checks.
- AI visibility is the readiness checks alone. When live testing is requested, it becomes 50% readiness and 50% the result of the test: each answer counts 1 if the site is cited as a source, 0.6 if it is only named, and 0 otherwise.
- Content is the average of four criteria scored by an AI model on each analyzed page. It is only measured when AI analysis is requested.
What happens when an area is not measured?
An area that could not be measured, such as content without AI analysis, is left out, and the remaining weights are scaled up to 100%. An unmeasured area never counts as zero.
Can two audits be compared?
Yes, when they were scored with the same scoring version. Every report states its version. When checks or weights change, the version number goes up, and reports stop comparing scores across versions.
What are the limits of the method?
- An audit covers a limited number of pages, so problems on other pages are not seen.
- Speed is measured in a test environment from one location. Real visitors on other networks and devices will see different numbers.
- AI assistants answer differently from one request to the next, and their apps can differ from their programming interfaces. Live test results are a sample, best read as a trend over several audits.
- The content analysis and the action plan are written by an AI model. They can be wrong and should be checked before they are applied.
- Automated accessibility testing finds only part of all accessibility problems. A score of 100 does not prove a site is fully accessible.