The short version
We read your homepage and up to seven other pages in a real browser, run a Google Lighthouse test on a simulated phone, and check your security, search and email set-up. Each problem we find gets a severity. Each area (security, SEO, speed and so on) starts at 100 and loses points for its problems. The overall score is the weighted average of the areas, with extra deductions for the few problems that stop a site working at all.
Eight areas, weighted
| Area | Weight | What it covers |
|---|---|---|
| AI readiness AI search: whether AI crawlers (ChatGPT, Perplexity, Claude) are let in by robots.txt and the firewall, whether pages can be quoted, content that only appears after JavaScript, structured data naming the business, question-and-answer content, llms.txt. AI agents: whether agents can discover, read and use the site (Markdown for agents, Content Signals, Link headers, DNS discovery, and API, OAuth and MCP discovery on sites with a login or API). | 20% | AI search: whether AI crawlers (ChatGPT, Perplexity, Claude) are let in by robots.txt and the firewall, whether pages can be quoted, content that only appears after JavaScript, structured data naming the business, question-and-answer content, llms.txt. AI agents: whether agents can discover, read and use the site (Markdown for agents, Content Signals, Link headers, DNS discovery, and API, OAuth and MCP discovery on sites with a login or API). |
| Security HTTPS and certificate health, security headers, exposed files, email spoofing protection (SPF/DMARC), cookie consent. | 16% | HTTPS and certificate health, security headers, exposed files, email spoofing protection (SPF/DMARC), cookie consent. |
| SEO Titles, descriptions, headings, duplicates across pages, sitemaps, indexing mistakes, structured data. | 16% | Titles, descriptions, headings, duplicates across pages, sitemaps, indexing mistakes, structured data. |
| Speed A Lighthouse run on a throttled phone, server response, page weight, images, blocking scripts. | 16% | A Lighthouse run on a throttled phone, server response, page weight, images, blocking scripts. |
| Accessibility Colour contrast, labels, alt text, keyboard access, checked in a real browser with the axe engine. | 12% | Colour contrast, labels, alt text, keyboard access, checked in a real browser with the axe engine. |
| Mobile Sideways scrolling, tiny text, viewport, tap-to-call. | 8% | Sideways scrolling, tiny text, viewport, tap-to-call. |
| Content & conversion Calls to action, contact routes, trust signals, readability, stale content, local-business signals. | 7% | Calls to action, contact routes, trust signals, readability, stale content, local-business signals. |
| Technical health Broken links and pages, failed files, JavaScript errors, domain and certificate expiry. | 5% | Broken links and pages, failed files, JavaScript errors, domain and certificate expiry. |
AI readiness carries the most weight: more and more people ask ChatGPT, Claude or Perplexity instead of searching, and being the business those assistants find and recommend now matters as much as ranking on Google. Security, SEO and speed come next because they decide whether people can find, trust and use a site.
How serious is a problem?
The site is failing at something basic: it can't be found, can't be trusted or can't be used.
A clear problem that costs visitors, rankings or trust right now.
A real improvement that makes the site stronger, but nothing is broken.
Polish. Shown, but never counted against your score.
Severity depends on how much of the problem there is. One link with no accessible name is a small detail. Forty of them is a real accessibility problem. The same goes for missing image text, unlabelled form fields, low-contrast text, broken links and blocking scripts. A problem found on every page you have costs more than the same problem on one page (between 60% and 100% of its points, depending on how many pages it affects).
The formula
- Each area: add up the points for its problems, then the area score is
100 × e(−points ÷ 60). It falls quickly for the first few problems and then slows, so one more problem on an already-struggling area matters less than the first did, and no area ever hits a hard zero. - Overall: the weighted average of the eight area scores, using the weights above.
- Critical problems: take an extra 10 points off the overall score each (at most 25 in total).
- Showstoppers: a few problems mean the site effectively doesn't work, so the overall score can't go above 45 while one is present.
- Small details never change the score.
Overall bands: 90–100 Excellent (grade A) · 80–89 Good (grade B) · 68–79 Fair (grade C) · 50–67 Needs work (grade D) · 0–49 At risk (grade F).
AI readiness
AI readiness is scored like every other area: each thing that's missing is an issue with a severity, and the area score comes from the same formula. It has two parts, shown on the same page of your report.
- AI search: whether AI assistants can reach your pages, tell who you are and quote you. Blocking the AI search crawlers, or telling search engines not to quote your pages, is serious; most of the rest is worth improving, and a few (like llms.txt) are small details.
- AI agents: whether agents can discover, read and use the site. A missing Markdown version or missing Content Signals is worth improving; Link headers and DNS discovery are small details. On a site with a login or API, missing API, OAuth and MCP discovery count too; on a plain content site they don't apply. robots.txt and the sitemap are scored once, under SEO.
- Shown but not scored: commerce protocols (x402, UCP, ACP and others), Web Bot Auth and A2A agent cards.
The showstoppers
- There is no real website at the address (a hosting placeholder or an empty page)
- The homepage tells Google not to index it
- robots.txt blocks every search engine
- The security certificate has expired
- The domain name has expired
- The site isn't served over HTTPS
A worked example
A site with two moderate SEO problems, one serious accessibility problem, one serious speed problem and one moderate content problem:
| Area | Points off | Area score | Weight |
|---|---|---|---|
| Security | none | 100 | 16% |
| SEO | −6 | 90 | 16% |
| AI readiness | none | 100 | 20% |
| Speed | −8 | 88 | 16% |
| Accessibility | −8 | 88 | 12% |
| Mobile | none | 100 | 8% |
| Content & conversion | −3 | 95 | 7% |
| Technical health | none | 100 | 5% |
Overall (the weighted areas): 95 (Excellent). Add one critical content problem and it drops to 83. Make that critical problem “the site isn't served over HTTPS”, a showstopper, and the score is held to 45.
How we calibrated it
We first scored 40 real small-business websites (clinics, dentists, law and conveyancing firms) and found our early scoring was too harsh: none reached 80, and typical healthy sites looked mediocre. We rebuilt it so that the problems that matter drive the score, the small stuff doesn't, and the ranking still matches common sense: unpublished, hidden or insecure sites land at the bottom, polished ones at the top. We re-check this whenever the scoring changes.
Measured, estimated and verified
These three words mean different things in your report, and we keep them apart.
- Measured. The score at the top comes only from a scan we ran. It never changes unless you scan again.
- Estimated. When you tick “Mark fixed”, that is your own note. We may show what your score could be if everything you ticked really is fixed, always labelled as an estimate. It is never added to your measured score.
- Verified. After you re-scan, we compare the two scans. An issue the first scan found and the new scan doesn't is shown as “no longer detected”. That verifies the result on your site. It does not tell us what you changed.
What “tested fix” means
We only show a fix once we've verified it. Right now that's 35 fixes. We have drafted more (65 others) that you will never see until they're verified; for those you can ask us to review them for you. Each verified fix carries a record of how, visible on the fix itself. There are three kinds of evidence:
- Automated test. We build a page with the problem and confirm our scanner flags it. We apply the fix's own code snippet, not a hand-written stand-in, and confirm the flag clears and no new problems appear.
- Real websites. We take the real pages of live sites, apply the fix to a copy, and re-scan it to check the issue clears there too, not just in our own test pages.
- Official documentation. Where a fix involves a platform's settings (WordPress, Wix, Squarespace, Shopify), we check the menu steps against that platform's own help pages. Steps we have checked are marked ✓ on the fix; the rest are general guidance, because menus change between versions and plans.
What this does not prove: that a fix suits every site. A fix can still depend on your theme, plugins, host or plan. That's why every fix tells you to re-scan afterwards. If the issue doesn't clear, tell us. Evidence last gathered 30 September 2026.
What the score can't tell you
- It is automated. It finds real, measurable problems but can't judge whether your design is good or your copy persuasive.
- The speed test runs from Google's servers on a throttled phone, so sites hosted far from the US can read slower than most of their local visitors experience. Treat it as a comparison, not a stopwatch.
- A finding describes what we observed. “Why it can matter” describes possible consequences, not outcomes we have measured for your business. We don't claim an issue will cost you rankings, customers or money, and nothing here is legal advice.
- We read up to eight pages. A very large site is sampled, not fully audited.
- Some sites block automated visitors. When we can't read a site we say so instead of guessing.
Think a score is wrong? That's useful to us. Email [email protected] with the address and what looks off, and we'll check it.