Methodology
How we score your website
Every number on a SnapSiteScan report can be traced back to something measurable. This page explains exactly where the data comes from, how the Website Health score is built, and how the AI decides which single issue to put in front of you.
1. Where the data comes from
Each scan runs a Google Lighthouse audit through the PageSpeed Insights API against the URL you enter, on both mobile and desktop. We do not maintain our own rendering engine or crawler, and we do not adjust Google's raw category scores. Results are cached anonymously per URL for 12–24 hours so repeat checks are instant; a forced fresh scan bypasses that cache.
Lighthouse is lab data. It simulates a specific device and network profile, which means two runs minutes apart can differ by a few points. Treat single-point movements as noise and trends as signal.
2. How the Website Health score is built
Four Lighthouse categories are combined into one number so you do not have to weigh them yourself:
| Category | Weight | Why this weight |
|---|---|---|
| Performance | 40% | It is the category visitors physically feel. Slow pages lose people before anything else matters. |
| SEO | 25% | Technical SEO decides whether the page can be found at all. A fast page nobody sees is still invisible. |
| Accessibility | 20% | Accessibility failures block real users and increasingly overlap with usability and legal risk. |
| Best Practices | 15% | Security, HTTPS and modern web standards. Important, but rarely the reason a site underperforms. |
In plain arithmetic: health = performance × 0.40 + seo × 0.25 + accessibility × 0.20 + best practices × 0.15, rounded to the nearest whole number.
| Score | Label | What it means |
|---|---|---|
| 90–100 | Excellent | Nothing urgent. Rescan after significant changes. |
| 75–89 | Healthy | Solid foundation with room for a few targeted wins. |
| 50–74 | Needs care | Real friction for visitors. Worth fixing this month. |
| 0–49 | At risk | Issues that are very likely costing you traffic today. |
3. How the AI picks one thing to fix
A report with 30 issues is a report nobody acts on. After the scan completes we pass the Lighthouse opportunities, category scores and Core Web Vitals for your URL to a language model, with instructions to choose exactly one issue by weighing expected impact against implementation effort, and to describe it without jargon.
- • The model selects only from issues Lighthouse reported for your page — it cannot introduce new ones.
- • Impact and effort are shown as star ratings so you can disagree with the ranking.
- • "Why we picked this issue" states the reasoning explicitly rather than hiding it.
- • When the scan is inconclusive, the report says so and marks the confidence as low.
4. Predicted scores and percentiles
The predicted score ("87 → 94") is a deterministic estimate, not an AI guess. It is derived from the estimated savings Lighthouse attributes to the highlighted opportunity, converted into a performance gain, capped at a maximum of 12 points, and fed back through the same weighting formula. We cap it intentionally: an estimate that overpromises is worse than no estimate.
The percentile compares your health score with the health scores of all sites previously scanned with SnapSiteScan and held in our anonymous result cache. It describes our sample, not the whole web, and it only appears once that sample is large enough to say anything useful.
Frequently asked questions
How is the Website Health score calculated?
The Website Health score is a weighted average of the four Lighthouse categories returned for your URL: Performance 40%, SEO 25%, Accessibility 20% and Best Practices 15%. The result is rounded to a whole number between 0 and 100.
Why weight Performance higher than the other categories?
Performance is the only category a visitor experiences directly and immediately. SEO, accessibility and best practices matter, but a page that takes eight seconds to render loses users regardless of how well it is marked up.
Where does the underlying data come from?
Every scan is a Google Lighthouse audit run through the PageSpeed Insights API on both mobile and desktop. We do not invent metrics or run our own crawler. Lab data can vary between runs because it depends on network conditions and the state of your server at that moment.
How does the AI pick the one thing to fix first?
We send the Lighthouse opportunities, category scores and Core Web Vitals from your scan to a language model with instructions to select a single issue based on expected impact versus effort, and to explain it in plain English. The model chooses among issues Lighthouse actually reported for your URL; it does not invent new ones.
What does the predicted score mean?
The predicted score is a conservative arithmetic estimate of what your health score would look like if the highlighted issue were resolved, derived from the estimated savings Lighthouse reports for that opportunity. It is an estimate, capped deliberately low, and never a guarantee. Rescan after deploying the fix to see the real result.
What is the percentile based on?
It compares your health score against the health scores of every site previously scanned with SnapSiteScan and stored in our anonymous result cache. It is a comparison against our sample, not against the entire web, and it is only shown once the sample is large enough to be meaningful.
See it on your own site
Run a free scan and check the numbers against the method above.
Grade my websiteAlso useful: SnapSiteScan vs GTmetrix