We run a free check that scores any website out of 100. Most tools that do this keep the rubric to themselves, which makes the number impossible to argue with — and a number you cannot argue with is not a measurement, it is a sales prop.
So here is the whole thing. Nine scored checks, weights adding up to exactly 100, plus two that score nothing on purpose. What each one is, what it is worth, and why it is worth that. If you never run our audit, you can still work through this list yourself.
The two reports are the same engine, run twice
The preview — free, instant
- Runs every check except the speed test
- Stops within 6 seconds
- Marks speed as “Locked”, never as a failure
The full report — after email
- Runs everything, speed test included
- Arrives by email; the speed test alone takes 10–30 seconds
- Scores across the complete set of checks
The preview does not fail the speed test. It marks it "Locked" and leaves it out of the score. Our own AI-readiness measure shows as "Not measured", because it has nothing to weigh without the speed test. Then the preview tells you how much of the total it was able to judge, in exactly these words: "9 of 11 checks measured (82% of scored weight)." Measuring less must never look like measuring worse.
The nine scored checks, in weight order
| # | What we call it | Weight |
|---|---|---|
| 1 | Ways to get in touch or book | 20 |
| 2 | Speed on a phone | 18 |
| 3 | Can search engines index it | 12 |
| 4 | Structured data | 12 |
| 5 | Structure and accessibility | 12 |
| 6 | Which AI crawlers your site allows | 8 |
| 7 | Language signals | 8 |
| 8 | Sitemap health | 5 |
| 9 | Link preview when shared | 5 |
1. Ways to get in touch or book — 20
Whether someone who has decided to act can actually do it: an enquiry form, a booking or reservation link, a phone number they can tap instead of copying, an email, or the messaging apps your market really uses — WhatsApp, Viber, Messenger. One of those is enough to pass, as long as it is not email alone: an email address is a real route, but it is the slowest one a phone offers, so a homepage where email is the only way through is marked as needing attention rather than passing. If all we find is an address, opening hours and a map, the check says so: a visitor can work out where you are, but there is nothing on the page to act on. The check reads the homepage as it is served. The full report also opens the page your booking or contact link leads to and lists what it carried, without scoring it. We never submit a form, so nothing here claims the step after the first tap works.
This is the heaviest check because it sits closest to the decision. Everything else on this list is about being found; this one is about what happens in the second after someone has decided.
2. Speed on a phone — 18
Most people find you on a phone, often on mobile data, often standing outside. If your page takes too long, they leave before they have seen anything — and they do not come back and try again later. It is the thing most likely to be costing you customers today without you ever knowing.
3. Can search engines index it — 12
Whether your site is telling Google it is allowed to appear in search results at all.
This is the one check that can cap the whole score
4. Structured data — 12
This is a small, hidden summary of your business written into the page in a format machines read: your name, address, opening hours, what you do. Without it, search engines and AI tools are left to guess all of that from your sentences. With it, the page states it for them. We check that it declares what kind of business you are, not that every detail inside it is right. A site that declares itself as an organisation that cannot be a business, such as an NGO or a public body, is not marked down for lacking a business type.
5. Structure and accessibility — 12
Whether the page is built in a way both people and machines can follow: one clear main heading, headings in a sensible order, a stated language, labelled buttons and forms. This is what a screen reader needs, and it is also what a search engine uses to work out what the page is about. The same work serves both.
6. Which AI crawlers your site allows — 8
ChatGPT, Claude and Perplexity read websites with their own automated readers, and Google offers a robots.txt setting that controls how Gemini may use a site. Some sites accidentally block them — usually nobody meant to. Those readers do different jobs. Some collect text for model training, and blocking those does not remove a site from an assistant's answer. Others fetch the page at the moment someone asks about you, and a block there does. We look for ten of these names in your robots.txt, and score only the ones whose block would keep a page out of an answer.
7. Language signals — 8
Whether the site tells search engines which languages it exists in and how the versions relate. We check where the links lead rather than trusting their labels: the list of versions has to include this page itself, under the language this page declares. In the full report we also fetch a few of the other versions to confirm they exist, though we do not read what language they are in. We are careful here: being in one language is not a mistake. On a site with no second language the check is marked not applicable and leaves the score entirely — it cannot count against you. If your customers arrive in three languages and the site only speaks one, that is a gap worth naming, and the report says it that way, as an opportunity.
8. Sitemap health — 5
A sitemap is a list of your pages handed directly to search engines so they do not have to discover everything by following links. We report what the file contains, and in the full report we fetch a handful of the URLs it lists and tell you what came back — a listed page that returns 404 is something you can check for yourself in a browser. Not having one is not a failure. Google says a small, well-linked site may not need a sitemap, so we record its absence as worth considering, not as a fault.
9. Link preview when shared — 5
What appears when someone shares your link on WhatsApp, Facebook or Viber — a title, a sentence and an image, or just a bare address. In markets where sharing happens in group chats, a bare link shows only the raw address; a proper preview shows a title, a sentence and an image next to it.
And two more that score nothing on purpose
Forgera AI-readiness — weight 0. Our own measure of how readable your site is to machines. It looks at layout stability — whether the page shifts under an automated reader while it is working out what to click — and it reports the structural conditions such a reader depends on, plus an llms.txt entry point if there is one. It is reported, not scored: the whole check carries no weight, so nothing it finds moves your number in either direction. That is deliberate. It is an experimental measure of ours, not Google's, and a signal we invented should not be able to move a claim we make about your business. The structure it reports is scored one check up, under structure and accessibility; charging you for it twice under two names would turn one problem into two. Google states it does not use llms.txt and marks it not applicable in its own audit, so marking you down for a file with no demonstrated consumer would be inventing a problem.
What the site is built on — weight 0. We report what platform we can detect — WordPress, Wix, Shopify and others, or no identifiable platform — and which server. It carries no weight in the score and the report says so plainly. Being on any particular platform is not good or bad; it is context, so the person reading knows what they are working with.
Scoring someone down for their choice of tool would be exactly the kind of fake authority this audit exists not to be.
The rules the engine follows
These are worth knowing before you read any report — ours or anyone else's.
Four rules, each of which costs us a nicer-looking number
Never a fail A check that crashes is never a fail. If our own code breaks, that is our problem, not your site's. It reports as unknown, with the reason. A bug of ours must never show up as your defect.
Never a pass Nothing we could not measure is counted as a pass. No coverage means no score for that check — never zero, and never a silent tick. A check that simply does not apply to your site is marked separately again, so our own blind spots are never padded out with questions that had no answer to give.
Always stated The score states its own coverage — "9 of 11 checks measured (82% of scored weight)." You can always see how complete the picture is.
One scored page, one moment The score describes one page at one moment, and the report says so in its footer. The full report reads up to two more pages and scores none of them. It is a measurement — not a forecast, and not a promise about rankings.
There is a fifth, and it only shows up on a bad day: if we could not retrieve your page at all, you get no score rather than a zero. "We could not measure your site" and "your site scores zero" are opposite claims, and printing the second when we mean the first would be a lie in our own favour.
Three things about the number that most scoring tools do not tell you
The weights are shares of what was actually measured, not fixed percentages of 100. If a check comes back unmeasured, it leaves both halves of the calculation — it cannot drag you down and it cannot flatter you. So "speed is 18% of your score" is literally true only when everything ran. In the free preview, where the speed test does not run, the remaining checks are scored across the weight that did.
Nine scored, eleven run. Two of the eleven — what the site is built on, and our own AI-readiness measure — are reported and carry no weight, which is why the coverage line counts eleven checks while the weights add to a hundred across nine.
The weights are our judgement, not a research finding. Nobody has established that speed is worth 18 and a sitemap is worth 5. We chose those numbers, we signed them off internally on 30 July 2026, and we revised them on 20 September 2026 — moving weight onto the check that asks whether a customer can reach you, and taking our own experimental AI measure down to zero. We can defend the reasoning, but it is an opinion about what costs a small tourism business customers, held by people who build these sites for a living. Anyone who tells you their audit weights are objective is selling you something. Changing ours changes every score the engine produces, so it is a decision, not a tweak. Any report that has been issued keeps the number it was given, stamped with the version it was scored under; we never recompute an old report to match new weights.
How the check actually runs
Method, not results — nothing here is a claim about any particular website.
- You give us one public URL
No login, no installation, nothing added to your site.
- We read that page
Plus the public files beside it, such as robots.txt and the sitemap. The full report also spot-checks a few of the pages your sitemap lists, and reads up to two more of your pages without scoring them.
- Eleven checks run independently
A check that crashes reports unknown; it cannot fail your site for us.
- Nine are weighted. Two are reported at weight zero
What the site is built on, and our own AI-readiness measure, are context — not grades.
- The score states its own coverage
“9 of 11 checks measured (82% of scored weight).”
Why we publish the rubric at all
Because the alternative is asking you to trust a number from a company that wants to sell you a website. We would rather hand you the list and let you check our work — including the parts where the honest answer costs us the more impressive result.