How the audit actually works.
Most audit tools are a quiz with a progress bar. You answer sixteen questions about yourself, and the number at the end is arithmetic on your own answers. Ours reads the site. Here's every check it runs and what each one is worth.
The three components
A hundred points, split by how much each part tends to cost a small business in practice.
The rule that shapes everything: a component we couldn't read scores nothing, not zero.
If we find no social links on the page we read, that isn't evidence you aren't on social. They might be in a footer we couldn't render, or behind a link-in-bio. So the component is marked unscored and left out of the total, rather than counted as a zero that drags your number down for something that may not even be true.
The same goes for analytics. A missing tag on the page we read might still be firing through a tag manager. We report what we found on the page, and leave the rest alone.
How much we saw governs how far we'll commit
Abstaining fixed one problem and created another. With the heaviest component unmeasured, a site whose two remaining checks happened to be clean could score 100 and an A. On the first real run of the scanner, one did exactly that off 60% of the picture.
Handing someone a perfect score off most of the picture destroys trust faster than a harsh one. It's the same error as scoring a zero, just in the flattering direction. So coverage now governs the grade:
- 90% or more of the picture — full confidence. The grade stands as scored.
- 60% to 90%— partial. An A is demoted to a B, because an A means “we looked at everything and it was excellent” and we can't say that.
- Under 60% — the letter grade is withheld entirely. You get the components we could judge and nothing more.
The score and the coverage are always shown as two separate things: a number out of 100 across what we could check, alongside how much of the full picture that represents.
Where the AI comes in, and where it doesn't
Every number above is arithmetic. None of it is generated, and none of it is a language model's opinion — the scores come from Lighthouse results and from what is or isn't in your HTML.
The model's job starts after that. It reads the findings and writes the part that takes judgement: what this means for a business like yours, which three things to fix first, and why the others can wait. It's explicitly instructed that an unscored component is not a bad one, and that it may not invent a finding that isn't in the data.
If the model is unavailable, you still get the full scored report without the narrative. The measurements never depend on it.
Common questions
How long does the audit take?
The technical read is near-instant. The Lighthouse run against Google can take up to a minute, and the written analysis a little longer, so a full report usually lands within a few minutes.
Do I need to install anything?
No. The audit reads your site the same way a visitor or a search engine does, over the public web. There is no tag, no plugin and no access to your analytics.
What if the audit cannot read my site?
Then it says so and scores nothing for that component. A site behind a login, a firewall or heavy bot protection will produce a report with large gaps, and we would rather show the gaps than guess at what is behind them.
Why did I not get a letter grade?
Because we could see less than 60% of the picture. A grade off partial data is worth less than no grade, so below that threshold the letter is withheld and only the components we could actually judge are shown.