How the Index is decided

How checks get chosen and weighted, how a score is computed, how long the evidence behind one is kept, and what happens when somebody thinks a finding is wrong.

We score other people's websites and publish the result. That only works if you can see how the scoring is decided, argue with it, and check our work — so this page says how, and the methodology itself is published in full rather than described.

How a check gets into the Index

A check earns its place by pointing at something a machine actually does. Every one of the 68 cites a primary source — an AI vendor's own crawler documentation, a specification, a standard — and that citation is on the page beside the check, not in a footnote we keep to ourselves. If the only argument for a check is that we think it's good practice, it does not go in.

We add checks when the thing they measure changes: a new crawler, a new well-known endpoint, a specification that stops being a draft. We remove them when what they measure stops mattering. Both show up in the Index changelog.

How the weights are decided

Each check carries a weight, and the weights add to 100. A check is worth more when failing it costs you more — being blocked outright at the door costs more than a missing timestamp, so retrievability outweighs instrumentation by a wide margin. Those judgments are ours, they are arguable, and they are all visible: every weight is in the methodology and in checks.yaml, which is published under the same open license as the rest of the Index.

We do not weight a check by how easy it is to sell a fix for.

How a score is computed

Every check is scored into a band from 0 to 4 against anchors written down in advance, so two people scoring the same evidence land in the same place. The band becomes a share of that check's weight, and the total is the share of the weight we could actually assess — a site where a third of the Index doesn't apply is scored against the two thirds that do, not against the whole thing.

Some checks are gates. Turning AI crawlers away at robots.txt caps the total however good everything else is, because everything else is unreachable.

A Quick scan reads one page and runs the 13 checks that can be answered from one page. It reports the points it banked on the Index's own 100-point scale rather than a percentage of what it looked at — a quick scan cannot award most of the Index, and a number that hid that would flatter you.

The evidence behind a finding

Every finding records what we actually saw — the URL, the header, the line of robots.txt, the missing property — and that evidence is shown under the finding in your report rather than summarized away. If we can't show you why a check failed, we treat that as our bug.

Evidence is kept with the scan for as long as the scan is kept; see the privacy policy for how long that is and how to have it deleted. A re-scan never overwrites an earlier one — the comparison between them is the point.

When you think a finding is wrong

Tell us: hello@ten7.com, with the report link. We will re-run the check against the evidence stored with that scan and reply with what we find.

Three outcomes, and we say which one it was. The scanner was wrong, and we fix it and re-scan you. The check was wrong — badly specified, or measuring something that no longer matters — and we change it in the Index, which changes it for everyone and lands in the changelog. Or the finding stands, and we show you the evidence it stands on.

Corrections to the Index are versioned. A change that would move existing scores gets a version bump, and every report records the Index version it was scored against, so an old report stays readable rather than silently meaning something new.

What we do not do

We do not adjust anybody's score by hand. We do not sell a better one, and being a TEN7 client does not change what your site scores. The scanner runs the same checks in the same order for a prospect, a client and for this site itself — we publish our own results, including the ones that are bad.

The Index is yours to check

The methodology and the 68 checks are published under CC BY-SA 4.0: you can implement the Index yourself, score your own site without us, and get the same answer. That is the strongest guarantee we can give that the scoring is not a black box — see licensing for what that covers.

Questions about any of this: hello@ten7.com.