Language index: which language the AI engines pick

How to read the number

The index is a ratio between two shares. A value above 1 means the engine uses the language more than your visitors do, a value below 1 that it uses it less.

Reported per engine, never as an average

In a measurement on our own site the values ranged from 1.11 to 4.21 for different engines on the same site over the same period. An average would have said everything was normal and hidden the fact that one engine fetched English pages four times more often than the search traffic warranted. The whole observation behind the metric is that the engines behave differently.

When no index is shown

The metric rests on the language of the pages being determinable, and that is not always the case. Rather than showing a confident number built on guesses, Paidin shows nothing at all when the basis is too thin.

• If the language cannot be determined for at least 60 percent of the fetches, only the distribution is shown, no index.
• Pages with an unknown language count in neither the numerator nor the denominator. They never fall back to the site's main language, because that would turn every unmeasurable page into a silent argument that you are under-indexed.
• Machine files such as robots.txt and sitemaps are not counted at all. On our own site 372 of 504 fetches from one engine were such files, and without the filter the metric would have been meaningless.
• Paths that return an error are counted, however. They are real fetches whose language could not be determined, and that should show up in the coverage.

Fetching and citation are kept apart

An engine fetching a page does not mean it uses that page in an answer. The two are therefore measured separately and never merged. The citation side also counts only your own domain, because the language on someone else's address cannot be determined from the address.