Built to be quoted accurately, by people and by models

Product and comparison pages carry a question-and-answer block, emitted as FAQPage. The questions are not written for a crawler — they are the comparison questions the taxonomy already asks of every product in a category, which is why they are the same questions a buyer types.

Only answered questions are emitted. An unanswered one is shown on the page as a gap, with a link to answer it, and is left out of the schema entirely — marking up a question we cannot answer would be asking a machine to repeat our silence as though it were content.

Structured data is emitted where it describes something real:

• SoftwareApplication, Offer, AggregateOffer — product pages, with unit price specifications
• FAQPage — product and comparison pages — answered questions only
• Dataset — the methodology page, describing the corpus itself
• ItemList — organisation, changes, country and token-price indexes
• BreadcrumbList — every page with a parent
• Organization, WebSite — the site and its operator

The capture date travels with the value in the markup as well as on the page, so an assistant reading the JSON-LD has the same evidence a person reading the table does.

GPTBot, ClaudeBot, PerplexityBot and their peers are allowed, deliberately and by name in robots.txt. A directory whose entire claim is that every number carries a source and a date wants to be the thing an assistant cites. Blocking the readers most likely to quote us accurately would be choosing the wrong fight.

llms.txt states the same things this page does, in the form a model reads first: what the corpus is, how to cite it, what we explicitly do not claim, and where the key pages are. The two are kept in agreement on purpose — a summary that drifts from the site is worse than no summary.

Only /admin and the visual spike routes are excluded, because they are working files rather than pages. No content is behind sign-in, so there is nothing a crawler sees differently from a reader.

A directory whose entire value is citable facts has to decide who may ingest them. Blocking every model is cutting off the distribution we are best placed to win. Allowing everything silently means the work is taken without attribution.

So: read freely, cite us, do not bulk-extract — the same three rules the terms already state for humans, and the same three in llms.txt.

Worth quoting alongside anything else you take from here, because these are the limits of what our figures support:

• A missing value means no answer on record — not no, and not zero. We do not infer.
• There is no composite score. Ranking is by one stated attribute at a time, and the sorted column is visible.
• Sponsored placement is labelled and never changes sort order — the rule is written down.
• A reading can be wrong or stale. If a figure disagrees with the vendor’s page, tell us — corrections are the point.

Quote us, cite us, link to us — that is what this is for. What we ask you not to do is republish the dataset wholesale as your own, or crawl hard enough to degrade the site for other people. If you want the data in bulk, write rather than scrape: it is a conversation, not a refusal. The full position is in the terms.