The AI Scam Query Index. What AI assistants ask about scams, and who they cite.

When Microsoft's Copilot answers a question about scams, it searches the Bing index first and, if it cites one of our pages, Bing tells us what it searched for. This is every one of those searches over 90 days: 180 grounding queries, 5,021 citations. 53% of them are about which payment app is safer. On 106 of the 180 queries, one small independent site supplied a quarter or more of everything the assistant cited.

The first export bundles everything that already existed on 2026-07-27. Later months are genuinely new queries. No query that ever appeared has since dropped out, which is what a rolling 90-day window on a young site looks like.

Method, and what this index does not claim

Source. Bing Webmaster Tools, AI Performance report, for this site. A grounding query is a search that Microsoft's AI surfaces (Copilot and Bing's generative answers) sent to the Bing index while composing an answer that then cited one of our pages. Bing reports the query wording, the number of citations, and our share of all citations for that query. Intent and topic labels are Bing's; the cluster label is ours, assigned by a published keyword rule and shipped with the data so it can be redone.

Window. 15 weekly exports kept since 2026-07-27. Each export is a rolling 90-day window ending on its date, so totals can only grow while the site is young; compare rates and shares, never levels. The query table is the export of 2026-09-08; first-seen is the earliest export in which a query appeared. The dashboard's own Total Citations tile read 12.8K on that day, larger than the 5,021 in the table, because the table lists only the queries Bing chooses to expose.

What it is not. A census of what people ask AI about fraud. Bing exposes a query only when one of our pages was cited for it, so every query here is one we already answer, and the index says nothing about the questions we do not. It covers Microsoft's surfaces only; ChatGPT, Perplexity and Google's AI Mode publish no equivalent report. The share figure is Bing's and its denominator is Bing's. The GA4 comparison is a cookie-consent-gated subset and is reported as a sequence, not a cause.

Why publish it. Site owners see this data and almost nobody outside the SEO trade knows it exists. For a fraud education site the demand shape is the finding: people ask an assistant to choose between payment apps and dating apps, and the assistant leans on whoever has done the comparison honestly. That is a different game from ranking on Google, and it is the one we are playing.

Free to cite

CC BY 4.0. Journalists and researchers are welcome to use this with credit. Two CSVs, a JSON file and the written method are in the deposit. Corrections to press@tuteladigitalis.com; we publish them.

Archived on Zenodo under the DOI 10.5281/zenodo.22657317. That is a concept DOI, so it always resolves to the newest version. To cite this exact revision, version 1.0.0, use 10.5281/zenodo.22657318.

Suggested citation: Tutela Digitalis (2026). The Tutela Digitalis AI Scam Query Index (Version 1.0.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.22657317