AI Visibility pillar / check ai-method-citations
Citations behind AI method claims
When an AI-agency page makes model, benchmark, or dataset claims, checks for links to the papers, datasets, or repositories behind them.
By Shimon Carroll, Founder, SEO for AI Agents · Last updated
What this check measures
Method references (benchmark, fine-tuning, retrieval-augmented generation, "our model", evaluation sets, datasets, LLMs). With two or more, we look for links to arxiv.org, paperswithcode.com, github.com, openreview.net, aclanthology.org, ACM, IEEE, Hugging Face, or doi.org.
Why it matters
Google’s helpful content guidance asks whether content provides clear sourcing and evidence of expertise. For technical buyers, a method claim with a linked paper or repository is verifiable; one without is marketing. The FTC also expects AI claims to be substantiated.
How we score it
A review prompt, not a legal verdict. Runs only on AI agency audits. When the cue is missing the finding is low severity at medium confidence, which lowers the ai visibility pillar by about 0.6 points. A pass is informational and never scores. Pages with fewer than two method references are not evaluated. In that case the finding is marked not measured: it does not score and does not count toward coverage. Each finding names who should review it, quotes what we saw on the page, and cites the primary source. Where the page shows a US address we name the inferred state; otherwise the wording is state-agnostic.
Confidence-flag rules
MEDIUM: the check reads the visible page text, links, and JSON-LD directly, so what it quotes is exact, but wording can carry the same meaning in phrasing we do not recognise, and one page does not show the whole site.
Common mistakes
- Naming a benchmark without linking the results.
- Linking a vendor blog instead of the paper.
- Describing a proprietary method with no evaluation at all.
How to fix it
Link the paper, public evaluation, dataset, or repository behind each method or benchmark claim, and date the results.
Primary sources
- Google Search Central, Creating helpful, reliable, people-first content
Google Search Central
- FTC, Crackdown on Deceptive AI Claims and Schemes (Operation AI Comply)
Federal Trade Commission
Changelog
- · New. This rule used to live only in the vertical configuration and never ran. It is now a registered review check with a cited primary source, framed as a prompt to review with a professional rather than a pass or fail legal verdict.