When a reader searches for an answer, the old bargain was easy to picture. A search engine offered a set of links. The reader chose one, and a publisher had a chance to earn attention, trust, a subscription, or a return visit. An AI summary changes the handoff. The reader may receive a useful orientation before opening anything, while the publisher may become an ingredient in an answer rather than the destination.
That is not a simple story of users winning or publishers losing. A summary can save time or point toward a source a reader would not have found. It can also compress context and reduce encounters with the page, its corrections, authors, and invitation to return. The reader question is two-sided: what is gained when the answer arrives on the results page, and what disappears when the click is unnecessary?
The evidence does not justify a forecast about every publisher. It supports a better unit than “traffic up” or “traffic down.” Call it the attention bargain: a platform offers speed and synthesis while a publisher supplies knowledge and may receive less direct relationship-building in return.
What the strongest public observation actually measured
Pew Research Center studied browsing behavior from 900 U.S. adults who agreed to share activity on a tracked device. The sample covered searches made March 1–31, 2025. Pew reconstructed those Google result pages April 7–17 and matched the next action to the links it found. It contained 68,879 unique searches, of which 12,593 produced an AI summary. Pew’s methodology and findings observe behavior rather than asking people what they remember doing.
In that sample, Pew reports a traditional result link was clicked on 8% of visits to pages with an AI summary, versus 15% without one. A summary link was followed on 1% of visits with a summary. Ending the session occurred on 26% of pages with a summary and 16% with traditional results only. About two-thirds of searches, with or without a summary, led people to continue searching or leave without clicking a result link.
These observations belong to a defined period, population, device setup, search engine, and reconstruction method. They are not September 2026 click rates, and do not establish that summaries caused a publisher’s revenue or traffic to decline. Searches that receive summaries may differ from others, and reconstruction can misclassify a changing page. A traditional result click and a summary-source click are separate measures; combining them overstates certainty.
The reader receives speed, but not necessarily the whole record
The gain is worth taking seriously. A reader may get orientation before deciding whether a topic deserves more time, compare unfamiliar terms, or avoid opening several pages on a small screen. These are plausible benefits of summarization, not outcomes measured in the Pew article.
The loss is not simply a missing pageview. A visit includes the surrounding explanation, sources, author, update date, correction, related tool, and the moment a reader decides to subscribe or return. A summary can cite a page while omitting qualifications that make its claim safe to apply. It can expose an idea without creating a durable relationship.
That is an inference, not a causal finding. Pew measured next actions, not trust, learning, subscription intent, or the financial value of a visit. Publishers should measure those possible losses, not announce them as facts.
Google’s position is opportunity plus a new control
Google’s current Search Central guidance describes AI Overviews and AI Mode as grounded in the Search index, using retrieval and prominent links to support an answer. It says these features may create opportunities to reach people who engage, spend time, or convert, and recommends valuable, people-first content. These are Google’s product explanations, not independent evidence that a site will receive more visits. The guide says no special markup or ideal page length is required.
Google’s June 3, 2026 announcement, updated August 31, adds a consequential choice. The company says a Search Console control lets owners decide whether their site appears in or grounds generative AI features. Sites that opt out will not receive traffic or impressions from those features, and the choice is not a ranking signal outside them. Google says the controls and insights rolled out worldwide August 31. The announcement is a company statement, not a neutral economic assessment.
The control makes the bargain visible. Inclusion may create discovery without a visit. Exclusion may protect a boundary but forgo the impressions and traffic Google describes. There is no universal answer because a nonprofit, subscription magazine, local service, and course creator value attention differently.
Keep an attention ledger, not one victory metric
The original contribution is a small ledger separating what a summary can deliver from what sustains a publisher.
| Stage | Question | What it can and cannot show |
|---|---|---|
| Presence | Was the page shown in a generative AI feature? | Exposure, not that a reader noticed or understood it |
| Citation | Did the feature link or name the page? | Attribution opportunity, not a visit or endorsement |
| Visit | Did a reader arrive through a link? | Referral, not trust, subscription, or revenue |
| Relationship | Did the reader return, subscribe, contact, or use a publisher-owned service? | A stronger sign of durable attention, still not proof of long-term value |
Google’s current Generative AI performance report can help with the first stage. It includes AI Overviews and AI Mode impressions, grouped by pages, countries, dates, and devices. New data can be preliminary, totals can differ because of aggregation, and Search Labs experiments are excluded. The documentation describes visibility, not a complete account of citations, clicks, or economics.
Pair the report with first-party analytics and subscription records where appropriate. Align date windows and record normal search, generative feature, newsletter, and direct visits separately. Do not turn an unavailable field into zero; label it unavailable.
A hypothetical decision at a small publisher
Imagine a small public-interest site publishes a sourced explainer on a complicated benefits rule. A Google summary cites it. Some readers may arrive with better orientation; others may stop at the summary. The publisher sees an AI-feature impression, perhaps a citation, and flat referrals. Without a causal experiment, it cannot say the summary produced either outcome.
The editor can still decide responsibly. Check for value that cannot be compressed safely: dated definitions, exceptions, sources, a correction policy, and a question route. Review the ledger over a matched period and mark observable stages. Then choose a boundary deliberately. Continue inclusion if discovery matters most; consider the control if direct relationship and context matter more. Record the policy and revisit it as the platform or goals change.
Protect the relationship that a summary cannot carry
The defensible response is not to make every page longer for a machine. Google says to make content for people, emphasize unique value, and avoid tactics designed only to manipulate generative responses. A page should be worth opening when a reader needs evidence, nuance, an update, or a human-made tool.
That may mean a source trail, date-sensitive change log, inspectable calculator, correction route, or newsletter inviting a return. None guarantees a click, but each makes the destination more than a container from which a summary lifts a sentence.
The attention bargain will change with interfaces and habits. Keep the record aligned: say what was measured, what the platform claims, and what remains inference. Keep the reader’s time and the publisher’s relationship in the same frame.
Sources and limitations
- Pew Research Center, “Google users are less likely to click on links when an AI summary appears in the results” — checked September 2, 2026. Supports the defined 900-person browsing sample, March 2025 search window, reconstructed results, click and session-ending observations, and methodology limits. Limitation: observational, Google-only, time-bounded evidence; it does not establish causation or current September 2026 rates.
- Google Search Central, “Google’s Guide to Optimizing for Generative AI Features on Google Search” — checked September 2, 2026. Supports Google’s description of retrieval, prominent links, people-first content guidance, and the statement that no special AI markup or ideal page length is required. Limitation: company guidance and product position, not independent traffic or revenue evidence.
- Google, “New opportunities, control and insights for website owners” — checked September 2, 2026. Supports the June 3, 2026 announcement, August 31, 2026 update, generative-feature control, opt-out consequence, and worldwide rollout statement. Limitation: company announcement; rollout and interface details should be rechecked before publication.
- Google Search Console Help, “Generative AI performance report (Search)” — checked September 2, 2026. Supports the report’s AI Overviews and AI Mode impressions, page/country/date/device dimensions, preliminary-data and aggregation cautions, and Search Labs exclusion. Limitation: the report does not by itself measure the full citation, referral, relationship, or economic ledger proposed here.