The results page stopped being a ranked list and became an assembled answer. AI Overviews sit above the links on close to half of Google searches, AI Mode breaks a single question into dozens of sub-queries and synthesizes them, and on June 3, 2026 Google gave Search Console a dedicated report for how often you appear inside those AI surfaces — measured as impressions, not clicks. Together those two changes redraw the target: you are no longer optimizing to rank a page, you are optimizing to be selected by a curator, and the metric that tells you whether it worked is presence rather than position. This guide explains what AI-curated visibility is, the exact SERP and Search Console changes that created it, how the curator chooses what to include, why query fan-out makes coverage breadth beat keyword targeting, the trap in a presence metric that reports impressions without clicks, and the production reality of competing for it.
For twenty years the job of content strategy had one shape: produce a page, get it to rank, earn the click. The ranking was a list — ten links ordered by relevance — and your position in that list decided your traffic. In 2026 the list is no longer the main event. On close to half of Google searches an AI Overview now sits above the blue links, reading across the ranking pages and assembling a synthesized answer that cites a handful of sources. AI Mode goes further, replacing the list entirely with a conversational answer built from many sources at once. The results page stopped being a ranked index of pages and became an assembled answer — and something has to assemble it. That something is a curator.
This is the shift that reorganizes everything downstream, and it is worth naming precisely because the response depends on it. A ranker orders whole pages; a curator reads across pages and chooses which passages to include and cite in one answer it writes itself. Optimizing for a ranker means being the best page for a query. Optimizing for a curator means being one of the sources it selects into the answer — a different target, measured a different way. The rest of this guide is about that target: what changed on the results page and in Search Console to create it, how the curator decides what to include, and what a content strategy built for selection rather than ranking actually looks like. For the specific click math this creates, AI Overviews are reducing organic clicks is the companion; this piece is about the strategic reframe underneath it.
Two developments in the last two years turned an incremental SEO annoyance into a genuine strategy change. One is on the results page; one is in the measurement layer. They compound, and understanding both is what separates a real response from an "optimize for AI Overviews" checklist.
Google launched AI Overviews to US users at I/O in May 2024, and by early 2026 they were appearing on close to half of all searches — Similarweb's data put it around 43%, other measurements higher, and dramatically higher for informational-heavy verticals: reported above 80% for health and education, far lower for entertainment. Because the answer box is large, it occupies roughly 40 to 50% of the visible screen on desktop and even more on mobile, pushing the organic list below the fold. So even when your page ranks, the first thing the searcher sees is a curated answer, not your link.
AI Mode is the fuller version of the same idea. It began as a Search Labs experiment in March 2025, rolled out to all US users through mid-2025, and expanded internationally after. What makes it structurally different from a ranked list is its retrieval method: query fan-out. Rather than answering the phrase you typed, AI Mode uses a language model to decompose your question into many related sub-queries — Google has described this as fanning a single query into dozens, sometimes more — dispatches them in parallel across its indexes, and has Gemini synthesize the returns into one answer. Google says its AI Overviews now reach around 2.5 billion users a month. The consequence for strategy is direct: the unit the system retrieves against is no longer your keyword, it is the set of facets your topic breaks into.
The second change landed on June 3, 2026, when Google gave Search Console a dedicated generative-AI performance report. It sits as its own tab in the Performance section and reports, for the first time as a separate view, how often your URLs appear inside AI Overviews, AI Mode, and generative-AI features in Discover. Crucially, it shows impressions only — no clicks, no click-through rate, no position, no query data yet — broken down by page, country, device, and date. Google was careful to say this is not new traffic: those impressions were always folded into your overall totals, and this report simply breaks them out. It rolled out in beta, starting with UK site owners, before wider expansion.
Read what that metric is actually measuring, because it is the whole point. It counts presence: were you included in the assembled answer, yes or no. It does not report where you sat, because inside an AI Overview there is no meaningful position — Google treats the entire Overview as a single position and every cited link shares it, and an impression is counted when your link is present and scrolled or expanded into view. It does not report a click, because in a curated answer the click is optional and often absent. For years the tell that AI Overviews were affecting you was an inference — impressions flat while clicks fell. Now presence in AI surfaces is a first-class number you can watch. The KPI Google chose to hand creators is inclusion, which is the clearest possible signal of what the game became: get selected into the answer. Whether that presence converts is a separate problem, and one the metric pointedly does not solve — a gap the broken content metrics guide covers in depth.
Put the two changes together and the working definition falls out. AI-curated visibility is how often your content is selected into an AI-assembled answer, rather than ranked in a list of links. It has three properties that make it a different object from a ranking. It is binary at the surface level — you are in the answer or you are not, with no long tail of "page two" to nurse. It is passage-level, not page-level — the curator lifts a specific chunk, so a strong extractable answer inside a mediocre page can be chosen while a great page with no clean passage is skipped. And it is measured as presence — impressions in the new report, citations in the answer — not as position or, directly, as clicks.
That reframing changes what "doing well" means. Under ranking, success was a number that went up: higher position, more clicks. Under curation, success is being present across the answers your audience triggers, on-brand, with a passage the model was willing to quote — and then doing something with that presence, because presence alone does not pay. The strategy splits into two jobs that used to be one: earn selection into the curated answer, and convert the resulting visibility into demand you can actually see. The next sections take each in turn. For the broader case that AI search visibility is a manageable growth channel rather than a mystery, AI search visibility makes the argument; here the focus is the mechanics the new metrics expose.
You cannot see the model's weights, but its behavior is consistent enough to optimize against. A synthesizing curator that has to answer in a few sentences and cite its sources favors content it can lift cleanly and trust quickly. In practice that means four things, and they are craft choices more than technical ones. Give each question a direct, self-contained answer placed near the top of its section, in the plainest form the answer can take — the curator is looking for a chunk it can quote without your surrounding setup. Structure the passage so it survives extraction: a clear heading that matches the question, one idea per paragraph, definitions and steps and comparisons formatted so a machine can grab a complete unit. Corroborate the claim — a specific, sourced, checkable statement reads as safer to cite than an unbacked assertion, and cautious models prefer safe. And write in a way that reads as expertise rather than filler, because the same qualities that made content trustworthy to a human reader are what a model trained on human judgment leans toward.
None of that is new advice in spirit — it is what good informational writing always was. What changed is the payoff structure. Under ranking, a padded 2,000-word page could still win on aggregate signals; under curation, the padding is dead weight the model reads past to find the one clean sentence it wanted. The content formats that actually get cited guide catalogs which structures earn selection most reliably. The through-line is that the extractable answer is the asset now, and everything wrapped around it is either scaffolding that helps the model find it or noise that buries it.
Selection craft gets you into a given answer. Query fan-out changes which answers you are even eligible for, and it is the part most keyword-era strategy misreads. Because AI Mode explodes one question into many sub-queries and answers the composite, the system is not matching your page to a phrase — it is checking whether the corpus can answer each facet it generated. A page perfectly tuned to "best CRM for real estate" is competing in a world where the model quietly asked itself a dozen adjacent things: pricing, integrations, small-team fit, migration effort, what actual users complain about. If your content answers the headline query and none of the fan-out, you are a thin source for the assembled answer even if you would have ranked first for the original phrase.
So the coverage unit shifts from keyword to topic-and-its-facets. The strategic move is to map the sub-questions a topic fans out into and give each a clean, self-contained answer — sometimes within one thorough page, sometimes across a cluster of pages and formats that collectively cover the space. This is why breadth of well-structured coverage now beats a single hero page, and why the old instinct to consolidate everything into one "ultimate guide" can backfire: the curator wants a specific answer to a specific facet, not a monument to a keyword. It is also the mechanism behind the "no clear owner" queries that no single page has claimed — fan-out surfaces facets the existing web answers poorly, and the source that answers them cleanly gets selected by default.
Here is the catch that the new Search Console report makes unavoidable, and it is the reason AI-curated visibility is only half a strategy. The metric reports impressions without clicks by design — and that is not an omission, it is an accurate description of the surface. You can be present in the assembled answer, watch your impression count in the generative-AI report climb, and receive no visit at all, because the searcher got what they needed from the curation and never left. Presence in an AI-curated answer is genuinely valuable — your brand and your claim reached the user, and being the cited source builds authority and later branded demand — but it is not a click, and treating a rising impression line as if it were traffic is the exact misread the metric invites.
This is where AI-curated visibility connects to the larger truth about the era: winning selection into the answer does not, by itself, produce the outcome the old click did. Google is increasingly resolving the query in place — with roughly two-thirds of US searches now ending without a click to the open web — so the click you used to convert into a visit, a signup, or a sale is often simply not on offer. The publisher traffic collapse guide traces where that traffic went. The strategic implication for AI-curated visibility is specific: earning presence is necessary but not sufficient. You also have to convert that presence into demand on surfaces where a click is not the required step — which is a distribution problem, not a search problem.
Assemble the pieces and the strategy is clear but heavy. To compete for AI-curated visibility you need extractable, corroborated answers to the many facets a topic fans out into — that is more, better-structured content than a single ranking page ever demanded. To convert the presence you earn into demand you can see, you need the same answers living on the feeds and in the inbox where discovery does not depend on a click Google is keeping — so the answer that appears in an AI Overview also appears as a short video, a carousel, a post, a newsletter, in front of people who never ran the search. And to make either work you need consistency, because a curator assembling an answer, and an audience forming an impression across surfaces, both reward a brand that shows up the same way everywhere, not one that drifts format to format.
That is a volume-and-consistency problem before it is a search problem, and it is where most teams stall. The plan — cover the facets, be present in the answers, mirror those answers onto every surface where a click is optional, and hold the brand steady across all of it — is correct and roughly no one can execute it by hand at the cadence it requires. Producing one authoritative page was hard enough; producing facet-level answers plus their feed-native and inbox versions, on-brand, on a rhythm, is several times the work. The winning strategy is real; the production ceiling is what makes it theoretical for a person or a small team. The same wall shows up in SEO in the age of AI Overviews — the durable move is only actionable if you can generate at the volume it needs without the quality falling apart.
Kompozy is built for exactly the two-job structure this shift creates — earn selection, then convert presence — and its relevance here is the production ceiling, not a promise to game the curator. It is an AI content generation-and-publishing engine — 18 output formats across eight social platforms plus blog and email — so from one dense source it can produce both halves of the work at once. Point it at the substance and it generates the extractable, facet-level answers a curator selects from as Blog Articles and Text Posts, and simultaneously renders those same answers as the feed-native and inbox formats that convert presence into demand off the results page: Persona Shorts and longer persona video for the demonstration, Carousels and Infographics for the walkthrough, Persona Tweets and Photo Posts for the quick take, an Email Newsletter for the owned audience no AI Overview sits in front of. The topic that used to be one page you hoped would rank becomes coverage across the facets a query fans out into and across the surfaces where a click is not required.
The consistency the strategy depends on is enforced by the engine rather than left to willpower, which matters because both the curator and your audience reward a brand that reads the same everywhere. Every generation descends from a single Persona Brief that governs voice and filters banned words, a face-locked persona pool keeps your presenter identical across video and images, and HyperFrames renders brand-exact styling — so the answer that appears in an AI-curated result and the versions that appear on ten feeds are recognizably one voice, not one asset reshaped nine ways. Then Autopilot schedules and publishes that spread across the eight primary social platforms plus blog and email from one queue, behind a per-post review gate so a human signs off before anything ships. That review step is where the corroboration and the specific, first-hand detail get added — the qualities that earn selection in the first place — rather than left to a model to invent.
The honest scope matters, because this page argues against checklists and Kompozy is not one. No tool can force an AI Overview to cite you or make Google return a click it has decided to keep — the selection is the curator's, and anyone selling guaranteed inclusion is selling the thing that does not exist. What Kompozy removes is the volume ceiling that makes the only durable response impossible for a small team: producing facet-level, extractable answers and their multi-surface versions, on-brand, at a cadence, so you are present in the assembled answers your audience triggers and converting that presence into demand on the surfaces where a click still happens. If your entire strategy rides on a handful of pages ranking for a few keywords, classic SEO discipline is the cheaper call. Kompozy earns its place when the results page has become a curated answer and the way to stay visible is to be selected into it everywhere at once — at the volume that requires. The list stopped deciding your reach. The answer does now, and the content that wins is the content present across every answer, and every surface, that answer never covered.
AI-curated visibility is how often your content is selected into an AI-assembled answer — Google's AI Overviews or AI Mode — rather than ranked in the classic list of blue links. The difference is who decides what the user sees: a ranking algorithm orders whole pages by relevance, while a curator reads across many pages and assembles a single answer from the passages it chooses to include and cite. Being visible in that world means being one of the sources the curator selected, which is a different target from being the page that ranks first.
On June 3, 2026 Google launched a dedicated generative-AI performance report in Search Console. It sits as its own tab in the Performance section and shows impressions only — no clicks, CTR, position, or query data yet — broken down by page, country, device, and date, across AI Overviews, AI Mode, and generative-AI features in Discover. Google was explicit that this is a breakout of data that was already counted in your overall totals, not new traffic. Practically, it is the first time creators can see their presence inside AI-curated surfaces as its own number.
Google treats an AI Overview as a single position, and every link cited inside it shares that position. An impression is counted when your link is present in the AI Overview and scrolled or expanded into view. If the same URL appears both in the AI Overview and in the classic blue links for the same query, Search Console counts it once, not twice. That means the impression number tells you that you were included in the assembled answer — presence — but not where you sat or whether anyone clicked.
Query fan-out is the technique behind Google's AI Mode: instead of answering the query you typed, the model decomposes it into many related sub-queries — sometimes dozens or hundreds — runs them across different sources in parallel, and synthesizes the results into one answer. For content strategy it flips the target. You are no longer trying to rank for a single keyword; you are trying to have a clean, extractable answer to each of the facets the model fans out into. Breadth of well-structured coverage on a topic beats a single page tuned to one phrase.
Write for selection, not ranking. Give each question a self-contained, quotable answer near the top of a clear passage; structure content so a machine can lift a clean chunk without your surrounding context; cover the facets a query fans out into rather than one keyword; and corroborate claims so a cautious model trusts the source. Then, because the new metric reports presence without clicks, convert that visibility into demand off the results page — put the same answers on the feeds and in the inbox where a click is not required.
AI-curated visibility is how often your content is selected into an AI-assembled answer — Google's AI Overviews and AI Mode — instead of ranked in a list of links. Two 2026 changes force the shift: AI Mode's query fan-out splits one search into many sub-queries a model synthesizes, and Google's June 3, 2026 Search Console report now measures your presence in those surfaces as impressions, not clicks. You optimize by writing self-contained, corroborated passages a curator can extract, then converting that presence into demand off the SERP.
Get started → · ← All guides · Compare Kompozy vs other tools