Google’s AI Overviews moved out of Search Labs and into general availability in the United States on May 14, 2024. The category we now call generative engine optimization — GEO — effectively turned one on that anniversary. After twelve months of public data and a year of practitioner reporting, it is now possible to write a working read on what AI Overviews actually changed about discovery, and on what survived more or less intact.
This is not a vendor piece and it is not a “GEO is dead” piece. It is a piece about what the public numbers show, what the working SEO leads at mid-market companies are reporting, and what the structural shifts mean for a marketing program in the second half of 2026.
What the public data says
The two best public datasets on AI Overviews behavior in the first year come from BrightEdge and from Semrush. They do not agree on every number, but they agree on the shape.
BrightEdge’s tracking, published across several quarterly reports through 2025 and into early 2026, has consistently shown AI Overviews appearing on roughly 13 to 20 percent of US Google queries by the end of year one, depending on the vertical. Health, finance, and B2B technology queries trigger Overviews substantially more often than entertainment, navigational, and transactional queries. The share has grown steadily, not by leaps; the introduction was rolled out cautiously after the high-profile early hallucination incidents in May 2024, and the cautious posture has held.
Semrush’s “AI Overview impact” research, updated through Q1 2026, reports a click-through-rate compression on positions one through three of the SERP when an AI Overview is present. The compression is in the 15 to 35 percent range, depending on whether the page is cited inside the Overview itself. Pages cited inside the Overview see less compression; pages displaced by the Overview see more. The data on this point is consistent with what the larger publishers have been reporting to industry trade press through 2025: Overviews are a click tax on the SERP for sites the Overview does not cite, and a click bonus for the sites it does.
Google’s own public disclosures on AI Overviews have emphasized the time-on-task improvements and the satisfaction metrics, neither of which the publisher community can verify. The publisher-side metrics that have been verified, by Chartbeat and SimilarWeb among others, are clearer about the cost: Google referral traffic to the news and reference verticals fell by a measurable single-digit-to-low-double-digit percentage through 2025, with the drop concentrated on the queries Overviews answer in-surface.
What changed: the structural shifts that held up
Four structural shifts in discovery have been corroborated across enough independent reporting to be treated as settled by mid-2026.
The high-intent informational query is now resolved in-surface for a meaningful share of the volume. This is the most consequential change. A buyer researching a product category, an executive researching a vertical, or a marketer researching a workflow now resolves a meaningful share of the high-intent informational journey inside Google’s AI Overview, ChatGPT’s web mode, Perplexity, or one of the integrated assistants. The page they would have clicked is the source the model cited; the click is the supplementary action, not the primary one. SimilarWeb’s traffic data on Wikipedia is the cleanest single proxy here: Wikipedia’s referral traffic from Google fell measurably through 2025, while Wikipedia citations inside Overviews and inside ChatGPT responses rose. The reading is straightforward. Users still consult Wikipedia; they consult it through the assistant rather than through the click.
Citation density inside answer surfaces has become a real discovery metric. A year ago, the GEO discourse was theoretical. By mid-2026, mid-market marketing teams have started tracking citation rate inside AI Overviews and inside the answer engines as a routine KPI alongside organic position. The measurement is imperfect — the third-party tools that scrape Overview citations are catching maybe 70 percent of what they are looking for — but the metric exists, has a name in the practitioner community, and has a budget line. Twelve months ago, none of those things were true.
Entity-anchored properties moved up the priority stack. The pages whose primary job is to define an entity — the About page, the Wikipedia entry, the Crunchbase profile, the LinkedIn company page, the structured-data scaffolding on the site — have measurably more leverage on the AI Overview surface than they had on the blue-link SERP. The reason is structural: the models grounding the Overview consult entity-defining sources more readily than the open-web ranking algorithm did. The teams that have rebuilt their entity layer in the past year are reporting a meaningfully better citation rate than the teams that have not. The Reuters Institute’s 2026 Digital News Report describes a parallel pattern on the publisher side.
The dark-funnel problem got measurably larger. A meaningful share of the buyer journey now happens inside conversational AI tools whose referral data does not pass through to the marketing team’s analytics stack. Most enterprise marketing teams reported, in the 2026 ANA “State of Marketing Measurement” survey, that the share of pipeline they cannot attribute has risen from the high-twenties percentage in 2024 to the high-thirties percentage in 2025. The Overviews change is part of that story, and ChatGPT, Perplexity, and Claude are the other part.
What didn’t change
Three things did not change as much as the discourse predicted.
Blue-link traffic did not collapse. It compressed. The doom predictions in the spring of 2024 — that AI Overviews would gut publisher traffic, that organic search was effectively dead, that SEO as a discipline was over — have not been borne out. Total Google referral traffic to most verticals is down modestly, not down catastrophically. The publishers and brands that did the entity-and-citation work are slightly up on their share of the smaller pie. The category as a whole is smaller. The discipline is not over.
Transactional and navigational queries are mostly unaffected. Users still Google “amazon,” still Google “best buy near me,” and still Google “[brand name] login.” These queries do not trigger Overviews, do not compete with answer surfaces, and have not seen meaningful traffic shifts. A non-trivial share of every marketing program’s SEO traffic falls in these query buckets, and the work to maintain that traffic is largely the same work it was in 2022. The doom narrative obscured this.
Long-tail informational queries still produce real click-throughs. The Overview is a fixed-cost feature for Google. It does not appear on every query. The long tail of niche informational queries where the answer is not yet in any model’s training distribution — emerging topics, recent events, hyper-specific procedural questions — still produces the kind of click-through pattern that long-tail SEO has always relied on. The volume per query is small. The aggregate is meaningful for the teams that built for it.
What the working SEO leads are doing
The mid-market SEO leads we have talked to in the past quarter are doing a consistent set of things. They have added an “AI Overview citation rate” metric to their dashboards. They have rebuilt their entity layer once, deliberately, with schema markup, a cleaned-up Wikidata entry, and a consistent canonical name across third-party properties. They have shifted budget from raw publishing volume toward authoritative depth on a smaller set of entity-anchored topics. They have started measuring branded-search lift and direct-traffic lift more carefully, because those metrics are now the dark-funnel signal they have access to. They have stopped trying to win the AI Overview wars on every query and started focusing on the queries where their entity has a real shot at being cited.
What the working SEO leads are not doing: panicking, throwing out the SEO playbook, or spending money on dedicated “GEO platforms” that have not yet earned their budget line. The discipline of organic search in 2026 is recognizably the discipline of 2022 with two new metrics, a re-prioritized property stack, and a more honest relationship with the dark funnel.
Where the category goes from here
The structural question that will define year two is whether the answer-engine surface stabilizes or whether it keeps fragmenting. If ChatGPT, Gemini, Claude, Perplexity, and the integrated assistants converge on similar citation behavior, the GEO discipline will look more like SEO than it currently does — one playbook, multiple surfaces. If they keep diverging, the discipline will look more like channel marketing — a different working playbook per surface, with the discovery teams managing a portfolio. The 2026 evidence is mixed. Our reading, on the working data, is that the surfaces are converging slowly, not diverging. The teams designing for portability across answer engines are making the right bet.
The other open question is monetization. Google has been cautious about advertising inside AI Overviews; the assistants outside Google’s ecosystem have started experimenting with sponsored citations. If the answer-engine surface starts carrying paid placement, the GEO category will look very different in eighteen months. The SEO community has been here before with the introduction of paid search; the right posture is to watch the development, not to bet against it.
What we can say, on the one-year anniversary, is that AI Overviews changed discovery in a structural way that was real but smaller than the doom narrative predicted. The teams that did the work in the past year are better off than the teams that did not. The teams that did nothing are still operating, with a smaller piece of a smaller pie. The discipline is, on the whole, doing what disciplines do: it is absorbing the shock, naming the changes, and getting on with the work.