AI Overviews vs Featured Snippets: What Actually Changed

A featured snippet is an extraction. A search engine copies a paragraph, a list or a table out of one indexed page, shows it at the top of the results, and links back to that page. Nothing is written and nothing is merged.

An AI Overview is a generation. A model runs several searches behind the one you typed, pulls passages from multiple pages, and writes a new answer with source links beside it. The sentence on screen exists nowhere on the web.

Everything else follows from that one difference: who is doing the writing. It changes the selection process, the number of winners, the placement rules, the measurement and the click math. The two also co-occur, so an AI Overview firing does not mean the snippet is gone.

Google's Search Console documentation defines it in one line: "A featured snippet displays information extracted from a single, specific web page. It includes a link to the source web page." Three words there do the work. Extracted, so the text is yours verbatim. Single, so there is exactly one winner. A link to the source, so it is a placement of your own listing rather than a summary standing in front of it.

You cannot opt in. Google's featured snippets documentation is blunt about how to mark a page up for one: "You can't. Google systems determine whether a page would make a good featured snippet for a user's search request, and if so, elevates it." No schema, no attribute, no submission form.

On reporting, the same help page says a featured snippet "has a link to the source property, and so it (typically) occupies position 1," and that standard impression rules apply. Hold that sentence. The AI Overview equivalent reads differently.

Four dates. Announced at Google I/O in May 2023 as the Search Generative Experience. Launched in the United States as AI Overviews on 14 May 2024. Expanded to more than 100 countries on 28 October 2024. AI Mode, the conversational surface, reached the US in May 2025.

The important machinery is retrieval, not writing. Google's AI features documentation says these features may use a "query fan-out" technique, described as "issuing multiple related searches across subtopics and data sources." Your query is decomposed into narrower machine-generated searches you never see, and the answer is assembled from what comes back. Where fan-out sits in each engine's pipeline is covered in how AI assistants pick sources.

Eligibility is short: to appear as a supporting link, "a page must be indexed and eligible to be shown in Google Search with a snippet." There is no way to mark up in. The documentation states you don't need "new machine readable files, AI text files, or markup to appear in these features."

Two reporting rules separate an AI Overview from a snippet completely. First: "An AI Overview occupies a single position in search results, and all links in the AI Overview are assigned that same position." Second, the one almost nobody quotes: "To be counted as an impression, the link must be scrolled or expanded into view."

Extraction rewards something liftable: a self-contained block under a heading that matches the question, answer before context, in a shape that survives being copied out. A short paragraph, an ordered list, a table. The engine is choosing a rectangle of your page to photocopy, and if your answer only makes sense after the two paragraphs above it, there is no rectangle to take.

Synthesis is a different problem, because the target is a claim rather than a query. An answer makes several assertions and each needs support, so your page competes for the slots it can support and no others. That is why a 700-word page can take a slot from a 4,000-word guide that mentions the same detail in passing. How a page gets cut into retrievable units is covered in AI answer engine retrieval, where the unit is a chunk and not a page.

The practical difference is how many winners exist. A featured snippet is one slot: you hold it or you don't. An AI Overview is a handful of support slots that engines spread across domains, so being second-best on a sub-question is often still worth a citation. Losing a snippet is total. Missing one citation usually is not. The two jobs are not opposed, though, because both reward a clear, self-contained answer under a heading that names what it answers. What differs is the payoff structure, not the craft.

This is the part of the comparison that gets skipped, and it changes how you read your own reports.

On 22 January 2020, Google deduplicated featured snippets. Before that, a page could hold the featured snippet and a regular listing on page one, giving eleven entries on a ten-result page. After it, the snippet counts as one of the ten and the source page's ordinary listing is removed. Search Engine Land, Search Engine Journal and Seer Interactive all documented the change at the time, and the rule has held since.

A featured snippet is therefore a trade, not a bonus. You swap a blue link carrying the title and description you wrote for a block of text the engine chose, placed first. Sometimes the engine picks the paragraph that answers the question so completely that nobody needs your page.

An AI Overview does not work that way. Nothing in Google's documentation removes your organic listing because you were cited in one, or because an AI Overview appeared at all. The block is added above the results. Your ten blue links are still ten.

Read that third row twice. Through 2025, plenty of sites watched average position slide on informational queries and filed it as a ranking loss. Wherever a snippet was replaced by an AI Overview, part of that slide is a deduplicated listing being handed back: position moves from roughly 1 to whatever the page ranks organically, often between 2 and 5, with nothing changing about the page's actual standing. Diagnose that as a demotion and start rewriting content, and you're treating a placement change as a quality problem.

Ahrefs published the cleanest available measurement of the trade, across 1,000,000 randomly selected US desktop results pages between January and June 2025.

The correlation between the two series was -0.9053, which Ahrefs describes as very strong, with a visible switch-over point in March. Their earlier work noted the two features commonly appearing alongside one another, so this is displacement in aggregate rather than a clean one-for-one swap.

Four caveats belong beside those numbers: US desktop only, one tool's SERP sample, a six-month window that ended over a year ago, and a market-level correlation that is not proof of substitution on any individual query. Treat the direction as durable and the levels as stale.

What holds up is the base rate. The snippet opportunity has shrunk sharply on the query classes where AI Overviews fire, which skew informational. Navigational and transactional queries are far less affected, which is why brand terms make such a useful control set in the audit below.

The strongest public evidence on AI summaries comes from the Pew Research Center, published 22 July 2025. Pew analyzed browsing data from 900 US adults who agreed to share their activity, covering 68,879 unique Google search queries in March 2025, of which 12,593 produced an AI summary.

Now the honest reading. This is a behavioral panel, not a controlled experiment, and the two groups are not comparable populations of queries. AI summaries fire more often on open informational questions, which had lower click rates than navigational queries long before generative search existed, so some of the gap is query mix. Google publicly disputed the study. The direction still matches what publishers reported independently, but quoting "AI Overviews halve your clicks" as a clean causal fact overstates the evidence.

Featured snippet click effects are harder to pin down, because nearly all the well-known snippet CTR studies ran before January 2020, measuring a world where the snippet holder also kept a blue link. That arithmetic no longer exists, and few circulating figures price in the lost listing.

The mechanism both features share is simpler than either study. When the answer arrives on the results page, the click doesn't happen. That is the dynamic covered in zero click searches and where your impressions go, and it predates generative search. What changed is scale and authorship.

Here is the trap inside every "featured snippet CTR vs AI Overview CTR" comparison you will read.

A featured snippet follows standard impression rules. It shows, it counts. An AI Overview link counts as an impression only when it "must be scrolled or expanded into view," per Google's own documentation. Links inside a collapsed AI Overview the user never opened generate no impression at all.

That biases the comparison in the opposite direction most people assume. AI Overview impressions are counted only over the subset who scrolled or expanded far enough to reveal the link, a more engaged population by construction. So any CTR you compute for AI Overview citations sits on a filtered denominator. Compare it to a snippet CTR, computed over everyone, and you are comparing two denominators and calling the difference a performance gap.

On 3 June 2026, Google added generative AI performance reports to Search Console, covering AI Overviews and AI Mode. Read the fine print before building a dashboard on it:

• Impressions only. No queries, no clicks, no click-through rate, no average position.
• Broken down by page, country, device and date.
• A breakout, not new data. Google confirmed AI impressions were already inside your existing totals, so your aggregate numbers do not change.
• Limited rollout. It reached a subset of site owners first, so an account without the report is not an account without AI impressions.

So you can now see where AI features surfaced your links, and still not what they cost you in clicks. There is no featured snippet filter under Search Appearance either, so neither surface can be isolated on the click side. For anything involving clicks, the results page itself is still the instrument.

Both look identical in a dashboard: impressions up, clicks down, position worse. They call for opposite responses.

Google's AI features documentation lists nosnippet, data-nosnippet, max-snippet and noindex as the controls that limit what is shown from your pages, and states these are the same controls used for Search generally. The featured snippets documentation names the first three for snippets.

The honest summary: no control suppresses AI Overviews while keeping featured snippets, none does the reverse, and none does either while leaving your ordinary description untouched. Google's preview controls treat all of it as one system. If a tool or an agency offers a clean AI Overview opt-out that preserves normal search presence, ask which documented directive delivers that.