What counts as a SERP feature
A SERP feature is anything on a search engine results page that is not a plain organic listing — the classic title, URL and snippet — and not a plain text advertisement. Everything else is a feature: the boxed answer at the top, the accordion of related questions, the strip of images, the panel of facts about a person or company down the right-hand side, the map with three local businesses on it, the AI-written summary above all of it.
The term is jargon, but the thing it names is the single biggest change in what a results page is. A page of ten links routes you somewhere. A page dominated by features answers you where you stand. Understanding which is which — and where each block of text was sourced from — is the difference between reading a results page and being read by one.
The features described here are documented, dated where possible, and named as their operators name them. They also change constantly, vary by country and device, and appear and disappear without announcement. Anything below should be read as a description of August 2026.
How often features actually appear: measured data
Most writing about SERP features asserts they are everywhere without measuring anything. This site has its own measurement, and it is worth stating precisely including its limits.
In an August 2026 SEMrush analysis of the 120 topical queries this domain ranks for, the feature counts were:
- Video results: 117 of 120 queries — 98%.
- AI Overviews: 104 of 120 — 87%.
- People Also Ask: 102 of 120 — 85%.
- Sitelinks: 92 of 120 — 77%.
Those numbers are high, and the honest caveats matter as much as the figures. This is one keyword set, for one domain, in one country, on one subject. The subject is search engines, which means the queries are overwhelmingly informational — "how does X work", "what is Y", "which engine does Z" — and informational queries are exactly the class where answer features cluster. A comparable measurement over transactional or navigational queries would look nothing like this. So the figures are not a claim about the web; they are a well-documented sample of one topical corner of it.
What they do establish, within that corner, is that the feature-free results page has effectively stopped existing. On this keyword set, a query with no AI Overview, no People Also Ask block and no video results is the exception at roughly one in eight. That is the useful finding, and it is measured rather than assumed.
Features generated from the index
The first family is text the engine produced by reading pages it had already crawled.
Featured snippets promote an extracted passage — a paragraph, a list, a table or a video segment — above the organic results, with a link to the source. Nothing is written by the engine; a passage is selected and quoted. When a featured snippet is wrong, it is usually because the source is wrong, or because a passage was lifted out of a context that qualified it.
People Also Ask is an accordion of related questions, each expanding to an extracted answer and a source link. Its defining behaviour is that it is generative in structure: expanding one question causes more questions to be appended below it, so the block grows as you use it. It is best understood as a map of the question space around a query rather than a fixed list, which is why two people expanding different branches see entirely different sets.
AI Overviews are the newest and, on the measured data above, now close to ubiquitous for informational queries. Unlike a featured snippet, the text is written — generated by a model grounded on the engine's index rather than quoted from one page — and it carries links to the sources it drew on. Google rolled AI Overviews out to all US users in May 2024, to the UK, India, Japan, Indonesia, Mexico and Brazil in August 2024, and to over a hundred further countries on 28 October 2024. On 27 January 2026, Gemini 3 became the default model behind them globally. There is no official account-level switch to disable them; appending &udm=14 to a search URL, or using the Web tab Google added in May 2024, filters a single search back to plain links.
The practical distinction to hold: a featured snippet is a quotation and an AI Overview is a paraphrase. Quotations can be checked against the source by reading it. Paraphrases have to be checked claim by claim.
Entity panels, blended verticals and navigation aids
Knowledge panels are the boxed summaries of a person, place, organisation or work, usually to the right on desktop. They are assembled from structured entity databases rather than from ranking web pages — Google's Knowledge Graph launched in the United States in May 2012 and expanded internationally by the end of that year, and by May 2016 Google said it could answer roughly a third of its hundred billion monthly searches. Much of the underlying material traces to Wikipedia and Wikidata, which is why panel errors frequently resolve to an edit in one of those rather than to anything a search engine did.
Blended vertical results are the strips and carousels drawn from an engine's non-web corpora: images, video, news, maps and local listings, shopping units, scholarly results. These were the original SERP features, predating everything else, and they are the reason video appeared on 117 of the 120 queries measured above — an informational query about a technology topic reliably pulls video in.
Navigation aids are the least discussed and among the most common. Sitelinks are the indented sub-links shown beneath a single result, typically for a navigational query where the engine has concluded you want a particular site and is offering to skip you deeper into it. They appeared on 92 of the 120 queries measured. Alongside them sit breadcrumb paths in place of raw URLs, "related searches" at the foot of the page, and "people also search for" chips.
Rich results are organic listings visually enhanced from structured data a site publishes in its markup. Google's public gallery documents the supported types, and the list gives a fair sense of how far this has gone: Article, Breadcrumb, Carousel, Course list, Dataset, Discussion forum, Education Q&A, Employer aggregate rating, Event, Image metadata, Job posting, Local business, Math solver, Movie, Organization, Product, Profile page, Q&A, Recipe, Review snippet, Software app, Speakable, Subscription and paywalled content, Vacation rental, and Video. Publishing the markup makes a listing eligible; it does not compel the feature to appear.
Why features exist, and who pays for them
Two forces produced the modern feature-heavy results page, and they point the same way.
The first is user preference. People asking a factual question want the fact, not a shortlist of documents that might contain it. Every feature that answers in place measurably improves that experience, and engines have the data to know it.
The second is economic. A results page that answers keeps the session on the engine. That is more inventory, more context for the next query, and no handoff to a third party. Neither motive is sinister and both are real, and they compound: the feature that pleases the searcher is the feature that retains them.
The cost falls on the sites the answers were extracted from. A query answered on the results page is a query that generates no visit to the source — the pattern usually called zero-click search. This is the central structural tension in search as of 2026, and it is not resolvable by anyone's good intentions: an engine that stopped answering would lose to one that answered, and a web that stops being visited has a harder time funding the material the answers are made of.
Worth noting that the ad units are features too. Text ads, shopping carousels and local ad packs occupy the same page real estate under different rules, and the labelling convention — a small Sponsored or Ad marker — is the only thing separating them from everything else described here.
Features on engines that are not Google
Feature sets differ sharply, and they follow from index ownership and funding.
Microsoft Bing runs its own crawler and index and offers a broadly comparable set — answer panels, entity cards, image and video verticals, its own generated summaries. DuckDuckGo takes its web links from Bing under a direct syndication contract and adds its own Instant Answers layer on top, which is the part it actually builds. It also publishes a documented AI-free endpoint at noai.duckduckgo.com, which is a clearer opt-out than any major engine offers.
Kagi has no advertising features at all, because it has no advertisers, and instead exposes controls conventional engines do not: raising, lowering, pinning or permanently blocking any domain from a subscriber's own results, plus filters for low-value listicle content. That is a feature set aimed at subtraction rather than addition.
Mojeek is the clearest counter-position on the web. In September 2025 it published a post titled "Mojeek is Not an Answer Engine," arguing that Mojeek is about search and that AI is not the answer. It does offer result summarisation using an open-weight Mistral model, but no generated answer is imposed above the results; the default surface is links. Whatever one thinks of the position, it is a stated strategy rather than an unbuilt roadmap item.
And Wolfram Alpha sits at the opposite pole: it does not index the web at all and returns computed answers from curated data. It is, in a sense, a results page made entirely of features and containing no results.
Two documented references worth reading directly are Google's rich results gallery and Mojeek's September 2025 statement on answer engines.
Frequently asked questions
What is a SERP feature?
Anything on a search results page that is not a plain organic listing or a plain text ad — featured snippets, People Also Ask, knowledge panels, image and video strips, local map packs, sitelinks, shopping units and AI Overviews. The term covers a wide range of blocks whose only common property is that they are not the classic title, URL and snippet.
How common are SERP features?
On this site's own August 2026 SEMrush analysis of 120 topical queries, video results appeared on 117, AI Overviews on 104, People Also Ask on 102 and sitelinks on 92. That is one keyword set, one country and one informational subject, so it is not a claim about the whole web — but within that corner, feature-free results pages have effectively stopped existing.
What is the difference between a featured snippet and an AI Overview?
A featured snippet quotes an extracted passage from a single page and links to it. An AI Overview is written by a model grounded on the engine's index and synthesises across sources, with links to those sources. The practical difference is verification: a quotation can be checked by reading the source, whereas a paraphrase must be checked claim by claim.
Where does People Also Ask get its questions?
From related queries and from passages in the engine's index, assembled into an accordion of questions with extracted answers and source links. Its defining behaviour is that expanding one question appends more below it, so the block grows as it is used. Two people expanding different branches will see entirely different sets of questions.
What is a knowledge panel and where does its information come from?
A boxed summary of an entity — a person, place, organisation or work — assembled from structured databases rather than by ranking web pages. Google's Knowledge Graph launched in the United States in May 2012 and expanded internationally that year. Much of the source material traces to Wikipedia and Wikidata, so panel errors often resolve to an edit there.
What are sitelinks?
The indented sub-links displayed beneath a single search result, usually on navigational queries where the engine has concluded you want a specific site and offers to take you deeper into it directly. They appeared on 92 of the 120 queries in this site's August 2026 measurement, making them one of the most common features despite receiving the least attention.
Do all search engines show SERP features?
No. Kagi shows no advertising features at all and instead lets subscribers raise, lower or block domains in their own results. Mojeek published a post in September 2025 titled "Mojeek is Not an Answer Engine" and keeps links as its default surface. Wolfram Alpha returns computed answers from curated data and indexes no web pages at all.
Why do search engines show features instead of just links?
Because searchers asking factual questions want the fact rather than a shortlist of documents, and because a page that answers keeps the session on the engine rather than handing it to a third party. Both motives point the same way. The cost falls on the sites answers are extracted from, in the pattern usually called zero-click search.