What Is a SERP? Search Results Pages Explained

By Bill Hartzer · Updated 19 August 2026

The definition, first

A SERP is a search engine results page — the page an engine returns in response to a query. The acronym is industry shorthand that leaked into general use; the plural, unhelpfully, is "SERPs", which people also use to mean rankings in general.

The most important thing about a SERP is the thing the name obscures. It is not a document sitting on a server waiting to be fetched. It is assembled on demand, in a fraction of a second, out of an index, a ranking function, an advertising auction, a set of feature modules, and a handful of facts about you — your approximate location, your language, your device, the time. Reload the page and it can differ. Ask a colleague to run the same query and it will differ more.

That has a practical consequence which matters far more than the terminology: a screenshot of a results page is evidence of what one person saw once, and not of what an engine "shows" for a query. Journalists, litigators and researchers get this wrong routinely.

The anatomy of a results page

Strip away the styling and a conventional results page has a consistent structure, in this order:

The query box, pre-filled with what you typed, usually with autocomplete predictions attached.
Vertical tabs or filters — Web, Images, Videos, News, Shopping, Maps, and often a date or region filter. These re-run the query against a different corpus or a different set of constraints, not against a different index.
An estimated result count. Not a count. It is a projection from sampled statistics and it varies between refreshes of the same query.
Paid listings, labelled, typically above the organic results and sometimes below them as well.
Organic results. The ranked list from the engine's index. Each entry conventionally carries a clickable title, the source URL or a breadcrumb path, and a snippet — an extract chosen to show the query terms in context. The snippet is generated by the engine at query time, not simply copied from the page.
Features interleaved among the organic results: answer boxes, question accordions, image strips, video carousels, entity panels, map packs.
Related searches and pagination or infinite scroll at the foot.

Two of these are widely misread. The result count is an estimate, so "about 4,320,000 results" is not a measurement of anything. And the snippet belongs to the engine: it is assembled from the page to fit the query, which is why the same page shows different snippets for different searches.

Organic and paid: the line, and how it got there

Organic results are ranked by the engine's own judgement of relevance. Paid results are bought. The distinction is the oldest structural fact about commercial search, and it is younger than the web.

It was invented, wholesale, by GoTo.com — Bill Gross's company, demonstrated at the TED conference in February 1998. Gross's design was that advertisers would bid against each other for a keyword, ranking would follow bid order, and the advertiser would pay only when someone clicked. Bids started at a cent. GoTo did not crawl the web; it backfilled unpaid algorithmic results from Inktomi beginning in June 1998, then sold its paid listings into other engines' pages — MSN in April 2001, Yahoo in November 2001, and AOL. Renamed Overture, it was bought by Yahoo in October 2003. Every search advertisement you have ever seen descends from that auction.

Because paid placement is bought rather than earned, the labelling of it is a consumer-protection matter and has been regulated and litigated for two decades. In practice the label is small, low-contrast and near the top of the unit, and studies of user behaviour consistently find people either do not notice it or actively avoid the results carrying it. The one reliable habit worth having is to look for the word Sponsored or Ad before clicking anything in the first screenful.

No two people get the same page

A results page is conditioned on more than the query. The inputs that reliably change it:

Location, inferred from IP address or device, which reorders anything with a local dimension and changes which country's version of a site is shown.
Language and country settings, which change both the ranking and which features appear at all — AI answer features in particular have rolled out country by country.
Device. A phone page has fewer results above the fold, different feature layouts, and different tap targets.
Signed-in history, where the engine keeps it and personalises on it.
Time. Indexes refresh, news cycles move, and auctions clear differently through the day.

Some engines deliberately remove most of these. Mojeek returns identical results to every user issuing the same query, with no personalisation and no behavioural signals, and says so as a design position. Marginalia Search likewise. That property is unglamorous and genuinely valuable for one use case: if you need a result you can cite, reproduce and hand to someone else, a non-personalising engine gives you something a personalising one structurally cannot.

From ten blue links to something else

The classic SERP — ten organic results, nothing else — was the norm for roughly the first decade of the commercial web, and it has been steadily displaced.

The stages are visible in the record. Blended vertical results brought images, news and video into the main list. Knowledge panels arrived in May 2012 in the United States, expanding internationally by the end of that year, drawing on Google's Knowledge Graph; by May 2016 Google said the Graph could answer roughly a third of its hundred billion monthly searches. Featured snippets promoted an extracted passage above the results. Then generated answers: AI Overviews rolled out to all US users in May 2024, to the UK, India, Japan, Indonesia, Mexico and Brazil in August 2024, and to more than a hundred further countries on 28 October 2024. AI Mode, a conversational surface, followed as a Search Labs experiment in March 2025, and on 27 January 2026 Gemini 3 became the default model behind AI Overviews globally.

As of August 2026, AI Mode is a surface alongside conventional results rather than a replacement for them. There is no official account-level switch to turn AI Overviews off in Google Search; appending &udm=14 to a search URL, or using the Web tab added in May 2024, filters a single search back to plain links.

The direction of travel is that the results page increasingly answers rather than routes. That is a change in what the page is for, and it is the reason "zero-click search" became a phrase — a growing share of queries end on the results page, which is good for the searcher who wanted a fact and structurally difficult for the sites the fact came from.

Results pages that do not look like Google's

The shape of a SERP follows from where the results come from and how the operator is paid.

Microsoft Bing runs its own crawler and index and a broadly similar feature set, with its own ads marketplace. DuckDuckGo takes its web links and images from Bing under a direct syndication contract and layers its own instant answers and interface on top, so its page has Bing's coverage and DuckDuckGo's presentation and privacy posture. Startpage is the cleanest illustration of the distinction: it serves Google's results with Google's tracking removed, so the ranking is Google's and the page is not.

Kagi carries no advertising anywhere, because there is no advertiser — it is subscription-funded — and it offers per-domain controls that let a user raise, lower, pin or permanently block any site from their own results. That is a results page whose ranking the reader can edit, which no ad-funded engine can offer. Marginalia returns a plain list of mostly non-commercial pages and nothing else.

At the far end, Perplexity and similar answer engines return a written answer with numbered citations rather than a ranked list. Whether that counts as a SERP at all is a definitional question, and the useful way to hold it is that the results page has always been a negotiation between listing sources and answering the question. It has simply moved a long way toward answering.

Reading a results page critically

Four habits make a results page much more legible.

Identify what was bought. Anything labelled Sponsored or Ad is placement someone paid for at auction; its position tells you about the advertiser's bid, not about the merits of the page.

Identify what was generated. An AI Overview or featured snippet is text the engine produced or extracted, not a page you are being sent to. Both carry links to sources, and following them is the only way to check the claim against what the source actually says.

Identify what was licensed. Sports scores, flight times, stock prices, weather and dictionary definitions typically arrive in answer boxes through commercial data feeds rather than from crawling. An answer box is not necessarily evidence that a web page said the thing.

Identify what came from an entity database. Knowledge panels are assembled from structured sources including Wikipedia and Wikidata, which is why panel errors often trace back to an edit rather than to the search engine.

None of this requires special tools. It requires knowing that a results page is four or five different systems printed onto one sheet of paper, and that they have different reliability, different incentives, and different owners.

Frequently asked questions

What does SERP stand for?

Search engine results page — the page an engine returns for a query. The plural, SERPs, is also used loosely to mean search rankings generally. The term originated as industry shorthand and is now common in journalism and academic writing about search, though most people who use a search engine have never encountered it.

What is the difference between organic and paid results?

Organic results are ranked by the engine's own assessment of relevance and cannot be bought. Paid results are bought at auction, ranked partly by bid, and must be labelled — usually as Sponsored or Ad. The model dates to GoTo.com in February 1998, which introduced keyword bidding with payment only on a click.

Why do search results look different for different people?

Because the page is assembled per query from inputs beyond the query itself: approximate location, language and country settings, device, signed-in history where the engine keeps it, and the time. Feature availability also varies by country. Mojeek and Marginalia deliberately do not personalise, returning identical results to every user, which makes them reproducible for research.

Is the number of results shown at the top accurate?

No. It is an estimate projected from sampled index statistics, not a count, and it commonly varies between refreshes of the same query. Engines also cap how deep results can be paged, so it is not possible to page through to the end and verify the figure. Treat it as an order of magnitude at best.

What are the ten blue links?

Shorthand for the classic results page: ten organic listings and nothing else, the standard format for roughly the first decade of commercial web search. Blended vertical results, knowledge panels from May 2012, featured snippets and then AI Overviews from May 2024 progressively displaced it. Conventional results still exist but rarely occupy the whole page.

Do all search engines have the same kind of results page?

No, and the differences follow from funding and index ownership. Startpage serves Google's ranking without Google's tracking. DuckDuckGo layers its own interface over Bing's links. Kagi has no advertising at all and lets subscribers raise, lower or block domains in their own results. Answer engines such as Perplexity return a cited written answer rather than a ranked list.

Can you turn off AI Overviews on a search results page?

There is no official account-level setting to disable them in Google Search as of August 2026. 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. DuckDuckGo, by contrast, publishes a documented AI-free endpoint at noai.duckduckgo.com.

What is a SERP?