ChatGPT Search: the definition

News — Since July 2026 we know ChatGPT Search does not rely on a single engine: it switches between several internal retrieval systems (Labrador, Bright, Oxylabs, SERP). When the system changes, the overlap of cited URLs drops from 0.273 to 0.149, roughly half the shared sources gone (Search Engine Land, 8 July 2026).

Ask ChatGPT a precise question and, instead of reaching into its frozen memory, it will sometimes go and read the live web to answer you, with links to back it up. That capability has a name: ChatGPT Search. It turns ChatGPT into a search engine in its own right, with its own rules for choosing and citing sources. This page gives you a clean definition, its origin, how it really works, how it differs from plain ChatGPT and from Google, and what it changes for a brand. For the wider discipline that grew up around these surfaces, our GEO agency pillar is the place to start.

ChatGPT Search: the one-sentence definition

ChatGPT Search is the search engine built into ChatGPT. For questions that call for it, it queries the web in real time instead of answering from training memory alone, then writes a natural-language answer with clickable sources. OpenAI launched it on 31 October 2024.

A plain chatbot answers from what it learned during training. That knowledge stops at a date, and it holds neither today's news, nor the price posted this morning, nor the page you published an hour ago. ChatGPT Search fills that gap: when a question needs fresh or checkable information, ChatGPT goes and fetches it on the web, then composes an answer that cites the pages it read.

The switch is concrete. Without search, ChatGPT tells you what it thinks it knows. With ChatGPT Search, it shows you where the information came from, through links you can open. OpenAI first opened it on 31 October 2024. For a brand, that changes the game: the question is no longer only whether you rank well in Google, but whether you are the source ChatGPT reads and names at the moment it writes its answer.

Since when ChatGPT Search has existed

OpenAI launched ChatGPT Search on 31 October 2024, first for Plus and Team subscribers and people on the SearchGPT waitlist. It was extended to all logged-in users in December 2024, then to everyone in early 2025.

ChatGPT Search grew out of a prototype called SearchGPT, tested over the summer of 2024 with a small group. At the official launch, on 31 October 2024, OpenAI described it as a way to get fast, timely answers with links to relevant web sources, according to the announcement published on OpenAI's site. Search Engine Land's coverage confirms the date and the initial access reserved to paid tiers.

The rollout was fast. According to Search Engine Journal's ChatGPT timeline, access widened to all logged-in users in December 2024, before becoming available to everyone, with no subscription, in early 2025. In a few months, a niche tool became a search front door for hundreds of millions of people.

How ChatGPT Search works

ChatGPT Search first decides whether the question needs the web. If it does, it queries one or more retrieval systems, then writes an answer that cites the pages it read as links. The user can also force the search manually.

At the start, ChatGPT Search ran on a model OpenAI described as a fine-tuned version of GPT-4o, paired with search results from third-party partners, Microsoft Bing among them. The principle holds in two moves:

1. The trigger. ChatGPT judges whether the question needs live information. A question about a stable fact is answered from memory; a question about today's price, a recent release or a specific business gets sent to the web.
2. The write-up. It reads the retrieved pages, drafts the answer from them, and drops citations inline. A "Sources" button opens a panel listing every page it used.

What decides whether your page is one of them is unglamorous: it has to be in the index and allowed to show a snippet. I have watched a site quietly disappear from these answers for a season because a leftover nosnippet tag was blocking the very passage the engine would have quoted. Nobody had touched the content; the door was just shut. That is the whole trap in one line: a page you cannot see in a normal Google snippet, ChatGPT Search cannot lift either.

Why the cited sources keep changing

The retrieval part is more complex than it looks, and it is the heart of the July 2026 news. According to Search Engine Land's analysis, ChatGPT leans on several distinct retrieval systems, known internally as Labrador, Bright, Oxylabs and SERP, and it switches between them over time. When the system changes, the sources it cites change with it: the overlap of cited URLs moves from 0.273 to 0.149, and the overlap of domains falls by about 42%. In other words, there is no single stable result to measure, which makes visibility tracking harder.

The models move too, and that shifts the citations. Search Engine Journal, in June 2026, relayed a measurement from the SISTRIX tool: after the move to GPT-5.5, the average number of sources cited per answer dropped from 30.9 to 28.4. A detail on its face, but it means fewer slots available to be cited, so harder competition between candidate pages. You can see the same squeeze elsewhere: Google's AI summaries eat into organic traffic, as we documented on the drop in traffic measured with AI Overviews.

ChatGPT Search vs plain ChatGPT vs Google

Plain ChatGPT answers from memory, with no links. ChatGPT Search answers from the live web, with cited sources. Google shows a list of links, sometimes topped by an AI summary. The three differ on freshness, answer format, and the room left for a click.

Confusing these three modes leads to the wrong expectations. The table below sums up the gaps that matter for a brand that wants to be visible.

Note that the boundary is moving. Google rolled out its AI summaries, the AI Overviews, from May 2024, which brings its experience closer to ChatGPT Search. For the generated answers on Google's side, see our dedicated page on AI Overviews. And if you are weighing the two big assistants, we compared their logic in ChatGPT vs Perplexity on visibility.

What ChatGPT Search changes for your visibility

ChatGPT Search serves a single answer where Google served a list. Being in it becomes more decisive than a good ranking with no citation. And because you talk to ChatGPT in natural language, it is often long, precise, low-volume questions that trigger the search.

The click collapses

The first effect is measured in clicks. A Pew Research Center study from July 2025 found that when an AI summary appears in Google, only 8% of users then click a link, against 15% without it, and 26% end their session right after reading the answer. The same shift shows up in ChatGPT Search: the answer is read inside the conversation, and the click to your site becomes secondary. What counts is being the cited source.

The long tail is where it feeds

The second effect is about the nature of the queries. You do not dictate a dry keyword to ChatGPT, you ask it a real question. These long phrasings target long-tail topics, often barely searched individually. This is ground I measure directly: according to Cicero Studio's internal data, of the 5092 French keywords we analysed, 35% get fewer than 100 monthly searches. The long tail dominates, and that is precisely where ChatGPT Search draws its answers. A brand that has only covered its big generic keywords stays absent from the conversation. If the vocabulary here is still fuzzy, our GEO definition and the longer generative engine optimization definition lay out the groundwork.

There is a pattern I see from the other side of the desk too. Across the 1216 SEO and GEO audits Cicero Studio has produced, the recurring problem is almost never a shortage of content. Most of these businesses publish. What they publish is written for a human who has already decided to trust them: the answer lands in paragraph six, the claim carries no source, the page assumes context the machine does not have. A retrieval system cannot lift anything clean out of that, so it lifts from someone else, and nothing in a rankings report ever tells you it happened.

Form beats volume

Finally, the form of the content weighs heavily. The founding study "GEO: Generative Engine Optimization", published in late 2023 by teams from Princeton, Georgia Tech, the Allen Institute for AI and IIT Delhi, showed it on a bench of 10,000 queries: citing sources and adding statistics markedly raises a page's visibility inside a generated answer, while keyword stuffing no longer helps. That result matches what I see by hand. When I run a client's real business questions through ChatGPT Search and read who gets named, the winners are almost always the pages that answer in the first two sentences and back the claim with a figure and a named source; the vague, sourceless pages I audit sit right next to them in Google and never get lifted. The gap is not authority, it is extractability. It is the same logic Google applies to commodity content that AI no longer cites. To dig into the concrete mechanics, I detail how to get cited by ChatGPT and how to appear in ChatGPT answers on two dedicated pages.

Alexis Dollé

CEO & Founder of Cicero Studio

I test AI visibility the slow way, by hand, one real business question at a time, across hundreds of sites. Every week I run ChatGPT Search on live client queries to see who gets cited and why. The pages these systems name are almost never the loudest ones. They are the clearest: a plain answer, stated early, backed by a source you can check. That is the standard we hold every page to at Cicero Studio.

LinkedIn →

What ChatGPT Search does not do

ChatGPT Search has limits that are better stated plainly.

The Cicero approach

At Cicero Studio, we treat ChatGPT Search as an extension of the content, not a gadget. The work starts with an audit that measures where your brand is cited, and where it is not, on Google as well as inside the assistants. Then comes editorial production that turns your expertise into reference pages built to be read and cited, with named sources and your own data. Automated semantic internal linking finally ties those pages together so the engines understand your territory. The promise fits in one line: agency-quality work, software-grade productivity.

Going further

We publish our method in the open, because in this business it is the only proof worth much.