SearchGPT Announced — OpenAI Enters the Search Market

On 25 July 2024, OpenAI announced its AI-search prototype, SearchGPT. The post was headed "SearchGPT Prototype" and described the thing itself as "a temporary prototype"; access went to "a small group of users and publishers to get feedback," with everyone else pointed at a waitlist. OpenAI told reporters the test group was about 10,000 users — a figure that appears nowhere in the announcement.

The post also said, up front, that OpenAI "plan[ned] to integrate the best of these features directly into ChatGPT in the future." SearchGPT was never meant to survive as a product; it was a public dress rehearsal.

The industry read the message differently. OpenAI was entering search. The company that owned the generative-AI spotlight through GPT-4 and ChatGPT was walking directly onto Google's home turf.

Why Search

A year and a half on from ChatGPT's November 2022 launch, OpenAI faced three structural problems.

Stale knowledge. GPT models know only the world up to their training cutoff. "Today's news", "the current share price", "the latest paper" — none of these are answerable. From the moment ChatGPT shipped, "today's X" queries returning yesterday's answers were the loudest user complaint.

No citations. LLM answers come with no source. Fact or fabrication: the user cannot tell. Perplexity had already solved this with footnoted citations in December 2022.

The size of the prize. Alphabet booked US$175.0 billion in "Google Search & other" revenue in 2023. If search was about to be redefined by AI, OpenAI needed a direct user surface to plug into that pool.

SearchGPT was OpenAI's joint answer to those three problems.

The Prototype Design

The July 2024 prototype interface carried over almost intact to the later ChatGPT Search.

Input. A natural-language query — question, or conversational instruction. Answer. A written response generated by the LLM, with bullets or tables where appropriate. Sources. To the right of, or at the bottom of, the answer, cards link to the web pages cited. Each card shows a thumbnail, the site, the title, and a short excerpt. Follow-up. Further questions stay in the same thread; context is preserved.

The UX core is "answer first, list afterwards" — essentially the form Perplexity had introduced in 2022. The difference was scale: OpenAI could ship the pattern through ChatGPT's already enormous user base.

OpenAI has never published the architecture. The October launch post gave up exactly two facts: the search model is "a fine-tuned version of GPT-4o, post-trained using novel synthetic data generation techniques, including distilling outputs from OpenAI o1-preview," and it draws on "third-party search providers, as well as content provided directly by our partners." That the third-party provider is principally Microsoft Bing is reporting, not an OpenAI statement. Separately, OpenAI does publish its OAI-SearchBot crawler and states that search indexing is distinct from generative-model training — a site that opts out of training can still surface in results.

The Publisher Deals

Alongside SearchGPT, OpenAI also emphasised publisher partnerships.

The licensing deals had been accumulating since well before SearchGPT: the Associated Press (July 2023), Axel Springer (December 2023), Le Monde and Prisa Media (March 2024), the Financial Times (April 2024), News Corp, Vox Media and The Atlantic (May 2024), Time (June 2024). By the October launch post the acknowledged partner list ran to fourteen: Associated Press, Axel Springer, Condé Nast, Dotdash Meredith, Financial Times, GEDI, Hearst, Le Monde, News Corp, Prisa (El País), Reuters, The Atlantic, Time and Vox Media. Terms were undisclosed; aggregate value was reported to run into hundreds of millions over several years.

This was also a response to the December 2023 lawsuit from The New York Times, which alleged that OpenAI had used Times articles as training data without permission. The deals tried to do three things at once: defend against the NYT litigation, secure legitimate content to display in SearchGPT, and set the economic template for AI search's relationship with publishers.

The presentation stressed that SearchGPT would preferentially cite partner content and would send traffic onward to source sites through the citation links. The message: AI search is not the publisher's enemy, it is their referral channel.

Promotion to ChatGPT Search — 31 October 2024

About three months after the prototype announcement, on 31 October 2024, OpenAI did what it had said it would — shipping the feature not as a standalone product but as ChatGPT search, inside ChatGPT.

Rollout timeline:

• 31 October 2024: "All ChatGPT Plus and Team users, as well as SearchGPT waitlist users, will have access today," on chatgpt.com and in the desktop and mobile apps. Enterprise and Edu were promised "in the next few weeks," free users "over the coming months."
• 16 December 2024: day eight of "12 Days of OpenAI" — rolled out worldwide to logged-in free users.
• 5 February 2025: the sign-in requirement dropped on chatgpt.com; search worked with no account at all. The native apps still required a login.

The product name "SearchGPT" dissolved into "the search feature inside ChatGPT". But the concept — LLM × web search × citations — was promoted to OpenAI's default UX.

A Four-Cornered Contest

From late 2024, AI search formed a four-cornered market.

Google AI Overviews (general US rollout May 2024): AI summaries placed above the existing search results page. The largest user base by default. Publishers complain that users read the Overview and leave without clicking through.

ChatGPT search (shipped October 2024): search folded into the ChatGPT experience. OpenAI reports weekly actives, not monthly — 300 million in December 2024, 400 million in February 2025 — and has never disclosed what share of that traffic is search.

Perplexity (launched December 2022): the search specialist, anchored on explicit citation and an answer-first UX. Its user numbers come from fragmentary company statements and third-party estimates that do not agree even to an order of magnitude, so none is quoted here.

Microsoft Copilot / Bing (New Bing February 2023): the Bing index covered by GPT-4-class models, integrated through Microsoft 365 and Edge.

OpenAI's entry rewrote the picture from "Google versus everyone else" into a multi-polar four-way contest.

Economic and Legal Aftershocks

SearchGPT and ChatGPT Search added new stress to the economics of search.

Publishers split. Partner publishers (News Corp, Vox, The Atlantic etc.) became advantaged; non-partners (notably the NYT, still in litigation) saw their referral chances narrow. Contract relationships, not algorithms, were now picking winners and losers.

Google reacted. Through 2024, Google extended AI Overviews across the United States and then to over 100 countries — defending search-ad revenue while AI-ising the product, a continuous balancing act.

The monopoly ruling. It was the court, not the Justice Department, that made the finding. On 5 August 2024, Judge Amit P. Mehta of the D.C. district court concluded, in a 286-page opinion in US v. Google (No. 20-cv-3010), that "Google is a monopolist, and it has acted as one to maintain its monopoly. It has violated Section 2 of the Sherman Act." The remedies decision followed on 2 September 2025: no forced divestiture of Chrome or Android, but a ban on exclusive default-search contracts and an order to share search index and user-interaction data with rivals. Whether AI search — ChatGPT included — counts as real competitive pressure was central to that phase. Google appealed in January 2026; the DOJ cross-appealed in February.

Copyright litigation continued. NYT v. OpenAI and Microsoft (filed December 2023; motion to dismiss largely denied in April 2025, then discovery) and Dow Jones and NYP Holdings v. Perplexity (filed October 2024) both ran on. Neither had reached trial by August 2026, so the legal status of AI search's use of web content remains unsettled.

Search Becomes the Battleground

The announcement of SearchGPT meant that search had become the central battleground of the AI industry.

In November 2022, when ChatGPT launched, search was one of LLMs' notable applications. From July 2024 onward, search itself had become the core of the LLM competition. Judge Mehta's findings put Google at roughly 90% of desktop searches and about 95% of smartphone searches at the time of suit; against a position of that size, OpenAI, Perplexity and Microsoft were now attacking simultaneously, from different angles.

And once AI handles the search query, the foundational structure of the web economy — "sites live by clicks" — begins to tremble. Publishers, e-commerce, personal blogs, knowledge communities — every content provider faces the scenario in which the AI writes the answer and the user never reaches the original.

The question SearchGPT opened is one of the largest structural shifts in the history of search. As Yahoo!'s 1994 launch discovered that "the web needs a map", SearchGPT's 2024 announcement declared the start of a new way to use that map — read the answer, not the map. What kind of web economy survives on the other side of that change is one of the largest open problems of the late 2020s.