SearchGPT

Long definition

SearchGPT was OpenAI's first dedicated search interface — a stripped-down chat that pulled real-time web results, cited sources inline, and answered with the GPT-4o model on top. The prototype launched July 25, 2024 to a 10,000-user waitlist and ran for three months.

Then it folded into the main product. On October 31, 2024 OpenAI shipped ChatGPT Search to Plus and Team users, and on December 16, 2024 it went out to free-tier users with the same web-grounding stack. The "SearchGPT" name effectively retired with that integration — current docs refer to the feature as "ChatGPT Search" or simply web browsing inside ChatGPT.

How it works under the hood:

• ChatGPT calls a search index (initially partnered with Bing, supplemented with OpenAI's own crawl via OAI-SearchBot and the older GPTBot).
• Retrieved documents feed back into the model as grounded context.
• The model composes an answer with inline numbered citations.
• Follow-up turns reuse the conversation context — true multi-turn search.

Citation patterns in ChatGPT Search lean toward authoritative high-traffic sites: Wikipedia, major news outlets, established review sites, and brand-owned domains for product queries. Long-tail content sites with strong topical authority can win citations on niche queries where the major sites don't cover the angle.

For SEO/GEO, ChatGPT Search is meaningful traffic but hard to measure. The chatgpt.com referrer header strips most query data, and click-through rates per citation are low compared to Google AI Overviews. Track it via referrer logs and brand-mention monitoring rather than via traditional rank tracking.

OpenAI has also shipped agentic browsing (Operator, ChatGPT Agent) on top of this stack — the same retrieval surface, but the model can now act on results, not just read them.

Common misconceptions

• "SearchGPT is still a separate product." It isn't. The standalone surface was a 3-month prototype. The functionality lives inside ChatGPT as ChatGPT Search since October-December 2024.
• "Blocking GPTBot keeps me out of ChatGPT Search." GPTBot is the training crawler; OAI-SearchBot is the live retrieval crawler. To remove yourself from the live answer set you need to block both, plus consider that ChatGPT also uses Bing's index for retrieval. Robots.txt is a partial defense, not a complete one.
• "ChatGPT Search citations drive significant traffic." Per-citation click-through is low — single-digit percentages in most studies. Volume comes from ChatGPT's massive user base, not from per-query CTR. Treat it as a brand-visibility surface first, traffic source second.