OpenAI launched a unified cross-search system for ChatGPT on Tuesday that lets users query their entire history of conversations, uploaded images, documents, and Projects from a single entry point in the sidebar. The feature went live today across ChatGPT's web, iOS, and Android apps and is available on every plan tier — Free, Go, Plus, Pro, Business, and Enterprise — simultaneously at global launch.

The update crosses a meaningful threshold. ChatGPT has spent nearly four years functioning as a session-bound conversation tool, where content generated in one chat stayed in that chat unless a user manually organized it into Projects. With a searchable index now spanning everything a user has ever created or uploaded inside the product, ChatGPT has effectively become a personal knowledge base — a searchable record of a user's accumulated thinking, questions, documents, and files held and indexed by OpenAI. That is a different kind of product with different implications for what it holds.

Cross-Search Makes Every Chat and Project Instantly Findable

The new search experience is accessible directly from the ChatGPT sidebar. Users can query across all past conversations, Projects, uploaded images, and documents in a single search, then narrow results using filters by content type — finding only images, for instance, or only files attached to a specific Project. Selecting any result opens the relevant chat, Project, or file directly in ChatGPT, without losing context.

The breadth of the index is what distinguishes this from anything ChatGPT has previously offered for navigation. Until now, users were reliant on scrolling through a chronological sidebar list or manually organizing conversations into Projects — OpenAI's workspace feature, which groups related chats and files with shared instructions and a persistent memory scope. The new unified search spans all of that, including content inside Projects, without requiring users to know which Project a file or conversation belongs to.

Why Free Plan Access Signals a Strategic Shift

The decision to launch on all plans simultaneously — rather than gating access behind paid tiers first — is a meaningful competitive signal. OpenAI has typically used significant feature releases as conversion levers, giving paid subscribers early access before opening to free users. A simultaneous free-tier launch suggests the company is prioritizing breadth of engagement and platform stickiness over immediate monetization, a posture that reflects growing competitive pressure from Google Gemini and Anthropic's Claude, both of which have been expanding their own cross-session memory and retrieval capabilities.

The feature also fits into a broader platform build-out that accelerated significantly in the first two weeks of July. On July 9, OpenAI launched ChatGPT Work — a long-horizon task agent for producing finished documents, spreadsheets, and web apps across connected tools — and simultaneously merged Codex, its coding agent, into a new unified ChatGPT desktop app. The unified search feature slots into this architecture as the retrieval layer that makes the growing volume of content users generate across these surfaces actually findable. Without it, ChatGPT Work's output would accumulate in a searchable void.

The same July 9 release also began formal deprecation of Atlas, OpenAI's standalone AI browser application. Atlas is scheduled to stop working on August 9, 2026, as OpenAI consolidates browser-based capabilities into the ChatGPT desktop app's built-in browser. Users who relied on Atlas for persistent browser memory should export bookmarks and saved passwords before that date; the data will not transfer automatically. GPT-5.6 — the new model family comprising Sol for high-complexity reasoning, Terra for balanced performance, and Luna for high-volume workloads — is also rolling out globally across ChatGPT, Codex, and the API in parallel with today's search release.

How Does ChatGPT's Search Actually Work — and What Can It Not Find?

Understanding what cross-search indexes — and what it does not — matters as much as knowing what it does. OpenAI has not published a technical specification for the index architecture, but the feature's behavior and the structure of adjacent ChatGPT systems point toward a design built around how semantic embeddings work: a process in which content chunks are converted into dense numerical vectors that capture meaning rather than just keywords, enabling the system to surface a document about "machine learning infrastructure" in response to a query about "AI compute setup" even when the exact words don't match.

The critical constraint is scope. The index covers only content created or uploaded inside ChatGPT itself: past conversations, Projects, images, and documents. It does not extend to content inside the tools that ChatGPT's newly launched Plugin system connects to — including Slack, Microsoft Teams, Google Drive, and SharePoint. A user who uploads a document to ChatGPT, asks Work to analyze it, and saves the resulting report inside a Project can find all of that through cross-search. A user who asks Work to pull a document from Google Drive and summarize it cannot find that source document through cross-search, because it was never indexed inside ChatGPT — it was fetched and used, but not stored.

This distinction matters for how the feature competes with the enterprise knowledge management platforms it most resembles. Tools like Glean, Atlassian Rovo, Notion AI, and Guru are built specifically to index knowledge across an organization's entire tool stack — 100 or more enterprise applications — with a knowledge graph that models relationships between people, content, and activity, and with real-time permission inheritance from each source system. According to Glean's enterprise search evaluation, published in February 2026, human graders preferred Glean's answers 1.9 times more often than ChatGPT's results for complex queries that required understanding company-specific jargon, multi-hop reasoning, or retrieving the authoritative version of a document rather than any version.

For individual users and small teams who generate most of their knowledge work inside ChatGPT itself, the cross-search feature effectively replaces the need for an external retrieval layer. For larger organizations with knowledge distributed across dozens of platforms, it deepens the personal retrieval experience without yet bridging the gap into indexed organizational memory at the network level. OpenAI has not announced a roadmap for extending cross-search to content in connected external apps.

What OpenAI Holds When You Search Gets Bigger

The shift from chat tool to personal knowledge base is not only a usability story. It is a data story.

Until today, ChatGPT's data model was primarily a chronological log: conversations stored sequentially, findable only by scrolling or by memory of approximate date. A semantic index changes what that data is. When a retrieval system can surface "the Q3 planning document I uploaded eight months ago" or "the conversation where I worked through a technical architecture decision last March," what OpenAI holds is no longer a log — it is a searchable record of a user's professional and intellectual work, organized by meaning and accessible by natural-language query.

OpenAI's standard data handling policy for consumer plans (Free, Plus, Pro) allows conversations to be used to train future models unless a user explicitly opts out through Data Controls. Deleted conversations are removed from OpenAI's systems within 30 days under normal circumstances. Those circumstances, however, have already been suspended once: a May 2025 federal court preservation order in the ongoing New York Times v. OpenAI copyright lawsuit required OpenAI to retain all ChatGPT conversations indefinitely, including conversations users had already deleted. In January 2026, US District Judge Sidney Stein ordered OpenAI to produce 20 million de-identified ChatGPT logs to news plaintiffs. Data from the April–September 2025 window remains in secure storage pending ongoing litigation.

The court order established something the standard privacy policy cannot guarantee: the right to delete can be suspended by judicial action without user consent, and preserved data can be disclosed to third parties through discovery. As the volume and semantic richness of ChatGPT-native content grows — which is precisely the point of making it a searchable personal knowledge base — so does the significance of that precedent.

Users who want to limit what ends up in the index have several practical options: opt out of model training in Data Controls (Settings → Data Controls → toggle off "Improve model for everyone"), use Temporary Chat mode for sessions containing sensitive content (Temporary Chats are not saved to history), delete specific conversations or Projects after use, and review and delete saved memories separately through the memory summary page, since deleting a conversation does not automatically remove memories extracted from it. Business and Enterprise workspace admins can set custom data retention policies with a minimum of 90 days.

ChatGPT Work and the Broader Platform Consolidation

The search launch is part of a platform architecture shift that OpenAI has been assembling throughout July, and understanding the architecture helps explain why cross-search exists now and what role it plays.

ChatGPT Work, which launched July 9, is an agentic layer that takes a goal, gathers context across connected apps and workflows, breaks the task into steps, and returns finished output — a spreadsheet, a slide deck, a document, a hosted web app — rather than a conversational reply. Work introduced Scheduled Tasks, which allow users to automate recurring work (a weekly summary of new Slack messages, a daily check of a connected dashboard) without manual prompting. The Codex coding agent, previously a separate application, is now part of the same unified desktop app that runs Chat, Work, and Codex as a single product on macOS and Windows.

What this architecture produces is an increasing volume of ChatGPT-native content: reports generated by Work, code produced by Codex, documents uploaded as context for Projects, images created inside conversations. Cross-search is the retrieval layer that makes all of that findable. Without it, the platform consolidation would have created a problem identical to the one it was solving — a large, growing collection of valuable work with no efficient way to navigate it.

The competitive implications extend to enterprise software. ChatGPT's expanding memory and retrieval footprint puts it in closer proximity to platforms like Glean, Atlassian Rovo, Notion AI, and Guru — tools built specifically to surface institutional knowledge distributed across documents, messages, and project history. What OpenAI is doing is currently narrower — unified search across content inside ChatGPT itself, not a cross-app enterprise knowledge graph — but it addresses the same core user frustration at the personal productivity level: the growing difficulty of finding something you already did or uploaded as AI-assisted work accumulates at scale.

OpenAI has not announced when or whether cross-search will extend to content in connected external apps. The Plugin directory, which launched alongside ChatGPT Work on July 9, already connects ChatGPT to external tools including Slack, Microsoft Teams, Google Drive, and SharePoint. Whether the index will eventually span those connected sources remains an open question — and the answer will determine how directly the feature competes with enterprise knowledge management platforms rather than simply complementing them.

Frequently Asked Questions

How do I search my old ChatGPT conversations and files?

Starting today, the unified search is accessible directly from the ChatGPT sidebar on web, iOS, and Android. Type your query into the search box and it will return results across all your past conversations, Projects, uploaded images, and documents. You can then filter results by content type — for example, to show only images or only files from a specific Project. Selecting a result opens the original chat, Project, or file directly in ChatGPT without losing your current context. The feature is available on all plan tiers, including Free, at no additional cost.

Can ChatGPT find content in Slack, Google Drive, or my other connected apps?

Not yet. The cross-search index covers only content created or uploaded inside ChatGPT itself — past conversations, Projects, images, and documents. It does not extend to content in the external apps that ChatGPT's Plugin system connects to, even if those apps are actively connected to your account. If you want cross-search to find a document, it needs to be uploaded directly into ChatGPT or generated inside a ChatGPT Project or Work session. OpenAI has not announced a timeline for extending cross-search to external connected sources.

What does ChatGPT becoming a searchable knowledge base mean for my privacy?

As cross-search makes ChatGPT's index of your work more valuable and more navigable, it also makes the data OpenAI holds more sensitive. For consumer plan users (Free, Plus, Pro), conversations are used to train OpenAI's models by default unless you opt out in OpenAI Data Controls settings. Deleted conversations are removed from OpenAI's systems within 30 days under standard policy — but a 2025 federal court preservation order in the New York Times v. OpenAI lawsuit demonstrated that this timeline can be suspended by judicial action, and in January 2026 a federal judge ordered production of 20 million de-identified ChatGPT logs to news plaintiffs. To reduce your exposure: opt out of model training, use Temporary Chat mode for sensitive sessions, delete conversations you no longer need, and manage saved memories separately in Settings, since deleting a chat does not delete the memories extracted from it.

Does ChatGPT's cross-search replace tools like Notion, Glean, or Guru?

For individuals and small teams whose work lives primarily inside ChatGPT, yes — cross-search effectively replaces the need for a separate personal retrieval layer. For larger organizations, the answer is more nuanced. According to a 2026 enterprise knowledge management tool comparison, platforms like Glean, Atlassian Rovo, and Guru index knowledge across 100 or more enterprise tools simultaneously, with organizational knowledge graphs, real-time permission inheritance from each source system, and retrieval models trained on company-specific jargon. A February 2026 evaluation found enterprise search tools outperformed ChatGPT on complex organizational queries by a factor of nearly two to one. ChatGPT's cross-search is a strong personal retrieval tool for ChatGPT-native content; it is not yet a cross-organizational knowledge graph. Whether that gap narrows depends on whether OpenAI extends the index to connected external apps.