OpenAI Introduces Company Knowledge for ChatGPT Business, Enterprise, and Edu

TL;DR

OpenAI announced Company Knowledge for ChatGPT Business, Enterprise, and Edu, a GPT‑5‑powered feature that pulls context from connected internal apps (Slack, SharePoint, Google Drive, GitHub, etc.) and returns answers with clear citations, while respecting existing permissions and enterprise‑grade security.

What Company Knowledge Does

  • Aggregates internal data: Connects to tools such as Slack, SharePoint, Google Drive, GitHub, Intercom, HubSpot, Linear, Asana, GitLab Issues, ClickUp, and more.
  • Provides citation‑backed answers: Every response lists the exact source snippets and lets users click through to the original document.
  • Runs on a GPT‑5 variant: The model is trained to search across multiple repositories, resolve conflicts, and apply date filters for time‑sensitive queries.
  • Available today: All users on ChatGPT Business, Enterprise, and Edu can enable the feature immediately.

How to Use It

  1. Tap Company knowledge under the message composer.
  2. Connect the desired work apps on first use.
  3. Enable the toggle for any conversation; the sidebar shows live searches and the final citation list.
  4. Click a citation to open the source for deeper inspection.

Core Technical Capabilities

Multi‑source Search and Conflict Resolution

Company Knowledge can execute several parallel searches, compare overlapping information, and surface differing viewpoints. This helps answer ambiguous questions such as “Where did we land on company goals for next year?” by summarizing discussions and highlighting disagreements.

Context‑aware Summarization

When asked for a report (e.g., “customer feedback from the mobile launch”), the system pulls sentiment data, support tickets, and direct customer quotes, then suggests next steps, delivering a more balanced view than a single‑source summary.

Time‑based Retrieval

The model can filter by date, rank results by recency, and retrieve information from specific periods (e.g., “company performance in Q1”). It aggregates recent Slack messages, Google Docs, and Sheets to present the freshest snapshot.

Privacy, Security, and Control

  • Permission‑aware access: ChatGPT can only read data the user already has permission to view.
  • No default training on your data: OpenAI does not use company data to train its models unless explicitly opted‑in.
  • Enterprise controls: Admins can set custom roles, group‑level permissions, and manage app connections via the Enterprise and Edu admin consoles.
  • Encryption & access management: Data is encrypted in transit and at rest; SSO, SCIM, and IP allow‑listing are supported.
  • Compliance: Conversation logs are accessible through the Enterprise Compliance API for audit and regulatory reporting.

Practical Use Cases

  • Turn customer insights into strategy docs: Synthesize Slack feedback, Google Slides survey results, and Intercom tickets into roadmap recommendations.
  • Create up‑to‑date reports: Pull HubSpot contacts, Google Docs post‑mortems, and email highlights to generate performance summaries.
  • Build release plans: Scan GitHub TODOs, Linear tickets, and engineering Slack channels to list outstanding work and identify gaps.

Current Limitations

  • Manual activation: Users must enable Company Knowledge per conversation; otherwise, the model may use connected apps without citation depth.
  • Feature isolation: When enabled, the model cannot browse the web or generate charts/images. Users can toggle it off within the same thread to regain those capabilities.
  • Connector coverage: Although many integrations are available, some internal tools may still lack connectors.

Roadmap and Future Enhancements

  • Unified experience: OpenAI plans to merge Company Knowledge with the full suite of ChatGPT capabilities (web browsing, visual generation) in the coming months.
  • Expanded connectors: New integrations for Asana, GitLab Issues, ClickUp, and additional tools are being rolled out.
  • Deeper AI reasoning: Ongoing work aims to improve the model’s ability to synthesize conflicting information and provide actionable recommendations.

All information is based on OpenAI’s official announcement dated 2025‑10‑23.

Sources