OpenAI is pushing ChatGPT further into the workplace with Company Knowledge, a new feature meant to help users search internal company information across common work tools. The feature is designed for business, enterprise, and education users, and it connects ChatGPT to sources such as Slack, SharePoint, Google Drive, and GitHub.
The idea is straightforward: employees ask questions in natural language, and ChatGPT returns answers with source references. The harder question is whether a large language model can reliably act as a search engine for complex work data.
What Company Knowledge is meant to do
Company Knowledge is powered by GPT-5. OpenAI is positioning it as a way for ChatGPT to search internal company information rather than only respond from a general model context.
According to the source article, the feature can work with vaguely worded questions. That matters because workplace search is often messy. People may not know the exact file name, thread, repository, or document where an answer lives. They may only remember part of a project, a decision, or a discussion.
The feature also includes source references. In a work setting, that detail is central. A useful answer is not just a summary; it should also help the user understand where the answer came from and whether it is grounded in the right material.
There are limits built into the current feature. Company Knowledge must be enabled manually. It also does not support web searches, and it does not generate images or diagrams.
Why the workplace search pitch is appealing
The appeal is easy to understand. Work information is often spread across chat messages, shared folders, code repositories, and internal documents. Even when the information exists, finding the right source can take time.
A ChatGPT feature that searches across Slack, SharePoint, Google Drive, and GitHub could help users surface relevant material faster. It could be especially useful for exploratory searches, where the user is trying to locate the right sources before making a decision or completing a task.
That is where large language models already tend to be useful: they can help users navigate a fixed context, summarize material, and point toward potentially relevant information. In that role, ChatGPT can act less like a final authority and more like a guide through a large pile of work data.
But the more ambitious version of the pitch is harder. If ChatGPT becomes the place where employees ask broad questions about internal knowledge, the system has to retrieve the right material, interpret it correctly, cite it clearly, and avoid mixing signals from unrelated sources.
The citation problem is still unresolved
The source article is clear that LLM-based systems still struggle when citations span broad, open data sources. Pulling from multiple sources at once can produce answers that are unclear or wrong. The problem is not limited to one product; every LLM-based system faces citation challenges at this scale.
These systems can give inaccurate details, omit important information, or misunderstand context. That risk becomes more serious when the answer appears confident and includes references, because users may assume the response has been fully checked.
The article also points to a recent study that calls some poor AI output "AI workslop". The source says this is already costing companies millions and hurting morale.
For workplace use, that is not a small concern. A flawed internal answer can waste time, send people toward the wrong document, or create false confidence around a decision. The issue is not only whether ChatGPT can answer quickly. It is whether users can trust the answer enough to act on it.
Context quality may decide answer quality
Another major issue is context. Research cited in the source article shows that irrelevant information in long contexts can hurt model performance. More input is not automatically better input.
This is why context engineering is becoming more important. The source describes it as carefully selecting and structuring the information given to the model. If the context is too broad, too narrow, or poorly organized, answer quality can drop and costs can rise.
Google DeepMind researcher Nikolay Savinov notes that choosing the right context is essential, even though today’s models can handle hundreds of thousands or even millions of tokens.
That point matters for Company Knowledge. Searching internal company data is not just a retrieval problem. It is also a filtering and framing problem. The system has to decide what information belongs in the answer and what information should be left out.
What companies should keep in mind
Company Knowledge could be useful when treated as an aid to search and exploration. It can help users ask loose questions, gather source references, and move faster through scattered workplace data.
But the source article argues that companies need to make sure employees understand the limits. Human expertise and review remain critical. Current technology can still make mistakes, and relying entirely on the model will inevitably lead to errors.
That creates a practical rule for adoption: use ChatGPT to help find and frame information, but do not treat it as the final reviewer of company knowledge. The value is in faster discovery. The risk is in mistaking a fluent answer for a verified one.
OpenAI’s official announcement, according to the source article, does not mention these risks. For teams considering Company Knowledge, that makes internal guidance important. Employees need to know when to trust the tool, when to check the references, and when to bring in human judgment.