How to Teach ChatGPT About Your Company

Teach ChatGPT about your company by giving it approved facts, current reference material, and clear instructions for the job you want done. Put that material where the relevant conversation can use it, then check the answers against its sources. This supplies working context; it does not mean every future chat automatically knows your business.

fenbs keeps shared decisions and current project context outside ChatGPT, so colleagues using different connected assistants can read the same record. Start with one real task, such as drafting a service description. The team brain library explains how that small record can support wider team work.

What is the simplest way to get started?

1. Choose a task and its audience. Writing a customer introduction needs different background from preparing an internal operating checklist.
2. Collect a short company fact sheet and the approved source documents for that task. Remove outdated copies before adding more material.
3. Write the rules for using those facts, including what ChatGPT must ask about rather than guess.
4. Put the material in the relevant conversation or project, and ask ChatGPT to summarize its understanding before drafting.
5. Test a fresh conversation in the intended setting. Correct missing or conflicting context, then repeat the test.

OpenAI’s Projects guide↗ (opens in a new tab) explains how to collect chats, reference files, and project instructions together. It also explains that project instructions apply within that project. Check the available options in your account and workspace rather than treating a project as a company-wide setting.

Which company facts should you include first?

Begin with facts that change the answer: what you sell, whom you serve, where you operate, and what you do not offer. Add the official names of services and a short glossary of company terms. Each changing fact should have a source and a person responsible for keeping it current.

Say a small US architecture studio wants help drafting service pages. Its pack might identify the states where it works, its residential project focus, the approved description of its design process, and the person who reviews public copy. Do not invent client names or credentials to make the example feel complete.

• Company description: an approved paragraph that explains the business plainly.
• Products or services: current names, intended customers, and important exclusions.
• Voice: a few approved examples with a short explanation of what makes them suitable.
• Sources: links or files, their purpose, and when changing information was last reviewed.
• Unknowns: questions that require an owner’s answer before work can continue.

Keep the pack focused on the task. A complete archive is harder to review than a short set of authoritative references. Our ChatGPT for business guide covers the broader choice of work to support.

Where should facts, references, and rules go?

Separate source material from directions. A service handbook describes what the company offers. An instruction tells ChatGPT how to use it: quote the approved service name, distinguish standard work from optional work, and ask when two sources conflict. A draft generated in yesterday’s chat is neither an approved source nor a new policy.

Use project instructions for the behavior needed on that project. OpenAI also provides Custom Instructions↗ (opens in a new tab) for account-level guidance. Keep client-specific commitments and changing project facts out of general preferences, so a useful rule for one job does not become a misleading assumption elsewhere.

A reference URL is useful for people, but check whether the assistant actually accessed its contents. If it cannot read a page, provide an approved excerpt or file you are allowed to share. Tell it which source wins when several documents cover the same subject.

For connected-source retrieval, use the company knowledge connections guide; teaching a technical subject belongs in the new programming language guide. Neither is the same job as curating an approved company reference pack.

How do you turn company policy into useful instructions?

Translate a broad policy into an action the assistant can follow. “Be accurate” is hard to test. “Use the approved service list, cite the source for changing claims, and ask the operations lead when a requested service is missing” gives the assistant a source, a boundary, and a next step.

Make the scope explicit. A company rule might require human approval before promising a delivery date. A project decision might approve one specific date for one customer. Keep both, but do not let the project decision erase the continuing approval rule. Record who chose the decision and why.

Ask ChatGPT to identify uncertainty in its understanding before producing the deliverable. It might know your preferred wording but lack the current list of supported services. That is a useful gap to discover early. In fenbs, Decisions and rules hold the approved instructions for connected assistants to read at the start of work.

How can you tell whether ChatGPT understood the material?

Prepare questions with answers you already know. Ask who the company serves, which service fits a sample request, and what it must not promise. Include one question whose answer is deliberately absent. A sound result should identify that gap rather than invent a convincing company policy.

Run the test in the setting colleagues will actually use. If the material lives in a project, start a new chat there. Ask for source references and open them yourself. Correct the reference pack or instructions when an answer fails, then retest the failed question along with a normal one.

Repeat this check after changing assistants. The Claude and ChatGPT comparison and Copilot and ChatGPT comparison can help you choose an app, but the company’s approved facts still need a maintained home.

How to do this with fenbs

1. Create a free account. One person can start free, with no credit card; use a team kanban board when colleagues need access.
2. Add the first project and a task for the real deliverable. Put company terms in the Glossary and relevant working facts in AI context notes.
3. Record the rules a person approved in Decisions and rules. Add the source references and leave unanswered policy questions open for a person to decide.
4. Follow the Integrations guide to connect a supported assistant. Ask it to read the context, explain the rules relevant to its task, and save progress before the session ends.

fenbs does not import ChatGPT chats or train a model on your company. Keep long handbooks in a suitable document tool and share only information you are authorized to use. If you only need help with one document in one conversation, supplying that document directly may be enough.

What keeps the company context useful over time?

Give each changing source an owner and review it when the underlying business changes. Replace outdated instructions, record new decisions, and remove unsupported claims from the working pack. Before finishing a task, save what was checked and what still needs an answer. That habit makes the next explanation shorter because the record is current.