How to Give ChatGPT Reference Material It Can Check

To give ChatGPT reference material, paste the relevant passages, attach the source files, or provide sources through an available connection. Label each source and explain which claims the answer must support from it. Then open the cited passages yourself before relying on the result.

For ongoing work, fenbs keeps reviewed source links beside project context and accepted decisions, so the next connected assistant can find the same starting point. A good reference package explains both what the material says and why it matters to the task.

What should you send before asking for an answer?

1. Write the question you need answered, including the audience and intended use.
2. Choose the smallest set of relevant sources. Keep the full originals available for checking.
3. Give each source a distinct label, title, date, and location, such as a section name or page number.
4. State how to handle disagreements, missing facts, and sources that are out of date.
5. Ask for a source location beside each important factual claim, then verify those locations yourself.

For example, a design studio preparing a proposal might provide an approved scope, a review note, and a list of open questions. Its request should say that the scope defines the promised deliverables while the review note supplies background. Otherwise a tentative comment can become an unsupported commitment.

OpenAI’s ChatGPT usage guide↗ (opens in a new tab) describes supplying files and relevant context. The related question of writing ChatGPT prompts concerns the request itself; here the priority is the evidence behind the answer.

How do you keep instructions separate from evidence?

Put your task and rules first, then introduce a clearly labeled source section. Say that the source contains material to analyze. This makes an instruction such as “compare these options” visibly different from a quoted sentence inside a report.

Use plain labels: Task, Evidence, Accepted decisions, and Open questions. If a reference includes examples or old instructions, identify them as quoted material. Do not mix your current directions into the middle of an unlabeled transcript.

OpenAI’s prompt engineering documentation↗ (opens in a new tab) explains using boundaries to distinguish instructions from context. Clear labels help you review the request too: you can see when a source statement is being promoted into a rule without anyone choosing it.

• Task: “Draft a client update from the supplied sources.”
• Evidence: “Source A is the approved scope; Source B is the review note.”
• Decision: “The client approved the shorter form; keep that choice.”
• Uncertainty: “The launch date is unresolved. Do not invent one.”
• Output check: “Name a source and location for each promised deliverable.”

Should you provide an excerpt, a document, or a link?

Do not infer access from a link appearing in the conversation. Ask ChatGPT to name the document it opened and the relevant section. If it cannot reach the source, provide an authorized excerpt or a readable file rather than asking it to guess what the page says.

OpenAI’s working with files guide↗ (opens in a new tab) covers using source material and reviewing the files produced from it. Keep the source separate from the output so you can compare a generated claim with the original wording.

How do you request answers grounded in the sources?

Specify what counts as support. For a factual comparison, request a claim, its source label, and its location. Ask ChatGPT to distinguish a direct source statement from an inference, meaning a conclusion drawn from that statement.

Include a useful missing-evidence rule: “If the supplied sources do not answer the question, say what is missing.” This is especially useful when the material sounds relevant but does not establish the point you need.

A citation alone is not proof. OpenAI’s citation formatting guide↗ (opens in a new tab) treats source representation and validation as parts of reliable citations. For everyday use, the practical check is simple: open the location and read enough surrounding text to see whether it supports the claim.

Keep reviewed sources with their purpose in fenbs Links and listings, so a future assistant can distinguish the approved scope from a general background article.

What should you do when references disagree?

Ask for the disagreement before asking for a recommendation. Identify which sources differ, what each one actually establishes, and whether they concern the same version or situation. A later date alone does not make a casual suggestion an approved replacement.

In the studio example, an early review might mention five pages while the signed scope covers three. ChatGPT can describe the conflict and prepare a question for the client. A person still needs to decide whether the scope changes.

Source checking matters whichever assistant you use. Our Claude and ChatGPT comparison and Gemini and ChatGPT comparison cover choosing between apps without treating a change of model as a substitute for evidence.

How to do this with fenbs

1. Create a free account and add a project for the work the references support.
2. Add reviewed source locations to Links and listings. Use a context note to explain which source answers which question and which version is current.
3. Record what a person accepted in Decisions and rules. Keep unresolved choices open and identify the missing evidence.
4. Use the ChatGPT guide in Integrations to connect your assistant, then ask it to read the project context before using the sources.
5. Save progress before stopping, including which conclusions were checked, which remain uncertain, and the next source to review.

fenbs does not automatically read every linked page or import your ChatGPT chats. If you need a document reader for one isolated question, supplying that document directly may be sufficient. fenbs becomes useful when the reviewed sources and resulting work must remain available across sessions.

What is the final check before you use the answer?

Choose the claims that would change someone’s action and verify each against the original. Check dates, qualifications, and whether a proposal has become an apparent fact. Correct the answer before reusing it in a proposal, report, or message.

For team use, the ChatGPT for business guide covers a broader working setup. Keep this narrower habit in every setup: the source should explain why a claim belongs in the answer, not merely decorate it. More examples of selecting that context are in our context engineering hub.