How do you prompt an AI to stick to only the information you give it, without adding outside knowledge

Why This Comes Up

By default, an AI model draws on everything it learned during training, not just whatever material you’ve handed it in the current conversation — which is a problem when you specifically need an answer grounded only in a particular document, dataset, or set of facts, without general background knowledge mixed in.

The Core Instruction That Helps Most

Explicitly telling the model to answer using only the material provided, and not to draw on outside general knowledge, is the single most direct way to reduce this — stating it plainly rather than assuming it’s implied noticeably changes how carefully the model sticks to the given source.

Asking It to Flag Gaps Instead of Guessing

A closely related instruction — asking the model to explicitly say when the provided material doesn’t cover something, rather than filling the gap with an inferred or general answer — helps surface exactly where it would otherwise be tempted to blend in outside knowledge without flagging that it’s doing so.

Give It the Actual Text, Not a Description

Pasting the actual relevant source material directly into the prompt, rather than just describing what the material says, gives the model something concrete to stay anchored to — it’s much easier for a model to stay grounded in text it can directly reference than in a paraphrased summary of that text.

Bottom Line

Explicitly instructing an AI to use only the material you provide, to flag gaps rather than guess, and giving it the actual source text directly all meaningfully reduce unwanted outside knowledge creeping into its answer — though verifying important claims independently is still worth doing.

Build a Better Prompt

Turn a simple form into a clear, ready-to-copy prompt with our free Prompt Builder — no blank page required.

Go deeper

• Why Do AI Models Have a Knowledge Cutoff Date?
• What is retrieval augmented generation and why does it reduce hallucination?
• How did expert systems rise and then fall out of favor?