Using Reference Text in AI Prompts

Ground a model in supplied documents while keeping instructions, source text, and requested analysis clearly separated. This guide treats AI as a drafting and reasoning aid, not as an authority. The goal is to make the task explicit enough that you can review the result against real source information.

The core workflow

1. Label source blocks — define this before asking for a polished final response.
2. State whether outside knowledge is allowed — define this before asking for a polished final response.
3. Preserve quotations accurately — define this before asking for a polished final response.
4. Ask for line/section references when possible — define this before asking for a polished final response.
5. Identify missing source coverage — define this before asking for a polished final response.
6. Avoid unsupported synthesis — define this before asking for a polished final response.

What to put in the prompt

For Using Reference Text in AI Prompts, start with the exact task and intended user. Then provide the source facts the model cannot safely infer. Put critical requirements close to the task, separate pasted source material from instructions, and say what the response should do when information is missing.

• State one concrete deliverable or decision.
• Supply relevant context and authoritative source material.
• Name facts, commitments, or definitions that must remain unchanged.
• Specify the output structure so the result is easy to inspect.
• Require assumptions, unknowns, or unsupported claims to be visible.

A reusable prompt pattern

How to review the first response

Do not judge the first answer only by whether it sounds polished. For Using Reference Text in AI Prompts, review whether the model followed the source, respected the stated limits, and produced something that can be checked. Highlight every claim that depends on a fact, date, number, policy, quote, citation, credential, or technical conclusion.

• Check label source blocks.
• Check state whether outside knowledge is allowed.
• Check preserve quotations accurately.
• Check ask for line/section references when possible.
• Check identify missing source coverage.
• Check avoid unsupported synthesis.

Common failure pattern

Iterate deliberately

When a response is weak, identify the largest specific failure and revise that instruction. Preserve the parts that are already correct. For Using Reference Text in AI Prompts, a useful second pass might add missing source material, tighten one scope boundary, change the requested structure, or require a claim-by-claim verification list.

Practice exercise

1. Choose a real low-risk task you understand well.
2. Write down the expected facts or decisions before prompting.
3. Run the prompt and mark where the response follows, omits, or invents information.
4. Change one instruction and compare the second result.
5. Save the final prompt only after you understand which instruction produced the improvement.

Primary documentation

Primary references

For current ChatGPT-specific prompting guidance relevant to Using Reference Text in AI Prompts, consult OpenAI’s official Prompt engineering best practices for ChatGPT and How do I create a good prompt for an AI model?. Product behavior can change, so current product documentation should take priority over older tips or screenshots.

Mark the boundary between source and instruction

Reference text should be visually and logically separate from the directions you give the model. Use labels or delimiters such as SOURCE MATERIAL, END SOURCE, TASK, and OUTPUT REQUIREMENTS. Then tell the model whether it must rely only on the source, may use general knowledge for background, or should flag anything the source does not establish. This reduces the chance that language inside a pasted document is mistaken for an instruction or that outside assumptions are blended into a source-bound answer.

Prepare long source material before asking for a final draft

If the source is lengthy, first ask for an evidence map rather than a finished answer. Have the model identify sections relevant to the task, extract important facts with location markers, list conflicts or missing information, and summarize the allowed evidence. Review that map, correct mistakes, and only then request the final output. This extra step is especially valuable when the source contains multiple versions, tables, exceptions, footnotes, or language where a small qualification changes the meaning.