A complete starting workflow for defining a task, supplying context, setting constraints, requesting a reviewable output, and verifying the result. 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
- Name one deliverable — define this before asking for a polished final response.
- Separate facts from instructions — define this before asking for a polished final response.
- State what the model must not invent — define this before asking for a polished final response.
- Choose an output structure you can inspect — define this before asking for a polished final response.
- Iterate on a specific failure — define this before asking for a polished final response.
- Verify consequential claims — define this before asking for a polished final response.
What to put in the prompt
For Start Here: A Practical Guide to 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 Start Here: A Practical Guide to 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 name one deliverable.
- Check separate facts from instructions.
- Check state what the model must not invent.
- Check choose an output structure you can inspect.
- Check iterate on a specific failure.
- Check verify consequential claims.
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 Start Here: A Practical Guide to 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
- Choose a real low-risk task you understand well.
- Write down the expected facts or decisions before prompting.
- Run the prompt and mark where the response follows, omits, or invents information.
- Change one instruction and compare the second result.
- 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 Start Here: A Practical Guide to 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.
How to decide whether the prompt is the problem
When an answer is disappointing, do not immediately make the prompt longer. First identify the failure type. If the model misunderstood the task, clarify the objective. If it lacked a fact, add the source. If it ignored a hard rule, move that constraint closer to the task and make the consequence explicit. If the requested output is hard to review, change the format. If the answer is factually uncertain even with good sources and constraints, the limitation may be the model or the task rather than the wording of the prompt.
A practical prompt review rubric
Before reusing a prompt, score it on five questions: Can another person tell exactly what the task is? Are the facts the model must not guess supplied or referenced? Are the non-negotiable limits easy to find? Is the requested output structured so a human can check it? Does the prompt say what to do when evidence is missing? A prompt that passes those checks is usually more reusable than one built around clever phrasing.