The useful result
Turn a long prompt into a numbered checklist and a self-audit that exposes missed requirements.
- Turn a long prompt into a numbered checklist and a self-audit that exposes missed requirements.
You write a careful prompt with eight requirements and the answer follows seven. Which one was missed is the part you only notice later. Long paragraphs of requirements are easy to skim, for people and for models alike.
Two changes help. Number each requirement so none hides inside a sentence, and make the assistant audit its own answer in a table before it finishes. Google’s guidance suggests splitting complex tasks into smaller steps; this is the lightest version of that idea.
The examples below are illustrative exercises, not recorded results from any tool. Sources were checked on 24 October 2026; menus, limits and prices change, so confirm details on the linked pages.
Before you start
- A prompt with several requirements
- An output that missed at least one of them (optional)
Review evidence
Documentation-based review · 2026-10-24
Primary documentation and official guidance checked on 24 October 2026. Examples, checks and exercises are original illustrations, not recorded tool tests. Cover artwork is a conceptual illustration.
Number the requirements, one per line
Rewrite your instructions as a list R1, R2, R3 and so on, with one requirement on each line and no hidden extras inside a sentence. Make each one checkable, such as “under 120 words” instead of “short”. Put the most important ones first.
Check your result
Every requirement has its own number and could be marked met or not met.
Ask for the answer and a requirement-by-requirement audit
Tell the assistant to write the answer, then add a table with each requirement number, whether it was met, and a short piece of evidence from its own answer. If a requirement is not met, it must revise the answer before finishing.
Task: [task]. Requirements:
R1. [requirement]
R2. [requirement]
R3. [requirement]
Write the answer. Then add an audit table with columns: requirement, met (yes/no), evidence quoted from your answer. If any row is “no”, fix the answer and show the corrected version and a new table.Example / expected result
Illustrative example; not a recorded model result.
“R2 under 120 words | yes | Count: 108 words.” Check counts yourself.
Check your result
The table lists every requirement with evidence, and any “not met” row was fixed.
Verify the risky requirements yourself
A model’s audit can be wrong, especially about counts and forbidden words. Check the two or three requirements that matter most, such as word counts, banned phrases or required items, with a counter or a find tool.
Check your result
Your own check confirms the high-risk requirements, and any disagreement with the audit is noted.
When it goes wrong
More than about ten requirements make the audit long and unreliable; split the task instead. Self-audits can claim success falsely, so spot-check. Conflicting requirements cause misses, so remove contradictions.
Make it your own
Take a prompt that missed something, number its requirements, run the audit and count how many rows were wrong when you checked them yourself.
Useful next resources
Sources and review
Last checked Oct 24, 2026. Features and account limits can change. Verify important results against your original material.
Did this help you complete your task?
Tell us what needs improving. No account or private output required.
Report incorrect information or a broken linkGenius Forges Editorial
Practical AI lessons with checked sources, labelled examples and human verification.
