Free small business AI resource

AI Prompt Quality Scorecard

Before a prompt becomes a team habit, test whether it controls facts, matches your voice, reduces editing, avoids risky claims, and can be reused by someone else.

Use the scorecard Copy the worksheet

Five-point prompt scorecard

Score each prompt from 1 to 5

Run the prompt on two or three old examples before using it on live work. If a prompt scores under 3 on any category, improve the prompt before handing it to a team member.

1. Fact control

5: Uses only the supplied facts, flags missing information, and avoids invented prices, dates, policies, guarantees, or results.

3: Mostly follows the facts but needs a careful human edit for assumptions.

1: Adds unsupported claims or sounds confident about information that was not provided.

2. Tone match

5: Sounds like the business: clear, helpful, and appropriate for the customer relationship.

3: Understandable, but generic or too polished for the brand.

1: Sounds robotic, pushy, hype-heavy, or unlike the business.

3. Editing effort

5: A human can approve it with light edits.

3: Saves some time but still needs rewriting.

1: Takes longer to fix than writing from scratch.

4. Risk control

5: Avoids legal, medical, financial, HR, warranty, pricing, privacy, and policy risks unless a human supplies exact approved wording.

3: Includes a useful draft but needs a risk review before sending or publishing.

1: Creates promises, pressure, or sensitive-data exposure that could damage trust.

5. Repeatability

5: Another team member can reuse it because the inputs, variables, output format, and review rule are clear.

3: Works when the original prompt writer runs it, but needs clearer variables.

1: Produces unpredictable results or depends on hidden context.

Copy/paste worksheet

Prompt quality test sheet

Prompt name: [NAME]
Workflow/job: [CUSTOMER FOLLOW-UP / SERVICE PAGE / CALL NOTES / SOP / OTHER]
Tester: [NAME]
Examples tested: [2-3 OLD, NON-SENSITIVE EXAMPLES]

Score 1-5:
- Fact control: [ ]
- Tone match: [ ]
- Editing effort: [ ]
- Risk control: [ ]
- Repeatability: [ ]

What worked:
[NOTES]

What a human had to fix:
[NOTES]

Do not use this prompt for:
[LEGAL/MEDICAL/FINANCIAL/HR/POLICY/PRICING/SENSITIVE CUSTOMER DATA/etc.]

Decision:
[ ] Save as team prompt
[ ] Rewrite and retest
[ ] Retire

Keep / fix / retire rule

  • Keep: average score 4+ and no category under 3.
  • Fix: one category is weak but the prompt still saves time.
  • Retire: the prompt invents facts, creates risk, or needs major rewriting every time.

Improve the prompt

Four prompt edits that usually raise the score

  1. Add a strict fact boundary: “Use only the facts below; flag missing details instead of guessing.”
  2. Define the audience and tone in plain words instead of asking for “professional” copy.
  3. Ask for a human-review checklist after the draft.
  4. Specify what the AI must not do: no prices, policy promises, fake urgency, or sensitive details unless provided.

Related free resources

Use the prompt starter pack for five ready-to-test examples.

Use the 30-day rollout tracker to measure whether your best prompt actually saves time.

Compare prompts vs workflow templates before turning one prompt into a full process.

Want the full workflow kit? The Small Business AI Profit Kit expands this scorecard into prompts, worksheets, rollout planning, and workflow templates.

View the Small Business AI Profit Kit Request resource updates

Free resources stay available without subscribing. No phone, payment, or CAPTCHA required.