# Practical: run a prompt comparison

AI Content Creation | Module 03: Write prompts you can improve

## Your goal

Test two versions of an announcement prompt and keep a short comparison log.

Bring together the two lessons in this module. Test two versions of an announcement prompt and keep a short comparison log. Use the fictional Green Bench source pack, and keep missing details visible in your working notes.

Your finished work should include: Two prompts with one deliberate difference. Two labelled outputs or manual sample drafts, with a reasoned choice. Keep the actual text in the workspace. For media, describe your file or storyboard, its version, and the decisions it shows. Media files stay on your device; this form saves text only.

Try this review experiment: Score each draft for fact accuracy, reader fit, and a clear action. Keep the scores and the words that support them. Record what changed and why. You may use your own AI tool or work from the supplied sample drafts. Label which approach you used. The checklist checks completeness and your review record, not creative quality.

## Source pack

FICTIONAL PRACTICE BRIEF: Green Bench is a student repair club. It runs a free notebook-cover repair workshop on Saturday, 14:00 to 16:00, in Room B. There are 12 places. Learners bring one worn notebook; tools and scrap card are provided. Register using the club form. No date, form URL, access arrangements, impact figures, or testimonials have been supplied. Ask the organiser about missing details before publishing. This is a creative exercise, not a real event.

## Worked example

Test two versions of an announcement prompt and keep a short comparison log.

### Your finished work

Two prompts with one deliberate difference. Start from the examples in the two earlier lessons, then write your own complete version or a detailed account of your media asset.

### Evidence and review

Use this experiment: Score each draft for fact accuracy, reader fit, and a clear action. Keep the scores and the words that support them. Record the observations, source checks, and any unresolved questions.

### Revision and next step

Describe a specific change to this assignment and explain how it helps the reader. Keep the previous version for comparison.

## Guided practical

1. Revisit lessons 03-01 and 03-02 and collect your drafts.
2. Test two versions of an announcement prompt and keep a short comparison log.
3. Use your own tool or the supplied examples. Keep the source facts, prompt, and draft together.
4. Score each draft for fact accuracy, reader fit, and a clear action. Keep the scores and the words that support them.
5. Revise the work, complete the worksheet, and record any remaining publication blockers.

### Deliverables

- Two prompts with one deliberate difference.
- Two labelled outputs or manual sample drafts, with a reasoned choice.

## Your work

Test two versions of an announcement prompt and keep a short comparison log.

### Your finished work

Write at least 30 words.


### Evidence and review

Write at least 20 words.


### Revision and next step

Write at least 15 words.

## Review your work

- [ ] I included two prompts with one deliberate difference.
- [ ] I included two labelled outputs or manual sample drafts, with a reasoned choice.
- [ ] I recorded an actual review result and a specific revision.

Completeness checks do not assess creative quality or factual accuracy. Review the actual work yourself or with a mentor.

## Quiz

1. What belongs in the review record for run a prompt comparison?
   A. Score each draft for fact accuracy, reader fit, and a clear action. Keep the scores and the words that support them.
   B. A claim that the work is perfect without checking it.
   C. Only the name of the tool used.

2. Two prompts produce different results on a second run. What should your log record?
   A. Only the result you liked.
   B. A guarantee that the first result always repeats.
   C. The outputs, tool and date, and comparison limits.

## Quiz explanations

1. A: A useful review records an action, an observation, and what you changed.

2. C: Generated results can vary, so keep the evidence and limits.

## Further reading

[Anthropic: prompt engineering overview](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/overview)