A better prompt?
Prove it.
You already use Claude at work.
Now make your next answer measurably better.
Same enquiry.
Same rubric.
Better instructions.
Course exercise · 12 September 2026
Read the evidence
One customer enquiry. A price the brief never gave.
Sounding good
isn’t the test.
This case tempts Claude to invent a price. Both prompts below were tested in the same run of the course’s own exercise, against the same marking rubric.
Course exercise · customer enquiry · 12 September 2026
7 of 8 Pass
“You are answering a customer enquiry for the business described in the brief below. Use only the facts in the brief. If the brief does not answer something the customer asked, say plainly that you will check and come back, and never invent a fact, a price or a timescale to fill the gap. Write 90 to 150 words of plain British English, warm but not gushing. Answer every question the customer actually asked, in the order they asked it. Close by telling the customer what happens next.”
Facts, not guesses.
The brief is the boundary.
A gap stays a gap.
Check and come back. No invented price.
Sendable work.
Answer the questions. Say what happens next.
The difference is in the instructions.
The evidence is in the run.
One real job.
Six awkward cases.
Nothing to bluff.
You keep the same job through the course. Each lesson ends in a run you can score.
- 1
Pick your job.
Choose one of thirteen, from a customer enquiry or job advert to a proposal, board update, support reply or incident report.
- 2
Write. Then improve.
Write your prompt in the page, on top of the version you last saved.
- 3
Give it the awkward cases.
Six answers, produced separately. A complaint. A job the business cannot do. A temptation to invent. An easy case cannot carry the others.
- 4
Let the rubric decide.
Mechanical checks first. Then a grader: four rules the answer must satisfy and eight it should. It never sees your prompt.
- 5
Keep the receipt.
Version, case, pass or fail, score out of eight. Your run log shows whether the new version beat the old.
“It cannot
be flattered.”
Nell Garvey, on a grader that never sees your prompt.
Nell is the person behind these courses. She writes them, runs them and answers problems herself.
The grader gets the brief, the case and the answer. Your carefully worded prompt gets no marks for charm.
“Nothing here rests on being told a prompt is good.”
How the course checks itself
Every quotation carries its source page and the date it was read. A scheduled script re-fetches quoted pages and fails if a sentence disappears.
Lessons explain what features do, without interface screenshots that age as buttons move. Two claims are deliberately left unmade because nobody had tested them.
One prompt.
Improved twenty times.
Eight modules and a short start-here. You leave with a named, versioned prompt for your own work, and a run log showing the new version beat the old.
23 lessons
5–8 minutes each
~150 minutes in all
See what you’ll work through Start here to capstone
- Start hereWhat Claude is, and is not. Your first prompt and baseline score.
- ContextWhat to give it, source material, and when context is too much.
- TaskWhat, for whom, why, and the decision the answer supports.
- RequirementsYour definition of done. Facts, assumptions and unproven claims. Rules you enforce.
- ExamplesOne useful example and the near miss.
- Evidence and testingReceipts, versions, run logs and diagnosing a weak answer.
- Claude’s own featuresProjects, Skills, memory, search and emphasis.
- CapstoneYour own job, end to end.
- Next doorWhat Claude Code adds.
Also open alongside it
Your Own Job, Scored
16 lessons · About 105 minutes. Your job on two models, a rubric you have argued with three times, and a changelog.
Checking Claude’s Work
16 lessons · About 101 minutes. A checking prompt tested on drafts with known faults: what it caught, missed and wrongly flagged.
Your first prompt
is your starting point.
Free while the courses are in private beta.
Make an account and start Better Answers from Claude.
Before you start
Your account keeps the evidence. Prompt versions, runs and scores need somewhere to stay so you can compare them.
Runs are capped. Each exercise uses twelve model calls. The allowance covers every exercise with room to retry. The runner explains when an account or course ceiling is reached.
This is private beta. All lessons are there and marking runs end to end. The adversarial audit is still ahead. Free runs remain deliberately modest while costs are watched.