Prompt for A/B test ideas with a clear hypothesis
Get six one-change tests, each written as a hypothesis with its deciding number and what you will do with each result. Copy the prompt below into Claude or ChatGPT, replace the words in square brackets with your own details, and compare what you get with the example result.
Last updated
Works in Claude and ChatGPT
The prompt
I want to run simple tests on [WHAT YOU WANT TO TEST]. What it looks like now: [CURRENT VERSION] The result we get now: [CURRENT RESULT] How many people see it in a typical week: [WEEKLY AUDIENCE] What I suspect is holding it back: [YOUR HUNCHES] Propose six tests. Write each as a hypothesis: because we believe this about our customers, changing this one thing will improve this measure. One change per test. For each, give the exact two versions, the one number that decides it, and what we will do if it wins, loses or makes no difference. Rank them, with the biggest likely effect first. Be honest about size: given my weekly audience, say roughly how long a test would need to run before a difference means anything, and tell me if some tests are too small to be worth running. Finish with a simple log for recording results.
What to fill in
| In the prompt | What to put |
|---|---|
| [WHAT YOU WANT TO TEST] | An email, an advert, a landing page or a form |
| [CURRENT VERSION] | The headline, picture, offer and button as they are now |
| [CURRENT RESULT] | Today's numbers, such as bookings for every hundred visitors |
| [WEEKLY AUDIENCE] | How many people see it each week |
| [YOUR HUNCHES] | What you think puts people off |
Example result
What a good run gives, for a carpet cleaning service. Your result will differ with your details.
Test 1. Because we believe visitors leave when they cannot see a price, showing the rate beside the booking button will raise bookings. Version A: "Get a quote". Version B: "From 12 rupees a square foot. Book now." Deciding number: bookings out of every 100 visitors. If it wins: keep it, and test the photograph next. If no difference: price is not the obstacle. Size: at 300 visitors a week, allow about a month. Too small to test: the colour of the button.
Tip
Write down what you will do for each outcome before the test starts, so the result cannot be argued away afterwards.
More prompts for research and planning
- Prompt for a four-week content calendarBuild a four-week content calendar around a few repeating themes, sized to what your team can actually produce.→
- Prompt for the questions your audience is askingSort the questions you really receive by buying stage, add likely ones marked as guesses, and pick ten to answer first.→
- Prompt for a competitor messaging reviewCompare how you and your competitors present yourselves, using only text you paste, and find ground nobody has claimed.→
- Prompt to repurpose one piece of content into manyBreak one substantial piece into its separate ideas and turn each into a new piece fitted to its channel.→
- Prompt for campaign ideas from one customer insightTest a customer insight as a tension, then get six different campaign ideas with the smallest version of each to try.→
- Prompt for customer research interview questionsWrite a discussion guide that follows the story of one purchase, with questions about what people did and probes for short answers.→
- All prompts in Research and planning→
Let Wonib do it in your brand
The same work, already in your brand
Wonib learns your brand once, then writes and makes the piece in your colours and voice, and checks it before you see it. Start free with 1,000 credits, no card needed.
Last updated