A/B test hypotheses and prioritization

Turn a weak-performing page or email into well-formed test hypotheses, ranked by impact and effort, with clean success criteria.

Change note: Initial version

Prompt

You are a growth marketer designing experiments. Help me plan A/B tests for the asset below.

Asset: {{asset}}
Current performance (conversion rate, traffic, any known drop-off): {{performance}}
Goal metric: {{metric}}
Audience and traffic source: {{audience}}
Weekly traffic or sends available for testing: {{traffic}}
Known qualitative feedback (surveys, support, session notes): {{feedback}}
Current copy or description of the asset:
"""
{{current}}
"""

Deliver
1. Diagnosis: the 3 most likely reasons performance is weak, each tied to evidence in what I gave you. Say what data I would need to confirm.
2. {{num_tests}} test hypotheses in this format: "Because [observation], we believe changing [element] to [variant] will [effect] for [segment], measured by [metric]." Include the control and variant described concretely, with sample copy.
3. Prioritize with a table scoring Impact, Confidence and Effort from 1 to 5, and an overall rank. Explain the logic in a sentence.
4. For the top test: minimum detectable effect that is realistic for my traffic, rough duration, a guardrail metric, and what I should not change during the test. Tell me plainly if my traffic is too low for a reliable test and suggest an alternative approach (for example, sequential or qualitative testing).

Rules: one change per test. Do not promise lift percentages. Do not invent data.

Variables

• {{asset}}Assettext

  Default: Pricing page for a project planning app

• {{performance}}Current performancetext

  Default: 4,000 visitors per month; 1.2% start a trial; 55% leave without scrolling past the plans table

• {{metric}}Goal metrictext

  Default: Trial starts per visitor

• {{audience}}Audience and sourcetext

  Default: Agency owners from search and a monthly newsletter

• {{traffic}}Traffic for testingtext

  Default: About 1,000 visitors per week

• {{feedback}}Qualitative feedbacklong

  Default: Prospects ask if they can cancel anytime and whether pricing is per user; some say the three plans look the same.

• {{current}}Current copylong

  Default: Headline: 'Pricing'. Three plans: Starter $19, Team $29, Studio $49 (all per user per month). Button on each: 'Get started'. FAQ at the bottom of the page.

• {{num_tests}}Number of hypothesesnumber

  Default: 4

Example output

## Diagnosis
1. The headline gives no reassurance or framing, and 55% leave early (evidence: bounce rate).
2. Key doubts (per user pricing, cancellation) are answered only in a footer FAQ (evidence: prospect questions).
3. Plans look similar, so choosing is hard (evidence: feedback). Confirm with heatmaps or plan-click data.

## Hypotheses
1. Because prospects ask about cancellation, adding "14-day trial, cancel anytime" under the headline will increase trial starts. Control: "Pricing". Variant: "Simple pricing. Try it free for 14 days, cancel anytime."
2. Because buyers ask whether pricing is per user, labeling "per user per month" in bold on each plan and adding a team-cost example will reduce hesitation.
3. Because plans look alike, highlighting who each plan is for will help.
4. Because "Get started" is vague, "Start free trial" will increase clicks.

| Test | Impact | Confidence | Effort | Rank |
|---|---|---|---|---|
| 1 | 4 | 4 | 1 | 1 |
| 4 | 2 | 3 | 1 | 2 |
| 2 | 3 | 3 | 2 | 3 |
| 3 | 3 | 2 | 3 | 4 |

## Top test details
At 1,000 visitors per week and a 1.2% baseline, only a large lift would be detectable in a month; expect around 8-10 weeks for a modest change. Guardrail: trial-to-paid conversion. Do not change pricing or layout meanwhile. If that is too slow, combine tests 1 and 4 as a single "messaging" variant.

Notes

Tying each hypothesis to an observation prevents random-idea testing, and the traffic reality check stops underpowered tests. Share real analytics screenshots as text to improve the diagnosis.