Skip to content
Product · Playbook

Write the hypothesis, the metric, and the stopping rule before launch.

A plan for one A/B test that fixes what you are testing and how you will decide, before anyone sees a result.

IncludedNative Claude Skill + plain .md for ChatGPT, Gemini & every LLM
100 Playbooks in the catalogDelivered as Claude Skill + plain .mdReviewed Oct 22, 2026Instruction-only · verifiable

A weak test usually fails before it starts. The hypothesis is a hunch, the metric is picked after the results arrive, and the test ends when the graph looks good. Stopping on a good day proves very little.

A/B Test Plan writes the hypothesis with its reason, one primary metric defined exactly, guardrails, and the control and variant. It leaves the sample size and stopping rule as items to confirm with whoever owns your analytics, records what they say, and pre-writes what you will do for each outcome.

What you get

What it produces.

  • →A hypothesis in if, then, because form
  • →One primary metric and one to three guardrails, defined exactly
  • →A stopping rule slot filled with what your analytics owner confirms
  • →A decision for each outcome, including no difference
Who it's for
  • ✦Growth leads and marketers planning a test
  • ✦Product managers about to ship a change to a share of users
  • ✦Anyone who wants a test plan others can check afterward
What's included
  • →ab-test-plan.zip: a native Claude Skill. Upload to Settings → Skills in Claude, or unzip to ~/.claude/skills/
  • →ab-test-plan-skill.md: plain markdown. Works in ChatGPT Custom GPTs, Gemini, Claude Projects, agent system prompts, anywhere
  • →A README inside the zip with install steps for Claude, ChatGPT, and Gemini
Delivery
Emailed within 60 seconds of payment
Compatibility
Claude, ChatGPT, Gemini, any LLM
License
One-time purchase, yours to keep
Sample output

What it does with a real question.

The brief

I'm the growth lead at a small project management SaaS, and we want to test moving our annual billing toggle above the plan cards on the pricing page. Right now about 3.1% of pricing page visitors start a trial, and I think the toggle will lift that. We can run it for up to four weeks starting March 3.

Fix the stopping rule before March 3. Your hypothesis is workable: if the annual toggle moves above the plan cards, trial starts per pricing page visitor will rise from the current 3.1%, because visitors see the discount before they size up the plans. That reasoning is a belief, so record it as one. Make trial starts per unique pricing page visitor the single deciding metric, and watch two guardrails: trial-to-paid conversion and refund requests. A toggle that lures people into annual trials who then cancel would look like a win on the primary number and lose you money.

The weak point is the four-week window. Four weeks is a ceiling, not a plan. A 3.1% baseline means each variant needs a lot of visitors to show a modest lift, and I will not guess at that number. Ask whoever owns your analytics or testing tool to confirm the sample size, the fixed run length, and the significance rule, then write them into the plan exactly as given. If your traffic cannot reach that sample inside four weeks, you will know before launch, which is far better than discovering it on day 20.

Also agree now that nobody checks results early or stops on a good day. Peeking and stopping when the graph looks favorable is the most common way a test like this misleads you.

Next move: send your analytics owner the 3.1% baseline, your weekly pricing page traffic, and the four-week limit, and ask for the run length and stopping rule in writing this week.

The full Playbook runs the whole procedure and writes the complete deliverable, not this excerpt.

Written by claude-sonnet-5-5 with this Playbook loaded, 2026-10-11. Checked automatically for length and for copied framework text.

One-time purchase
$12.99
Yours to keep
No subscription · free updates within v1.x

Works as a Claude Skill (.zip) and as a plain .md for ChatGPT, Gemini & every LLM
Delivered to your email within 60 seconds · Secured by Stripe
Price includes UK VAT where applicable

Questions

Before you buy.

How is this different from just asking Claude directly?

Asked cold, Claude may give you sample size figures and a significance threshold as if they applied to your traffic. This plans the test, leaves those numbers for your analytics owner to confirm, and makes sure the metric and decisions are set before launch.

How is this different from a Tool?

Playbooks are systematically produced to a fixed template and priced accordingly ($12.99). Tools are hand-built, one at a time, and go deeper ($39 to $199). A Playbook gives you a reliable method for routine work; a Tool is for when the judgment matters more than the format.

Will this work with ChatGPT, Gemini, and other LLMs?

Yes. You get a native Claude Skill (.zip) that auto-invokes in Claude, plus a plain .md you can paste into a ChatGPT Custom GPT, a Gemini system prompt, or any LLM. The procedure travels with the file.

Can I share this with my team?

It's licensed for your own personal and professional use. The output it produces is yours to circulate. To put the Playbook itself in the hands of a whole team, email hello@authority.md about team licensing.

What if it doesn't work for me?

If the file hasn't been delivered, or arrives defective or empty, we refund in full: see our refund policy.

More Product Playbooks

Other Playbooks.