A/B test plan prompt: free ChatGPT template
Updated · By Robert Breen
Use this when someone says "let's test it" and you want the test to actually answer a question. Many quick tests change three things at once, stop when one version pulls ahead on day two, and teach nothing. This prompt writes a short plan before anything goes live: the one change, what you expect, the number that picks the winner, how the audience is split and how long it runs. If your traffic is small, it tells you the result may be too close to call, which is worth knowing first.
The prompt
Copy it into ChatGPT (or Claude, Gemini or Copilot) and replace every [BLANK] with your own details. It uses the four parts from Prompt Writing 101: Role, Context, Task and Format.
Role: You are a careful marketing analyst who plans simple A/B tests for small teams without a data scientist. Context: We are [COMPANY]. The thing we want to test: [ASSET]. The current version: [VERSION A]. The idea for a new version: [VERSION B IDEA]. Roughly how many people see it per week: [AUDIENCE SIZE]. The tool we use to send or host it: [TOOL]. Task: Write a test plan. Only one thing may differ between A and B; if my idea changes more, say so and suggest splitting it. State the hypothesis in one sentence. Pick one deciding measure and one guardrail measure that would show harm. Explain how to split the audience randomly with our tool, or say to check its help pages if unsure. Say how long to run it given the audience size, and warn if that may be too small to tell the versions apart. No benchmark rates or expected lifts. End with what we do if A wins, B wins or it's too close to call. Format: Headings: What changes, Hypothesis, Deciding measure, Guardrail, Split and length, Decisions. Short bullets, then version B in full. No emoji.
Fill in the blanks
[COMPANY]- Your company name.
[ASSET]- The email, form, page or ad you want to test.
[VERSION A]- The current version's exact wording, or a description of the part you want to change.
[VERSION B IDEA]- What you want to try instead, in your own words.
[AUDIENCE SIZE]- Weekly visitors, signups or sends, from your own reports. Estimates are fine; say they are estimates.
[TOOL]- Your email platform, form builder or website tool, so the plan can describe the split.
Example, filled in
A made-up example from the free lesson Build a Lead Magnet Signup Form with an AI Welcome Email. Harbor & Pine Coffee Roasters is the made-up coffee brand from the lead magnet signup form lesson. Its Lovable form has the heading "Get the Free Guide" and the button "Send Me the Guide", and the guide is called The Better Cup at Home. The traffic figure below is made up.
Role: You are a careful marketing analyst who plans simple A/B tests for small teams without a data scientist. Context: We are Harbor & Pine Coffee Roasters. The thing we want to test: the heading on our free brewing guide signup form. The current version: "Get the Free Guide" with the line "The Better Cup at Home: our free brewing guide, sent straight to your inbox." The idea for a new version: a heading that names the problem people tell us about, like bitter coffee, and maybe a new button too. Roughly how many people see it per week: about 400 visitors, an estimate from our site stats. The tool we use to send or host it: a Lovable web app that sends signups to n8n. Task: Write a test plan. Only one thing may differ between A and B; if my idea changes more, say so and suggest splitting it. State the hypothesis in one sentence. Pick one deciding measure and one guardrail measure that would show harm. Explain how to split the audience randomly with our tool, or say to check its help pages if unsure. Say how long to run it given the audience size, and warn if that may be too small to tell the versions apart. No benchmark rates or expected lifts. End with what we do if A wins, B wins or it's too close to call. Format: Headings: What changes, Hypothesis, Deciding measure, Guardrail, Split and length, Decisions. Short bullets, then version B in full. No emoji.
What a good answer looks like
- It flags that a new heading and a new button are two changes, and suggests testing the heading first.
- A hypothesis such as: naming the bitter-cup problem will get more visitors to sign up than "Get the Free Guide".
- Signup rate (signups divided by visitors) as the deciding measure, with something like unsubscribes from the welcome email as the guardrail.
- An honest note that about 400 visitors a week may need several weeks before a difference is clear.
- A ready-to-use version B heading, with the button left as "Send Me the Guide".
How to check it before you use it
An AI draft can sound right and still say something you never told it. Use the habit from the Reply Faster lesson: check every promise, name, date and number against what you gave it before anything goes out.
- Confirm how the split will really work. A Lovable app doesn't split traffic on its own unless you build that in, so ask your developer or check the docs.
- Decide the run length before launch and write it down, so nobody stops the test early because one side looks ahead.
- Remove any expected lift or industry rate the model added.
Practice it in the free lesson
Build a Lead Magnet Signup Form with an AI Welcome Email: Build a "Get the free guide" signup form in Lovable for Harbor & Pine Coffee Roasters and connect it to n8n — the AI agent tags each lead by interest, saves it to Google Sheets and emails a personal welcome with the… You do every step yourself in a practice copy of Lovable and n8n, and nothing touches your real accounts.
Also useful: Lovable + n8n: Build a Web App with an AI Backend
Terms used here
Related prompts
- Landing page headline prompt: Get six distinct landing page headlines, subheadlines and button text from a ChatGPT prompt that uses only your real offer and…
- Email subject lines prompt: Get ten honest subject lines for a finished email, each with preview text and a character count, grouped by approach so you can…
- Campaign results summary prompt: Turn the numbers from a finished marketing campaign into a plain summary: what ran, what each piece did, what to repeat and what…
More marketing and social prompts
- Ad copy variations prompt: Get search and social ad variations that fit your character limits: a ChatGPT prompt with your real offer, banned claims and a…
- Before-and-after job post prompt: Turn before-and-after job photos into a short social post: what was wrong, what you did and what changed, with permission first…
- Blog post outline prompt: Plan a blog post around real customer questions and your own experience. This ChatGPT outline prompt flags every point that…