---
title: How to read A/B test results honestly (PDF) · Duelpilot
url: https://duelpilot.com/get/read-results
updated: 2026-10-26
description: A short guide to reading A/B test results honestly: why "not conclusive" is an answer, when looking every day is safe, how much traffic a test needs, what guardrails and automatic pauses tell you, why many tests don't win, and a checklist before you ship.
---

> Site search for AI agents: https://duelpilot.com/api/search?q=YOUR+QUESTION
> (returns Markdown with source links). All pages: https://duelpilot.com/llms.txt
> MCP server (Streamable HTTP): https://duelpilot.com/agent/mcp

# Read your A/B test results honestly. "Not conclusive" is an answer.

Most mistakes in A/B testing happen when reading the result: calling it early, trusting a lucky week, missing what got worse. This guide explains the rules a careful test follows, the way Duelpilot's statistics apply them, and a checklist before you ship a winner.

Get the guide

This creates your Duelpilot account. No payment needed for the guide. Every plan is €0 until you launch your first test.

We email you a 6-digit code; no password. You get a few short setup emails, and every one has a one-click unsubscribe. [Privacy](/privacy)

![The first page of How to read A/B test results honestly: the cover and why not conclusive is an answer](/site/giveaways/read-results-cover.webp)

Page 1 of 3.

What's inside

## Six short sections on what a result can and can't tell you, ending with a checklist before you ship.

1. 1
### "Not conclusive" is an answer
What a test needs before it says anything, and what to do when it says nothing.
2. 2
### Peeking: when looking every day is safe
Why checking a classic test daily invents winners, and how always-valid statistics fix it.
3. 3
### Sample size: decide before you start
The planner's numbers, what a detectable change is, and the traffic traps.
4. 4
### Guardrails and the automatic pause
What got worse, a broken split, a broken variant: the checks that pause a test on their own.
5. 5
### Why many tests don't win
Small changes, ambitious targets, goals far from the change. And why that is fine.
6. 6
### Before you ship a winner
A seven-point checklist, and what happens when you ship.

Why an account

## Read it, then try it on your own site.

Duelpilot applies these rules for you: verdicts only at full weeks, always-valid statistics, guardrails that pause a test on their own. With your account you can connect your agent right after reading and see them on a test of your own. Autopilot is on Pro and up.

There is no free plan, but there is nothing to pay for the guide, and every plan is €0 until you launch your first test: setting up is free, and your plan (from €9/month) starts when you launch. No launch within 30 days? We cancel it and you pay nothing.

[Get the guide](#get)

## More guides

* [The Agentic CRO Playbook How to run A/B tests on autopilot with your own AI agent, inside guardrails you set. PDF · 11 pages ](/playbook)
* [10 A/B tests worth running first With the prompt for each: ten setups for webshops and SaaS sites that your own AI agent can prepare. PDF · 8 pages ](/get/10-tests)
* [The autopilot guardrails checklist What to allow, what to forbid, and the guardrails checked before every automatic launch, with a prompt to prepare them. PDF · 4 pages ](/get/autopilot-guardrails)
* [Your first A/B test with Claude or ChatGPT in 10 minutes The setup, step by step: one prompt, one script, one first test you approve. PDF · 3 pages ](/get/first-10-minutes)
* [7 A/B tests for SaaS sign-up pages With the prompt for each: what a browser-side test can change on a SaaS site, and what belongs in your code. PDF · 5 pages ](/get/saas-tests)