Skip to content
For Claude, ChatGPT and more

A/B testing on autopilot, run by your AI agent.

Your AI agent finds what to test, designs the variants and checks them on every device. Autopilot launches the ones that pass your guardrails, and harmful tests pause themselves. You ship the winners, or approve every test yourself: it's a switch.

Start for €0: you set it up yourself, with your agent's help. Connect your agent: your agent does it all and sends you a claim link.

€0 until you launch your first test · plans from €9/month

One script tag · 14 KB · no visitor tracking · any website

Works withClaudeChatGPTCursorGeminiAny MCP agent
Animation: your agent installs the Duelpilot script and verifies it, the Analyst puts three ideas on the board and moves the best one to Design, the Designer's variant replaces the original hero of a store and passes on desktop, tablet and phone, autopilot checks the guardrails and launches the test, visitors split between A and B until the interval says Variant B wins, and a person ships it. Illustrative: a fictional store on autopilot, from one prompt to a shipped winner. The only click is yours: ship.

Give this to your agent

$ Set up Duelpilot A/B testing on my website and prepare our first test. Instructions: https://duelpilot.com/agents.md

It connects, installs the script (WordPress, Shopify, Webflow, Wix, Next.js and more), verifies it in real browsers and comes back with ideas.

How it works

From one prompt to a winning test.

Your agent does the work, autopilot launches what passes your guardrails, and you make the call that matters.

01Connect your AI

verify_install

script in <head>, synchronous

Chrome desktop · runtime 2.0

iPhone Safari · runtime 2.0

Paste one prompt into Claude, ChatGPT or Cursor. Your agent installs the script and proves it works on phone and desktop.

02It finds what to test

Idea · ICE 6.3

Illustrated icons for the 12 accommodation types

2 types share a colour2.5 tiles visible on phones

The Analyst reads your GA4 through Duelpilot and the pages themselves, and puts evidence-backed ideas on your board.

03It designs and checks

#1 Illustrated iconsphone · desktop ✓
#2 Postcard tilesphone · desktop ✓
#3 Clearer countsphone · desktop ✓

The Designer makes 2–4 genuinely different variants in your brand, with a preview link each, checked on every device.

04It launches, you ship

Day 14 · conclusive

#1 Illustrated icons wins

Original
#1

Illustrative.

Autopilot launches the variant when every guardrail passes (or you approve it yourself). Harmful tests pause themselves; the Monitor tells you, in plain words, when there is a winner to ship.

Autopilot · Pro and up

Switch it on. The loop runs itself.

Your agent finds what to test and designs the variants. Autopilot launches the ones that pass every guardrail you set, harmful tests pause themselves, and results conclude on their own. You decide what ships. Prefer to approve every test? Leave it off: it's a switch per site.

  1. 1
    Your agent

    Finds what to test

    The Analyst reads your analytics and your pages every week and puts ideas on the board, each with evidence.

  2. 2
    Your agent

    Designs and checks

    It moves the best idea to Design (as many a week as you allow) and makes 2–4 variants with previews; Duelpilot checks each on phone, tablet and desktop and lints the copy.

  3. 3
    Autopilot

    Launches the safe ones

    The agent's top-ranked variant goes live against the original, only if every guardrail passes. If one fails, it waits for you, with the reason.

  4. 4
    Duelpilot

    Watches and decides

    Harm, a broken split or a broken variant pauses the test on its own. A conclusive result closes the test at a full week.

  5. 5
    You

    Ship the winner

    You decide what ships: the winner serves every visitor right away, until your developer builds it in.

Autopilot is ON

morrowandpine.com · policy v3

e7 · rank 1 of 3 proposals

Variant B · reviews next to the button

  • Change typestext · move · css
  • Page/products/* · not excluded
  • Traffic50% · 1 of 1 autopilot slot
  • Copy lintno new claims
  • Site checkphone · tablet · desktop
  • Goal and scriptactive · seen 4 min ago

Every guardrail passed · launched by autopilot

e8 · rank 1 of 2 proposals

Free-delivery note on the cart

Waits for you

Targets /cart, an excluded page, so it stays in Review with that reason. You decide.

Illustrative: a fictional store. Your guardrails live in Settings → Autopilot.

You set the guardrails

  • Which changes. Copy, styling, layout and images. Never custom JavaScript; redirect tests only if you allow them.
  • Which pages. Never checkout, cart, payment, account, login or booking pages, or the sensitive pages in your site profile.
  • How much traffic. At most 50% of visitors per test and one autopilot test live at a time, by default; you can allow up to three.
  • What it may say. No copy-lint errors: no prices or percentages that aren't on your page, no unproven superlatives or urgency, nothing on your forbidden-claims list.
  • Proof it works. A passing site check on every device the test targets, an active goal, and your script reporting in the last 24 hours.

Harm stops itself

Duelpilot checks every live test around the clock. Significant harm, a traffic split that doesn't add up or a variant that breaks for many visitors pauses the test on its own, autopilot on or off. A conclusive result closes the test at a full week, so nobody has to watch it.

Your call, always

  • Autopilot never ships a winner, resumes a paused test or changes a live one.
  • An AI agent never launches a test on its own: a person approves each launch, on the board or in the chat with their agent, unless an owner switched autopilot on. Duelpilot's server enforces both.
  • Every automatic launch is logged with the policy version it used.

Autopilot is on Pro and up · off until an owner switches it on

Live on this website

This website is being tested. Watch.

Duelpilot runs on duelpilot.com with the same script our customers install. See what's being tested, open either version yourself, and follow what the statistics say.

See it running →Real counts. Nothing simulated.

Your agent can start today.

Connect it with one prompt and set up your first test for €0: you pay when you launch it. Plans from €9/month.

€0 until you launch your first test

Real variants, real site

Not mock-ups. What the agents made.

For glampings.co, the agents drew new icons, redesigned a section and built a phones-only test, all in the site's own colours and fonts.

glampings.co · variant
Variant with illustrated icons per accommodation type
The Designer agent drew an illustrated icon per accommodation type, in each tile's own colours. Shown in a Duelpilot preview: never counted.

Want variants like these on your site?

Your Designer agent makes 2–4 per idea, each with a preview and a check on phone, tablet and desktop.

Set up your first test for €0

€0 until you launch your first test

Three agents. Your Claude.

A CRO team that runs on the AI you already have.

They run as routines in your own Claude, with instructions Duelpilot keeps up to date. No plugin, no extra AI bill.

Analyst

Weekly

Reads your GA4 through Duelpilot, finds pages and steps where people drop off, and proposes ideas with at least two pieces of evidence each.

Get your first ideas

Designer

When an idea moves to Design

Designs 2–4 variants that differ in approach, from a sharper headline to a bold redesign with new icons, and checks each one on phone, tablet and desktop.

See your first variants

Monitor

Daily

Watches every live test, pauses anything that hurts, and posts a plain update where you work: the board, Slack, email, through your own agent.

Get your first report

€0 until you launch your first test · no extra AI bill

Built like the enterprise tools

Works on any website.

Duelpilot changes the page your visitors see, so it doesn't care what it's built with. Tested in Chrome, Safari and Firefox on phone, tablet and desktop.

WordPressShopifyWebflowWixSquarespaceFramerHubSpotNext.jsNuxtSvelteKitAstroRemixAngularRails + TurboLaravelGoogle Tag Manager

No flicker

Only the elements under test wait, for milliseconds, never the page. Hydration-safe on React, Vue, Svelte and Astro.

Previews anyone can open

A link per variant, a small tab on phones, and Point & comment so feedback lands on the card.

Statistics you can trust

Always-valid sequential tests, sample-ratio checks and corrections for several variants. It says NOT CONCLUSIVE until it is.

Your analytics, automatically

GA4 dimensions set up for you and results pulled back, next to ours. Also Adobe, PostHog, Mixpanel, Amplitude, Segment, Hotjar, Plausible, datafa.st.

You set the guardrails

Autopilot (Pro and up) launches only what passes your rules: pages, change types, traffic, copy and site checks. Otherwise a person approves each launch, on the board or in the chat with their agent.

Privacy by design

No visitor profiles, no session replays, no fingerprinting. Google Consent Mode, TCF and a cookieless mode.

One script tag. Any website.

Your agent installs it, verifies it in real browsers and puts the first ideas on your board.

€0 until you launch your first test

We count. We don't track.

A scoreboard, not another analytics tool.

Most testing tools store every click of every visitor. Duelpilot keeps a tally per variant, per day and per device type. The detail goes to the analytics you already have, tagged with the variant. That is why it's light on your site, light on privacy, and cheap to run.

Start for €0

€0 until you launch your first test

14 KB

the whole script, from the edge

≤ 2

tiny pings per visitor per test

0

visitor profiles stored

Pricing

One price per site. That's it.

No visitor meters, no AI credits. Your agents run in your own Claude, and we don't store visitor data, so there is nothing to pass on.

Every plan: €0 until you launch your first test. Plans from €9/month.

€0 until you launch your first test. Setting up is free: connect your agent, install, ideas, designs and previews. No launch within 30 days? We cancel the plan automatically and you pay nothing.

Starter

One site, one test at a time, run by your own agent.

€7/ month

billed yearly · €84 / year

Start for €0

€0 until you launch your first test

  • 1 site, 1 live test at a time
  • Your agent installs and verifies the script
  • Ideas, designs and previews by your agents
  • Site checks on phone, tablet and desktop
  • GA4 results next to ours
  • You approve every launch (autopilot is on Pro and up)

Fair use: 100k pageviews per site per month.

Most teams start here

Pro

A/B testing on autopilot, inside guardrails you set.

€39/ month

billed yearly · €468 / year

Start Pro for €0

€0 until you launch your first test

  • Autopilot: tests that pass your guardrails launch without you
  • Unlimited tests and design rounds
  • Exclusion groups, results per device
  • Custom JavaScript with owner sign-off
  • Staging key, reset results, full audit log

Fair use: 2M pageviews per site per month.

Team

For several brands, domains or clients.

€119/ month

billed yearly · €1,428 / year

Start Team for €0

€0 until you launch your first test

  • Everything in Pro, for 5 sites
  • Unlimited team members and roles
  • Embed the board in your own HQ (SSO)
  • Priority support
  • Extra sites €29 / month each

Fair use: 10M pageviews per site per month.

Code

Tests in your own code too: server-side, zero flicker.

€159/ month

billed yearly · €1,908 / year

Start Code for €0

€0 until you launch your first test

  • Everything in Team, for 5 sites
  • Code tests: your agent builds them in your repo, behind Duelpilot flags
  • Server-side flags: booking flows, pricing logic, search ranking
  • Duelpilot never needs access to your code
  • Extra sites €29 / month each

Fair use: 10M pageviews per site per month.

Enterprise

For organisations with security, legal and procurement requirements.

  • Everything in Code, for as many sites as you need
  • Dedicated region, script on your own domain
  • SLA and a signed DPA
  • Invoice billing
Talk to us

Prices excl. VAT. Cancel any time; your tests stop and your site goes back to the original.

Fair questions, straight answers

Before you connect your agent.

Still unsure?

Run one real test on your own site, then decide.

Set up your first test for €0

€0 until you launch your first test

What do I need?

A website you can add one script to (or an agent that can, through your CMS or code) and an AI agent that speaks MCP: Claude, ChatGPT, Cursor or similar. Your agent connects to Duelpilot with one prompt.

Can it run my A/B tests on autopilot?

Yes, on Pro and up. An owner switches autopilot on per site, and from then on the loop runs without you: your agent finds ideas, moves the best ones to Design (as many a week as you allow) and designs the variants, Duelpilot checks them on every device, and autopilot launches the agent's top-ranked variant against the original when every guardrail passes. Harmful tests pause themselves and conclusive results close on their own; shipping a winner stays your call. Without autopilot, your agent never launches a test without your explicit yes, on the board or in the chat with it; autopilot is your setting, enforced by Duelpilot's server. Code tests are always approved by a person. Prefer to approve every test yourself? Leave it off; it stays off until an owner turns it on.

What guardrails does autopilot follow?

The ones you set in Settings → Autopilot, and a launch must pass all of them: which kinds of change (copy, styling, layout and images by default; never custom JavaScript), which pages (never checkout, cart, payment, account, login or booking by default, plus the sensitive pages in your site profile), how much traffic (by default at most 50% per test and one autopilot test at a time), clean copy (no prices or claims that aren't on your page), a passing site check on every device the test targets, an active goal and a script that reported in the last 24 hours. If one fails, the test waits in Review with the reason. Every change to the guardrails is a new policy version, and every automatic launch records the version it used.

Does it slow my site down?

The script is 14 KB from a global edge network and runs before the page paints. Only the elements under test wait, usually for a few milliseconds; the page itself is never hidden. In our tests it costs about one network round trip.

Will it work on my site?

If visitors see it in a browser, almost certainly. Duelpilot works on the finished page, so the framework doesn't matter: React and Next.js, Vue and Nuxt, Svelte, Astro, Turbo, single-page apps and sites with strict security headers are covered by our browser tests in Chrome, Safari and Firefox, and WordPress, Shopify, Webflow or Wix pages are ordinary HTML to it. What it can't change in the browser: content inside other companies' widgets (embedded iframes) and anything your server decides, like prices or search results. That is what code tests are for.

How does the agent know which element to change?

It finds them itself: the Analyst reads your analytics and the page structure and puts ideas on the board with evidence; the Designer inspects the page and checks every selector before it submits. You can point too: open 'Pick on my site' from your board and click any element on your live site to turn it into an idea, or use Point & comment on a preview to say 'make this bigger'. The pick link only works for signed-in members, expires after 30 minutes and only creates ideas, so nobody can change your site with it.

Do buttons, date pickers and forms keep working?

Yes for what most tests change: text, colours, styling, images, hiding and reordering keep the original element, with everything it does. New blocks a variant inserts are plain HTML, so a new button is a link, never a new script. Before anything can go live, the site check clicks the buttons, menus, date pickers and form fields in and around the change on phone, tablet and desktop, without sending anything, and compares the variant with the original: if something stopped responding, the check fails and you see it before you approve.

What about bigger changes, like a new booking flow?

Changes in how something works (a new flow, a different date picker, fewer form fields, pricing or search logic) belong in your code. On the Code plan your own agent builds each variant behind a Duelpilot flag in your repository and opens the pull requests with its own GitHub access; Duelpilot never needs access to your code. Duelpilot splits the traffic, checks every variant's preview, measures, and when a winner ships your agent opens the clean-up pull request.

What about cookies and GDPR?

Duelpilot stores no visitor profiles: only daily totals per variant, and a random visitor id stays in the visitor's own browser. It follows your consent setup (Google Consent Mode, TCF or your cookie banner) and has a cookieless mode. Whether it may run before consent is your call; your agent asks you.

Do you sign a DPA?

Yes, for every customer. Our Data Processing Agreement (GDPR art. 28) is part of the terms you accept when you choose a plan, so it is in place without paperwork; need a signed copy for your records? We send you the same text as a signed PDF. It lists what we process for you, our sub-processors, and that your visitors' data stays in the EU. Read it at /dpa.

Read the Data Processing Agreement
Is there a free plan?

No, but every plan is €0 until you launch your first test. You add payment details at signup and everything before the launch is free: connecting your agent, installing the script, your brand kit, ideas, designs, previews and site checks. Your plan starts when you launch your first test, and that first payment is taken then. No launch within 30 days? We cancel the plan automatically and you pay nothing. Plans from €9/month, flat per site.

Why do you ask for payment details at signup?

So your first test can go live the moment you approve it: €0 until you launch your first test, and launching it starts your plan in the same step. If your bank asks to confirm the payment, the test waits in Review until you have, then launches with one click. Card, SEPA Direct Debit and iDEAL, depending on your country.

What does the AI cost?

Nothing extra from us. The agents run in your own Claude (or other agent) subscription; Duelpilot never resells AI.

Which analytics do you support?

GA4: connect it in one click (Settings → Analytics). Duelpilot registers its custom dimensions in your property and shows GA4's view of every test next to its own counts, and your agent can read GA4 reports through Duelpilot. Variant events also go to Google Tag Manager, Adobe, PostHog, Mixpanel, Amplitude, Segment, Hotjar, Clarity, Plausible, Fathom, Matomo and datafa.st automatically; anything else gets them through a one-line bridge.

Your first test can be live this week.

Connect your agent and approve your first test. Then switch on autopilot and let the loop run inside your guardrails.

€0 until you launch your first test. Setting up is free: connect your agent, install, ideas, designs and previews. No launch within 30 days? We cancel the plan automatically and you pay nothing.