Picked the winning headline in 76% of 1,000 real A/B testsSee the benchmark

Find what breaks before your customers do.

Simulated people, AI profiles drawn from population statistics to match your audience, try your site in real browsers and tell you where they got stuck and why.

3 simulated people try to in a real browser. Takes a few minutes.
ExampleSign up for the free trial on Tallyloop, 5 simulated people
  1. 01Start your free trialNobody
  2. 02About your businessNobody, stuck on the company website field
  3. 03Your remindersNobody
  4. 04WelcomeDaniel, Joan, Joshua, Sandy, Rafael
All 5 finished.Top issue · 5 of 5URL field validation rejects valid URLs without clear error guidanceSee a full report
Tallyloop is a made-up product. The run is real: recorded on 26 Sept 2026 and replayed from its record.
claude mcp add --transport http mimiq https://mcp.mimiqai.com/mcp
Works withClaude CodeCodexCursorVS CodeWindsurfClaude Desktop

What you get back. What broke, who it happened to, what they said, and what to fix first.

How far each person got

Where each person stopped. All 5 finished Tallyloop's sign-up, in 5 to 9 steps.

Open a sample report
Top issueURL field validation rejects valid URLs without clear error guidance5 of 5 raised itSuggested fixFix the URL validation logic to accept standard URL formats (with and without https://, with and without www).
Top issue

What to fix first. The problem most people hit, who hit it, and a fix you can hand to whoever builds the page.

Joan Jones43, CFO · simulated
  1. 09Said they were doneWelcome“We're in!”
  2. 08Clicked “Start my free trial”Your reminders“Good, 'Firm' is selected and the preview looks right, invoice number, amount, due date, clear call to action.”
  3. 07Clicked “firm”Your reminders“Step 3 of 3, almost there!”
  4. 06Clicked “Continue”About your business“It's still showing Invalid URL even with [their website]”
In their words

Every step, and why. Each person says what they see and think as they go, so a drop-off comes with its reason.

Web page · 25 simulated peopleOnly 9 of 25 stayed to read.
Stayed or better9 of 2522% to 52% is likely
Would sign up0 of 250% to 10% is likely
First impression

A landing page, counted. Who stayed and who left in the first seconds, with the range real visitors would likely fall in.

How page tests work

From the two Tallyloop runs at the top of the page. What anyone typed is never kept.

What it tests. Anything a customer will see, shown the way they would really meet it.

How far to trust it. Mimiq measures that in public, and says where it is not there yet.

  • Mimiq's forecast75.7%
  • Best rule of thumb (the headline written first)61.3%
  • Asking an AI directly53.1%
  • A coin50.0%

On 1,000 held-out headline A/B tests: Mimiq's forecast 75.7%, Best rule of thumb (the headline written first) 61.3%, Asking an AI directly 53.1%, A coin 50.0%.

Measured in public

76% on headlines. Mimiq's forecast picked the headline more real readers clicked, on 1,000 A/B tests held out from its development. Ranking emails, ads and page versions is the work in progress.

The benchmark
PlanPublic before it runs
RunOn tests it never saw
ResultPublic, whichever way
Pre-registered

The plan first, the result after. Each benchmark is public before it runs and after, whichever way it goes.

Every plan and result
What Mimiq keeps01Filled in 3 fieldsStart your free trial
Not kept

What they type stays in the browser. A flow keeps each step in words. The values typed into your site are never stored.

Trust and data
Sandy Hudson63, Small Business Owner, HollywoodSimulated
A profile drawn from population statistics, played by an AI.
Always labelled

Simulated, and always called that. Simulated people are a fast first pass before real research, not a replacement for it.

Payments go through Stripe; card numbers never touch Mimiq's servers.

From your coding agent. Your agent built the page; let it show it to a crowd before you review the diff.

Terminal
claude mcp add --transport http mimiq \  https://mcp.mimiqai.com/mcp \  --header "Authorization: Bearer $MIMIQ_API_KEY"

Leave the key out for one free test. For more, create a key in Settings; it starts with mq_sk_.

Then ask it
Its eight tools
test_flow
A sign-up, onboarding or checkout flow in a real browser, step by step.
test_page
One page: what confused people, what they doubted, what would help.
compare_urls
Two pages, such as a preview and production, on the same people.
compare_copy
Two versions of copy or an email, on the same people.
test_copy
A headline, a tagline, a subject line or a call to action.
test_text
Positioning, feature descriptions, error messages, instructions.
test_component
A UI component, from its HTML or a description.
ask_audience
A multiple-choice question, answered with reasons.

Works with Claude Code, Codex, Cursor, VS Code, Windsurf and Claude Desktop. The server is open source (MIT). Set it up

Pricing. Pay per person tested. No seats, no subscription.

First testFreeWith a free account (Google sign-in, no card), which includes 100 credits.
150 credits$96 page tests of 25, or 30 people in flows
500 credits$2920 page tests of 25, or 100 people in flows
2,000 credits$9980 page tests of 25, or 400 people in flows
5,000 credits$199200 page tests of 25, or 1,000 people in flows

One credit is one person in a page, ad, email or copy test; a person in a flow uses 5, since each gets a real browser for minutes. Packs are one-time and never expire. Full pricing

Questions.

No. Every person in Mimiq is simulated: a profile drawn from population statistics (age, work, place, habits) and played by an AI. They are fast, cheap and candid, and they are not your customers. Use them to find problems before real people do.

You give a URL and a task, like signing up for the free trial. Each person opens the site in their own real browser and clicks, scrolls and types for up to 15 steps, saying what they think at each one. You see how far each got, where they stopped and why. A flow costs 5 credits a person.

It depends on what you test. On headlines, Mimiq's pre-registered forecast picked the real winner 76% of the time on 1,000 held-out tests. Ranking emails, ads and page versions is not there yet, and a flow shows where people get stuck rather than predicting your conversion rate.

No. A flow keeps each step in words (clicked this button, filled in two fields), never the values typed and never an address's query string.

No. People in a flow cannot sign in to an existing account, so they stop at a login page. Test the public part: the landing page, sign-up, pricing and checkout.

Your first test is free: sign in with Google and it starts right away. A free account includes 100 people, no card. Then one-time packs from $9 for 150 credits, which never expire. A person in a page, ad or copy test is one credit; a person in a flow is 5.

Mimiq is made by Victor Gulchenko. Something in a report looks wrong, or you need a test it does not run yet? victor@mimiqai.com

Try it on your site.

Your first test is free: sign in with Google and it starts right away. A free account includes 100 people, no card.

3 simulated people try to in a real browser. Takes a few minutes.