TheySaid

npx add-mcp 'https://ai.theysaid.io/api/mcp'

Runs inside the tools you already use, nothing new for your team to learn

Validation Debt is the cost of decisions you shipped without asking a customer

Every month your team ships more decisions — designs, tickets, PRs. The ability to check them with customers hasn't moved in a decade: schedule interviews, recruit a panel, wait two weeks. The gap fills with guesses, and guesses compound.

Validation Debt

Shipped

Validated

Jan Dec

Validation can't be a project anymore. It has to be infrastructure — running every time you ship.

• 01 Connected once, by one
  person — in 2 minutes
• 02 Nobody on your team logs
  into anything new
• 03 Nobody has to remember to
  run it — tickets trigger it

Customer validation where you build, not another app

Ask for feedback without leaving the canvas. When a ticket hits “Design Done,” customer validation runs automatically, attaching the evidence to the ticket—no one needs to remember to ask. TheySaid runs the test, returns the findings, and gives you the prompt to fix them.

Test design with our target users

Would you like me to fix the design problems identified?

Design Done

Example ticket

Customer validation

4 min ago

TheySaid showed your feature to 12 users, and found 1 issue that needs fixing before you ship.

See it. Hear it. Fix it.

• Watch

  A tester reviews your design or product, recorded like a real session.

• Hear

  Every hesitation, complaint and “wait, where is…” — transcribed, timestamped, clipped.

• Fix

  TheySaid shows what broke and gives a prompt to paste into Claude or your editor.

• Your own customers, recruited testers, or third-party panels

  Humans use your design, share screens, and think out loud. Use them for confidence before a big bet.

• Synthetic users trained on your customer data — not generic personas

  They follow tasks, share screens, and think aloud quickly. Use them to narrow options and set early design and product direction.

• Did we build what we said we'd build?

  An expert critique of the experience against your spec and best practice, before anyone else sees it.

Synthetic you can trust

Not generic personas. Testers built from real sessions and tuned to your customers.

1. Built on real behavior

   Every synthetic tester starts from a base model trained on 100,000+ hours of real usability sessions — how people actually hesitate, misread and give up.

2. Tuned to your customers

   Connect the data you choose — call recordings, surveys, support tickets, CRM — and testers learn how your buyers and users think and talk. No customer data yet? Start with a pre-built Audience for your market and refine it as you learn.

3. Checked against real people

   We run the same tests with synthetic and human testers on the same designs, and publish how often they agree — so you know how far to trust the result.

Synthetic testers find nearly as much as people do, in minutes, and catch issues people miss. For the big bets, run both. These results come from over 10 studies run both ways on the same designs.

Read the methodology

79% found by synthetic & human users

Run the same test across roles and compare how buyers, users and execs react to the same design — in one report.

Buyer End user Exec

Results in minutes — choose Fast Mode for a quick read or Full Mode for a complete session. Screen and voice are recorded with think-aloud, just like a real session, and you can rerun after every change with no extra effort.

Julia

Why is shipping only shown after I add to cart? I almost left I needed to know the total first.

Use your own customers, testers we recruit, or third-party panels — screened, scheduled and paid for you, with results in hours. All through the same test builder, same audiences, same report.

John 98%

Jean

Carl

22–28 yr

Junior designers at SaaS startups

22–28 yr

You decide what customer data tunes your synthetic testers, how AI interprets it and what it can act on. Screeners keep the wrong people out of human tests, and Response Quality Monitor keeps low-effort answers out of your analysis.

Training data sources

3 approved·1 excluded

Response Quality Monitor

SOC 2 GDPR

Discover

Experience hub

Experience Hub streams and scores feedback; evidence enters backlog attached to tickets.

Design

Synthetic audiences

Synthetic Audiences and Expert AI Reviewer run from your design tool when there's something to react to.

Build

Spec check

Test the working build against the spec before the PR merges.

Monitor

Critical experiences

Critical Experiences are scored across the journey after release. You know if it worked.

Answers in minutes, not weeks— for a fraction of the cost

Get 25x more customer validation for the same budget, on average. And it starts at just $1 per synthetic test.

	Talk to customers yourself	Pay a testing panel	TheySaid Human	TheySaid Synthetic
Time to complete	Weeks	Days	Hours	Minutes
Who you hear from	Whoever agrees to a call	Professional testers	Your customers or screened testers	Audiences built from your customers
Cost per user	Your team’s hours	$50–100	$12–24	From $1
Sample size	5, if you’re lucky	10–20	4x more for the same budget	As many as the decision needs
Rerun after a change	Start over	Pay again	Relaunch in one click	One click
Who has to run it	You	A researcher	From MCP, or automatically when a ticket moves	From MCP, or automatically when a ticket moves

Here's whatTheySaid

The questionseveryone asks first

How accurate are synthetic testers compared with humans?

In a typical project, of all the issues discovered, 79% will come from both human and synthetic audiences, 10% only from synthetic, and 11% only from humans. So tests run only with synthetic will uncover 89% of total issues, nearly as many as tests run only with humans (90%). And synthetic should always be included in testing because they catch problems that humans miss.

How are your synthetic audiences different from other companies that offer digital twins or synthetic users?

TheySaid is the first and only company with technology that allows synthetic testers to record their screen while speaking their thoughts aloud. The deliverable isn’t a wall of Claude text, it’s a video recording with audio of a synthetic customer, trained on your data, using your prototype, app, Claude concept, or website.

Is this another tool my team has to manage?

No. TheySaid runs inside the tools your team already uses — Claude Design, Claude Code, Cursor, Jira, and more. One person connects it once, in under 2 minutes. Nobody else logs into anything new, and nobody has to remember to run it — tickets trigger it.

Who are the people in human tests?

Your own customers or people we recruit from our huge panel of testers. Screeners make sure only the right people get in. We schedule and pay them for you, and results come back in hours, not weeks.

Can I test with my own customers?

Yes. You can test with your own customers, testers we recruit for you, or third-party panels. We screen, schedule, and pay participants for you, and results come back in hours, not weeks.

Which tools does TheySaid work with?

The ones you already use — nothing new to learn.

• AI design tools: Claude Design, v0, Bolt and Lovable
• AI coding tools: Claude Code, Cursor and Replit
• Ticketing tools: Jira, Linear, ClickUp and Monday.com

How fast do I get results?

Synthetic tests finish in minutes. Tests with real people come back in hours, not weeks. One person can connect TheySaid in under 2 minutes, and you can run your first test right away.

How are TheySaid’s synthetic testers different from generic AI personas?

Generic personas come from a prompt. TheySaid’s testers start from 100,000+ hours of real usability sessions, then learn from your own customer data. You can also run one test across roles — like buyers, end users, and execs — and compare how each one reacts.

How do you keep low-quality testers out of my results?

Screeners keep the wrong people out of human tests before they start, and Response Quality Monitor removes low-effort answers before they reach your analysis.

Can I run tests in other languages?

Yes. TheySaid supports more than 70 languages across user tests, interviews, and surveys, so you can hear from customers around the world in their own language. TheySaid 4.0 adds auto-translation, so you can run the same study across markets without rewriting it.

Is it secure enough for enterprise teams?

Yes. TheySaid is SOC 2 and GDPR compliant, and you control what customer data AI can see and act on.