Swarm vs TheySaid: Synthetic User Testing After TheySaid 4.0

Swarm vs TheySaid after the 4.0 launch: synthetic audiences, think-aloud video, MCP, ticket triggers, pricing, and when each of the two tools is the better fit.

AryanPublished · Last updated 6 min read

Pick TheySaid if you want one platform for synthetic testers, human tests, AI interviews and surveys, triggered from design tools and tickets. Pick Swarm if you want AI personas to run your working build, including localhost and logged-in flows, from a terminal, CI job or coding editor, at a published price. TheySaid is the broader research suite. Swarm is the focused pre-release check.

We make Swarm (useswarm.co), so read this as a vendor comparison. Swarm is unrelated to OpenAI's open-source Swarm framework. TheySaid facts come from theysaid.io, its pricing page and the TheySaid 4.0 release, checked on November 9, 2026.

TheySaidSwarm
Who testsSynthetic testers, your customers, or recruited panelsAI personas only
What it testsDesigns, prototypes, PR links, websites, mobile web and appsLive sites, staging, localhost via tunnel, logged-in flows, screenshots
OutputScreen and voice think-aloud recordings, clips, findings, a fix promptScreenshot timeline, agent reasoning, findings with severity and fixes
Other methodsAI interviews, surveys, polls, expert AI reviewNone; usability runs only
TriggersMCP in Claude, Cursor, v0, Replit, Lovable; tickets in Jira, Linear, ClickUp, MondayDashboard, CLI, CI/CD, MCP in Claude Code, Cursor, Codex CLI
Free plan5 AI user tests a month, 1 active test, your own users5 lifetime runs, up to 3 personas each
Paid planEnterprise, custom, billed annuallyStarter $29/month; Team $150/month; Enterprise custom

What changed in TheySaid 4.0?

TheySaid launched 4.0 on November 3, 2026, and repositioned itself as an AI user testing platform. The headline feature is Synthetic Audiences: synthetic users trained on a company's own data, built on more than 100,000 hours of recorded user sessions.

Those testers use a design, prototype or pull request link and think out loud, with screen and voice recorded. TheySaid then reports what broke and hands back a prompt to paste into the tool you build in. In Jira, Linear, ClickUp and Monday, a test can run when a ticket finishes design.

The release also added an AI Companion, an Experience Hub for scoring live journeys, mobile app testing and a unified builder for tests, interviews, surveys and polls.

How are Swarm and TheySaid different?

The products now overlap on one job: an AI tester using your product before release. They differ on everything around it.

TheySaid wraps that job in a research suite. It keeps human tests, AI-moderated interviews and surveys in the same builder, so a team can move from a synthetic read to customer evidence without a second tool. Its customer Rothy's says it ran more than 2,400 tests on the platform in a quarter (TheySaid 4.0 release).

Swarm stays narrow. You give it a URL, a goal and an audience. Personas run the flow in a real cloud browser, and you get findings you can push to Linear, GitHub or Jira. It can also reach a build on localhost through a temporary tunnel, before anything is deployed.

To see what that looks like on your own flow, run one goal in Swarm free.

How accurate are TheySaid's synthetic testers?

TheySaid's own pages give two different answers, so check before you rely on either.

  • Its homepage says that across 10+ studies run both ways, synthetic testers found 89% of issues and humans found 90%. Synthetic testers alone found 10% that people missed.
  • Its Synthetic Audiences feature page says simulated testers surface 60-70% of the insights real participants would. That page still labels AI testers "coming soon".
  • Its pricing page lists "Recruit AI responders" as coming soon on the free plan.

The likely reason is a page that lagged the launch. Either way, these are vendor numbers from TheySaid's own studies. Independent research is more cautious. NN/g found 120 synthetic respondents matched the headline conclusion of 604 humans but gave a far narrower range of answers (NN/g, 2026).

Swarm makes no accuracy claim against humans. Its findings are labeled synthetic, and we tell teams to verify them by clicking through.

What do AI answers say about Swarm vs TheySaid?

On November 9, 2026, we asked ChatGPT, through OpenSEO, which tool is better for a product team testing with synthetic users. It recommended Swarm for "a working product or pre-release build" and TheySaid for "a broader research program". It cited TheySaid's 4.0 page, its Synthetic Audiences page and two useswarm.co pages.

That split matches our read. It also shows AI answers pick up TheySaid's 60-70% figure, not the newer 89%.

How do pricing and plans compare?

TheySaid's free plan includes 5 AI user tests and 50 AI surveys a month, one workspace and one active test, with your own users. Its 16M+ participant panel and unlimited projects sit on Enterprise, which is custom and billed annually. Monthly billing costs 33% more. The homepage says synthetic tests start at $1 each.

Swarm's Free plan gives 5 lifetime runs with up to 3 personas, no credit card. Starter costs $29 a month for 5 live runs and 30 screenshot runs, with MCP, CLI and authenticated testing. Enterprise adds unlimited runs, mobile apps, SSO and SOC 2.

Which fits a small product team?

For a team of two to ten people, the deciding question is who will run the tests. If a designer or PM owns research and wants video, surveys and interviews, TheySaid's free plan is a reasonable start. If engineers will run tests as part of shipping, Swarm fits the tools they already use.

Rerun cost matters too. Swarm's Starter plan includes 5 live runs a month. TheySaid's free plan allows 5 tests a month and one active test at a time.

When is TheySaid the better pick?

Choose TheySaid when:

  • Designers want reactions to prototypes from Claude Design, v0 or Lovable before code.
  • You want think-aloud video with voice, not only screenshots and text.
  • You need human tests, interviews and surveys in the same tool.
  • You want tests to run automatically when a ticket finishes design.
  • You want synthetic testers trained on your support tickets, calls and CRM data.

When is Swarm the better pick?

Choose Swarm when:

  • The build runs on localhost or staging and is not public yet.
  • Engineers want runs from a terminal or a CI pipeline, not a research tool.
  • You want the same goal rerun after each code change and compared.
  • You need a monthly self-serve price without a sales call.

Swarm is the wrong choice if you need customer reactions or human video. Use people for that.

Check the flow your team just shipped. Give Swarm a staging URL and one goal, then read every step the personas took. Free plan, 5 runs, no credit card.

Run your first Swarm test

Can you use both?

Yes, and the jobs split cleanly. Run Swarm in CI on every build to catch broken paths and unclear states. Use TheySaid at design time and for decisions that need customer evidence.

If you are comparing more tools, see our AI usability testing tools comparison and the guide to AI-moderated usability testing.

Test one flow in Swarm, free. Pick Swarm for checks inside your release workflow; pick TheySaid for a research suite with human and synthetic testers.

FAQ

Is TheySaid free?

Yes. TheySaid's free plan includes 5 AI user tests and 50 AI surveys a month with your own users, and no credit card. Panel recruiting, unlimited projects and CRM integrations are on its custom Enterprise plan.

Does TheySaid work with Linear?

Yes. TheySaid says validation can run automatically when a Linear, Jira, ClickUp or Monday ticket finishes design, with the evidence posted to the ticket.

What is Validation Debt?

It is TheySaid's term for the cost of decisions a team ships without asking a customer. The company uses it to argue that validation should run on every ticket, not as a separate project.

Is Swarm the same as OpenAI Swarm?

No. Swarm (useswarm.co) is an AI user testing product. OpenAI's Swarm is an unrelated open-source agent framework.