How it works

How Plastic's synthetic user research works

Plastic runs synthetic focus groups. You build AI personas that know what real people in their position would know, and nothing more. They react to your materials, answer your questions, and produce a structured report with charts, quotes, and recommendations.

Plastic is most useful when you need to know what people will question, ignore, or resist before you ship. It applies frontier human decision simulation to model how different people reason, doubt, and decide. Personas respond from their own role, incentives, blind spots, and knowledge limits instead of falling into the usual helpful, generic AI voice.

Most teams use it to get early signal before a live study, stress-test materials before they ship, or check assumptions when scheduling real participants isn't practical.

How it works, step by step

1. Build personas with real depth

Each persona has a full profile: who they are, how they think, what they do for work, what frustrates them, and what they know and don't know. A logistics manager won't give you marketing advice. A first-time founder won't talk about enterprise procurement. These knowledge boundaries keep responses grounded in what each persona would realistically say.

You can create a persona in two ways. Most people start with Plastic Wizard: give it a short description of the kind of person you need, and it builds a reusable persona in about three minutes that is shared with the community, scores its realism from 0 to 10, and asks three follow-up questions until the profile feels solid enough to use. Only personas that reach the session-quality threshold can join sessions. For experienced researchers who need precise control, Custom Personas let you hand-craft every detail of the profile — they take longer to build but produce highly specific results.

2. Upload what you want to test

Bring whatever you'd bring to a real focus group: landing pages, pricing tables, onboarding screenshots, pitch decks, ad mockups, tweets, survey drafts. You can upload images, text documents, or PDFs (automatically converted to images). Personas see and react to the same things your audience would.

3. Ask your research questions

Write what you'd ask in a live session: "What's your first impression?", "Would you pay for this?", "What's confusing?" Each persona answers on their own, in character, based on their background and how they're feeling that day.

The answers are framed as inner voice rather than polished interview talk. That matters because people often have a harsher, more conflicted private reaction than the one they would say out loud in front of a moderator or salesperson.

That last part matters. Plastic picks from 17 mood states for each response. One persona might be relaxed and chatty, another stressed and short. If someone would be impatient with your question on a bad day, you'll see that too.

The prompts also include anti-generic safeguards. If a persona would not understand a domain, Plastic pushes them to stay confused, vague, or dismissive instead of inventing smart-sounding commentary. That preserves role-constrained reasoning instead of flattening everyone into the same articulate generalist.

4. Read the report

Every session runs two analyses at once: one for numbers, one for meaning.

You get a summary, three concrete recommendation cards, and the top three direct quotes with context. Each question gets its own breakdown. Scale questions ("How likely would you…") get automatic score distributions. Category questions ("What features matter most?") get grouped charts. Open-ended questions get theme summaries.

The report also flags fit. If a persona isn't your target buyer, the analysis calls that out so you can weigh their responses accordingly.

5. Clone, change, rerun

Changed the headline? Reworked the pricing? Clone the session, swap the materials, and rerun. The original stays frozen so you can compare results side by side.

That makes Plastic useful for repeatable synthetic research. You are not just collecting isolated reactions; you are comparing runs against a stable enough behavioral frame to see what likely changed.

What you can test

  • Landing pages and copy. Upload the screenshot, ask five personas for first impressions, see which version wins.
  • Pricing and packaging. Is your pricing clear? Does the free tier feel generous or cheap? Personas with different budgets give you different answers.
  • Purchase objections. Share your offer or sales page and let personas name the specific doubts that would stop them from buying.
  • Service offers and proposals. Run a pricing page, service package, or client proposal past your target persona and check whether the scope, terms, and price read as clear and fair.
  • Outreach messages. Cold emails, DMs, and LinkedIn sequences: find out whether your message earns a reply or gets ignored before you send it.
  • Positioning and messaging. Does your value prop land with your ideal buyer, or do they shrug while a different segment lights up?
  • Onboarding flows. Walk personas through your first-run screens. Find the step where a non-technical user says "I have no idea what this means."
  • Activation and checkout friction. Identify the gap between signup and first value — the step where someone stalls and quietly leaves.
  • Feature concepts. Before you build it, describe it. Ask: would you use this? How often? What would you pay?
  • Competitive alternatives. Ask personas who use a competitor how they'd evaluate your offer. They respond based on what they'd realistically know about both products.
  • Ad creative. Test angles across segments. Does the urgency hook work on enterprise buyers, or only on indie hackers?
  • Survey and interview questions. Run your draft questions on synthetic participants first. If a persona misreads the question, real ones will too.
  • Pitch decks. Build a panel of synthetic investors and run your deck past them. Find out which slides lose attention and where the hard questions come.
  • Tweets and social posts. Test a draft post against the audience you actually want to reach. See if the reaction matches your intent before you publish.

Best for

  • Founders who need pricing and conversion signal before a launch
  • Solo operators iterating on copy and offers without a research team
  • Consultants and freelancers stress-testing proposals and deliverables before presenting to clients
  • Agencies validating positioning and messaging across multiple client segments
  • Product managers who need user signal before committing engineering time
  • Marketers testing messaging across audience segments in the same afternoon
  • UX researchers pressure-testing study designs before spending the live budget

Not ideal for

  • Usability testing with interactive prototypes and click tracking
  • Research that needs participants' real purchase history or account data
  • Regulated contexts that require human-subjects review board approval
  • Replacing qualitative research entirely. Plastic is a fast directional tool, and works best alongside real studies

Frequently asked questions

How realistic are the persona responses?
Every persona stays in character and respects what someone in their position would actually know. A logistics manager won't offer opinions on UX design. Responses come in natural, spoken-word style, with hesitation, bluntness, or boredom when the question doesn't land. Plastic picks from 17 mood states per session, so responses include impatience, distraction, and skepticism alongside enthusiasm. When personas disagree or react in unexpected ways, the report calls that out — you see the tension, not a smoothed-over consensus. No single persona will match a specific real person, but they reliably surface the same objections, confusion, and preferences that live research does.
Can I customize the personas?
Yes, and there are two ways. Plastic Wizard is the fastest: describe the kind of person you need and get a scored, reusable persona in about three minutes. Wizard personas are shared with the community. Custom Personas are for experienced researchers who want precise control over every detail — they take longer to build properly but produce highly specific profiles. Available on Team and Researcher plans, private to your account.
How many personas can I include in a session?
Depends on your plan. Every plan supports multiple personas per session. They all respond at the same time, so more personas doesn't mean more waiting.
What file types can I upload?
Images (JPG, PNG, GIF, WEBP), text files (TXT, Markdown), and PDFs (up to 20 pages, auto-converted to images). Most teams upload screenshots, mockups, PDFs, or paste in copy they want reactions to.
Is my data private?
Your session materials, answers, and results stay private to your account. Custom Personas do too: they are available on Team and Researcher plans, give you full control over the full persona description, and are never shared outside your account. Plastic Wizard personas work differently: they are designed to be easy to create, reusable, and shared with the community. You can also create a share link for any finished session; it shows a read-only report and hides your raw materials, individual answers, and account details, and you can revoke it anytime.
How is this different from just prompting ChatGPT?
ChatGPT gives you one voice that tends to be helpful and agreeable. Plastic gives each persona their own identity, their own knowledge limits, and their own mood — with rules that stop them from just validating whatever you put in front of them. A skeptical logistics manager might say "I don't know what I'm actually paying for here, so I'd just Google the competitor first." You don't get that from a generic prompt. Sessions are good for repeated comparisons because you can rerun the same setup with new materials and inspect what changed. The output is a structured report with score distributions, recommendation cards, direct quotes, and per-question breakdowns you can share.
Start free trialSee pricing