Introducing Simulations →
RunPivot
Behavioral research by RunPivot

Synthetic personas

AI-driven behavioral research that predicts conversion before you spend.

RunPivot's synthetic personas run through your landing pages, products, and campaigns to predict who will convert, who will leave, and why. Run one before you spend the money on a live A/B test.

Start free simulationBook a walkthrough

Why synthetic personas matter.

A live A/B test waits for traffic. Synthetic personas give you a directional answer before you spend it. They power RunPivot Simulations.

  • Not a role-play prompt

    A behavior, not a job title

    Asking a model to review a page as a customer gets you a paragraph. A synthetic persona is a behavioral model: patience, reading depth, and intent decide what it notices and where it stops.

  • Built from your traffic

    The visitors you actually get

    On a connected site, the personas are built from your traffic, so the panel reflects the people who land on your pages, not a generic shopper.

  • Labeled as a prediction

    Directional, before the spend

    Every result is marked predicted. Use it to decide what is worth testing. A live experiment still decides what ships.

Under the hood: how RunPivot synthetic personas work.

Four things separate a synthetic persona from a model asked to pretend. Each one answers a problem with how synthetic research is usually done.

  1. 1

    Grounded personas, not generic agents

    The problem. Most synthetic tools wrap a standard model in a short prompt. The answer sounds realistic, but changing the job title changes the tone, not the decisions.

    The RunPivot difference. Each persona is defined by how it behaves on a page:

    • Patience: how long it stays and how much friction it tolerates
    • Reading depth: whether it scans headlines and buttons, or reads proof, pricing, and fine print
    • Intent: how ready it is to act when it arrives
    • Arrival context: a paid click checking the ad, a mobile skimmer, a returning visitor

    On a connected site, the panel is built from your traffic, and you can shape it further with your own customer research.

  2. 2

    Predictive validity you can measure

    The problem. A simulation that never meets a live result is a story. You need to know how often the prediction holds up.

    The RunPivot difference. In validation testing against live A/B tests, RunPivot synthetic personas correctly predict the winning variant 73–81% of the time, depending on traffic volume and test complexity. That means you can:

    • Drop low-probability ideas before they burn traffic
    • Prioritize high-confidence tests for your live experimentation queue
    • Reduce time-to-learn from weeks to hours

  3. 3

    Behavioral rigor beyond flat HTML

    The problem. A language model can read a whole page at once. A visitor cannot, and most pages are not static text.

    The RunPivot difference. Synthetic personas do not just read the copy. They interact with:

    • Dropdowns and multi-select menus
    • Tabbed navigation and accordions
    • Multi-step forms, including real-time error states and field-level friction
    • Dynamic pricing calculators and configurators
    • Modal flows and exit-intent popups

    Every simulation outputs hard metrics, not just qualitative feedback:

    • Predicted time-to-task completion, in seconds
    • Interaction-cost score, a 0–100 friction index
    • Scroll-depth drop-off points
    • Hesitation markers on pricing and checkout pages

  4. 4

    Guardrails against AI sycophancy

    The problem. Language models lean agreeable. Feedback that likes every page is useless for research.

    The RunPivot difference. RunPivot uses adversarial prompting and persona-specific guardrails, so each persona has its own goals, patience threshold, and dealbreakers:

    • Skeptical researcher: challenges weak copy, hidden pricing, and vague value propositions
    • Price-sensitive browser: bounces on unexpected shipping costs or forced account creation
    • Power user: skips onboarding fluff and hunts for advanced features
    • Mobile skimmer: reads headlines and buttons, and acts only if the next step is obvious
    • Paid-ad visitor: checks whether the page matches the ad, and leaves fast if it does not

    No persona is a yes-man. When they disagree, that split is the test.

Meet the default panel.

Four visitor types that want different things from the same page. Each one has its own patience, reading depth, and intent.

  • Paid-ad visitor

    Checking the promise

    Clicked an ad a few seconds ago and is checking whether the page matches what the ad promised. Leaves fast if it doesn't.

    Patience
    Reading depth
    Intent
    Arrives from an adFirst visit

  • Mobile skimmer

    One screen, then gone

    Reads headlines and buttons only, scrolls one or two screens at most, and acts only if the next step is obvious.

    Patience
    Reading depth
    Intent
    MobileScans, doesn't read

  • Skeptical researcher

    Wants proof first

    Compares several options on desktop, reads details, and looks for proof, pricing and fine print before committing.

    Patience
    Reading depth
    Intent
    DesktopNeeds proof

  • Returning visitor

    Wants the short path

    Has visited before and already understands the offer. Wants the fastest path to finish what they came to do.

    Patience
    Reading depth
    Intent
    ReturningWants speed

Customization and enterprise integration.

Shape the panel with your own research, then test the experience each visitor actually gets.

Bring your own data

Personas in your customers' words

Upload your actual customer insights to customize the default archetypes:

  • Customer persona documents (PDF, Notion, Google Docs)
  • User interview transcripts and session notes
  • Google Analytics behavior flows and funnel drop-off reports
  • Hotjar or FullStory session replay highlights

Synthetic personas adapt to your audience's real language, objections, and decision criteria.

Dynamic and personalized content

Test the pages you actually serve

Your site serves different experiences based on:

  • Geo-location, such as US vs. AU pricing, currency, or compliance copy
  • Referral UTM tags: paid search, organic, or partner
  • User state: logged-in or logged-out, free or paid tier
  • Device and traffic source, such as mobile paid social vs. desktop organic

Synthetic personas test those specific variations with persona-traffic matching.

Commercial and legal guardrails.

What you submit for a simulation is used for your simulation, and nothing else.

  • Training isolation

    Your data stays yours

    Any URL, mockup, or proprietary copy you submit for a simulation stays isolated. It is not absorbed into RunPivot's global model training data or used to improve other customers' simulations.

  • Business and Enterprise

    Enterprise controls

    Run private simulation environments with dedicated model instances, set data retention to auto-delete after 30, 60, or 90 days, and export simulation logs for internal audit and compliance.

What you get from a synthetic persona simulation.

Every simulation run gives you numbers to compare and a readable account of who acted on which version, and why.

Conversion probability score
A 0–100 score for each variant.
Friction heatmap
Where personas hesitate, scroll back, or abandon.
Persona debriefs
Qualitative insight in each persona's words, such as “I didn't trust the pricing page because…”
Recommended next tests
Ranked by predicted lift and confidence.
Exportable reports
PDF, CSV, or Notion, for stakeholder alignment.
A draft experiment
The candidates, the goal, and the notes, ready for a live A/B test.
Results

Pricing page CTA

40 simulated visits · Goal: Start a free trial

Predicted winner: BPredicted, not measured

Candidate B · journeys

Desktop

Simulated trial starts

A · Original7/20
B · Variant 112/20

By visitor type

  • Paid-ad visitorPrefers B · 4 of 5

    B repeats the ad's promise, so it stayed past the hero.

    Engaged: Hero headline · On A, 3 of 5 left at the hero

  • Mobile skimmerPrefers B · 5 of 5

    B's trial button is on the first screen. On A it's two scrolls down.

    Engaged: Primary button · On A, stopped scrolling before the CTA

  • Skeptical researcherPrefers A · 3 of 5

    Wanted pricing and proof before a CTA. B asked too early.

    Engaged: Pricing and logos · On B, left from the pricing table

  • Returning visitorPrefers Tie · 5 of 5

    Found the trial button on both. B was one click faster.

    Engaged: Primary button · No drop-off on either

Use cases for synthetic personas.

Before a test, after a flat result, while a campaign is still a concept, and wherever a price shows up.

  • Pre-test validation

    Is this test worth the traffic?

    Run a simulation before a live A/B test. If the personas cannot separate the candidates, the test may be low impact or need a sharper hypothesis.

  • Post-test diagnosis

    Why was the result flat?

    An inconclusive test can hide a split. The trace shows whether visitor types wanted different things, such as skimmers leaving a long form that researchers finish.

  • Campaign design

    Which concept goes to design?

    Compare concepts as mockups before high-fidelity work. Drop weak directions early and keep what each visitor type acted on.

  • Checkout and pricing

    Where does the number lose them?

    The skeptical researcher looks for pricing and fine print. Hidden costs and plans that are hard to compare show up as where it stops.

How to get started.

Free to start. No credit card needed.

  1. 1Connect your site with the RunPivot snippet, so the personas are built from your traffic.
  2. 2Start a simulation and add up to four candidates: a variant, a URL, or a mockup.
  3. 3Set the goal, such as starting a free trial. A metric is optional.
  4. 4Run it. Most simulations complete in 3–10 minutes, and you can leave the page while it finishes.
  5. 5Read the trace, then create the experiment from the run and launch it yourself.
Setup

1 · What are you comparing?

Pricing page CTA · No experiment attached

AOriginal

Control

Variant · The page you have

Bpricing-cta.png

Candidate

Mockup · Trial button on the first screen

Up to 4 candidates. Mockups up to 5 MB. Tall images are split into screens from top to bottom.

2 · What should visitors do?

Goal

Start a free trial

Metric, optional

None
Common questions

Synthetic personas, answered

What is a synthetic persona?
A synthetic persona is an AI model of a type of visitor, defined by behavior rather than demographics: how patient it is, how deeply it reads, how ready it is to act, and where it arrived from. In RunPivot, synthetic personas are built from your site's traffic and used to predict which version of a page gets more visitors to act.
What is the difference between synthetic personas and simulations?
Synthetic personas are the visitors. A simulation is the run: you give RunPivot up to four candidates and a goal, the personas go through each one, and you get a predicted winner with each persona's reasons. Simulations are the RunPivot feature. Synthetic personas are what powers it.
How is this different from traditional user testing?
User testing gives you 5–10 qualitative sessions over days, with recruitment and scheduling. Synthetic personas give you many simulated sessions in minutes, with consistent metrics and no recruitment overhead. They are a filter for what to test, not a replacement for watching a real person.
Can synthetic personas replace live A/B testing?
No. Synthetic personas are directional. They reduce risk and prioritize your live testing queue, but final decisions should always be validated with real traffic. RunPivot creates the experiment from the simulation, so the handoff is one step.
What's the turnaround time?
Most simulations complete in 3–10 minutes, depending on page complexity and the number of variants. You see an estimate before you start, and the run keeps going if you leave the page.
Do synthetic personas work on complex products like SaaS dashboards and marketplaces?
Yes. Synthetic personas handle authenticated flows, multi-step onboarding, and dynamic content. Provide test credentials or use sandbox mode for logged-in experiences. You can also compare mockups of screens that are not built yet.
How accurate are synthetic persona predictions?
Directional accuracy ranges from 73–81% in validation studies against live A/B tests. Accuracy improves with clearer hypotheses, higher-traffic pages, and well-differentiated variants. Every result is still labeled as a prediction, so a live test decides what ships.

Ready to simulate before you spend?

RunPivot synthetic personas give you behavioral answers in minutes, not weeks.

Start freeBook a 15-minute walkthroughRead the benchmark report