Play two positioning angles out across 400 agents over ten rounds.

Instead of a one-shot A/B read, the team releases both positioning angles, "savings-first" vs "freedom-first", into two parallel Simulate runs of 400 target-matched agents. Over ten rounds the agents post, quote each other, and shift, so the team sees not just which angle lands first but which one builds a following and which one hardens a "sounds risky" faction at 45+ by the final rounds. Ad targeting follows the trajectory, not the first reaction.

Run Simulate
AuthorMariete
AgentSimulate
RuntimeRuns overnight
IndustryFintech
  1. 1

    Describe the task

    Simulate takes two competing positioning angles and plays each one out inside its own synthetic population over multiple rounds, no live audience, no ad spend, no waiting on a slow A/B test to reach significance. It spins up hundreds of cognitive agents with realistic demographics, priorities, and financial attitudes, drops each angle into a simulated feed, and then lets the agents post, reply, and reshare, reacting to each other round after round rather than to the copy in isolation.

    That is the whole point of running this in Simulate rather than a single-shot panel: the first-round reaction is rarely the final one. An angle that reads well on contact can curdle once a few sceptics start quoting it; a slow-burn angle can gather a following as agents talk each other into it. Tell Simulate exactly who the target segment is, give it both angles verbatim, set the round count, and ask for sentiment tracked by sub-segment across rounds, and it will show you where each angle is heading, not just where it starts.

    Run two positioning angles as two parallel simulations, 400 agents each, matching our target segment, over 10 rounds. Segment: US adults 25-55, actively saving, mixed risk tolerance, defined in target-segment.json. The two angles, verbatim: - Angle A ("savings-first"): "Build the safety net you keep meaning to." - Angle B ("freedom-first"): "Money that finally moves at your pace." Let the agents react to each other each round. Track, by age band (under-35, 35-44, 45+) and by round: sentiment trajectory, whether a vocal faction forms and whether it grows or fades, and the phrase agents keep quoting. Flag the round where each angle’s narrative settles. End with a one-line targeting recommendation per band, based on where sentiment lands, not round one.

  2. 2

    Give Simulate context

    In a Simulate workspace, point the run at your segment definition so each synthetic population is built from real attributes rather than a generic "consumer," then set a round count high enough for the discussion to actually develop, a handful of rounds is enough to see a faction form and either spread or burn out.

    Give Simulate both angles in final copy, not a summary, phrasing is what agents seize on and quote to each other, so a paraphrase will hide the exact line that seeds a faction two rounds later.

    Required context
    Both positioning angles in final, verbatim copy
    A target-segment definition, age range, savings behaviour, risk tolerance, and income band
    Optional context
    A prior brand-voice note so agents weigh the angles against how they already perceive you
    Competitor taglines to benchmark each angle against the category norm
    An ad-platform export so the recommended segments map straight onto real targeting rows
    target-segment.jsonpositioning-angles.mdcompetitor-taglines.csv
  3. 3

    What Simulate creates

    Simulate runs each angle through its own 400-agent population over the rounds you set, letting agents influence one another, and tracks how sentiment moves per age band from the first round to the last. You get a per-angle trajectory, the round where a faction forms and whether it grows, the phrase that keeps getting quoted, and a targeting recommendation grounded in where each narrative settles.

    From Simulate: Ran 2 angles as parallel 400-agent sims over 10 rounds, tracked by 3 age bands.

    10Rounds per angle
    BuildsFreedom-first, under-35 over rounds
    Round 645+ risk faction consolidates
    Freedom-first: sentiment by round2 bands rising
    Age bandRound 1Round 5Round 10
    Under-35WarmPositive, agents amplifying itStrongly positive, self-reinforcing
    35-44NeutralWarmingPositive, softer than under-35
    45+: the “sounds risky” faction forms and spreadsgrows, then holds
    RoundFaction reach in bandWhat’s driving it
    Round 1A few sceptics“moves at your pace” read as volatility
    Round 6Roughly a third of the bandSceptics quoting each other; frame consolidates
    Round 10Holds; savings-first never triggers itSettled “could lose value” narrative

    "Freedom-first doesn’t just win on contact for under-45s, it compounds, agents talk each other further into it round over round. But at 45+ a ‘sounds risky’ faction forms by round 6 and holds to the end, while savings-first never sets it off. Want me to draft split ad copy by age band, or re-run with a hybrid angle that keeps the freedom hook but adds a safety cue before that faction consolidates?"

  4. 4

    Follow-up prompts

    Split the buy by the faction line

    Simulate found the age where freedom-first flips from a compounding win to a hardening liability, and the round it consolidates. Ask it to turn that into ready-to-load targeting rows so the media team ships two creatives instead of one compromise.

    Turn the results into an ad-targeting plan: freedom-first creative for under-45s, savings-first creative for 45+. Give me the audience definition and the headline for each row, ready to paste into the ad platform.

    Test a hybrid before the faction hardens

    The freedom hook compounds, the risk cue is what seeds the 45+ faction around round 6. Re-run with a third angle that keeps the momentum framing but adds a safety signal, and watch whether it holds the under-35 build without ever letting the 45+ faction form.

    Add a third angle: "Money that moves at your pace, protected the whole way." Run it as a third parallel sim against the same segment over 10 rounds and tell me whether the hybrid keeps the under-35 build while stopping the 45+ faction from forming.

  5. 5

    Tips and troubleshooting

    Feed final copy, not a gist

    The 45+ faction formed around five specific words that agents kept quoting to each other, not around the "freedom" concept in the abstract. A paraphrase would have hidden the exact line that seeded the cascade. Always test the copy you intend to ship.

    Read the trajectory, not round one

    The first-round reaction and the settled narrative were not the same thing here, freedom-first compounded and the 45+ faction only consolidated by round 6. Ask for sentiment across rounds up front; a single snapshot would have missed both the build and the faction.

    A synthetic run is directional, not gospel

    Simulate shows you where each narrative is heading and why, which is enough to pick a targeting strategy. Treat the trajectories as a strong prior to validate with a small live test, not as a final conversion forecast.

  6. Ready to try it yourself?

    Give Simulate both angles and your segment definition, and get a faction-aware, trajectory-based targeting call before you spend a dollar on live testing.

    Run Simulate