Where creatives competebefore your budget does.
Two creatives go in. Up to a thousand synthetic consumers argue it out. You launch the one they believed — before a dollar of media moves.
Free in early access · No card required
Up to 1,000 panelists per runVerdict in minutesNo fake CTRs, ever
Every campaign begins with a guess.Yours doesn't have to.
Splitroom is the missing step between “the creative is done” and “the campaign is live.”
The person who made the creative can't also be its audience.
You've seen your ad two hundred times. Your audience will see it for two seconds. Taste built the creative — taste can't judge it. A thousand strangers with different priorities can, and they don't care which one is your favorite.
Meta will happily charge you to discover your loser.
A real split test spends real budget through the learning phase — on both creatives, including the one that was never going to work. A simulation finds the likely loser before it earns a single impression. Your media budget confirms; it doesn't explore.
Ask a customer why they clicked and they'll invent a reason.
Post-launch metrics tell you what happened, never why. Our panelists aren't allowed to hand-wave: every reason they give must trace back to something measured on both ads. When the report says B feels safer, it can show you exactly what made it feel that way.
Creative windows don't wait for focus groups.
By the time a traditional panel reports back, the campaign shipped, the offer changed, and the trend died twice. A verdict in nine minutes fits inside the only gap that matters — between final export and campaign launch.
Opinions don't scale. Evidence does.
Both ads scored on the same dimensions, side by side. Nothing is compared that wasn't measured on both.
Trust. Three verification marks. The proof lands at a glance.
Emotion. Two people, morning light. Warmth arrives before words.
Readability. One benefit line. Nothing competes with it.
Brand fit. The voice matches the shelf. Familiar without trying.
Health cues. Clean white space reads clinical. Clinical reads safe.
Purchase intent. A clear claim over a clear ask. Hesitant buyers know the next step.
A thousand opinions. One decision.
Every panelist has an occupation, a buying style, and their own reasons. They don't agree with each other — that's the point.
- Buying style
- Careful Researcher
- What moves them
- Needs proof before price.
- Their reaction
- “The three check marks did it for me. I don't buy on vibes.”
Listen before you launch.
Panelists don't fill in a survey. They argue. You get the transcript from the focus group you didn't have to book.
Same product. Two ads. Let's find out where you split.
“B tells me what it is and proves it. A asks me to feel something first. I don't shop on feelings.”
“But you remember the feeling. Tomorrow, Meera remembers the breakfast. Nobody remembers a checkmark.”
“I remember checkmarks. It's kind of my whole thing.”
So trust versus warmth. Sana — you have both in your cart every week. Where do you land?
“I wanted to pick the pretty one. But the clean one tells me exactly what I'm giving my kids — 10 grams, three checks, done.”
Vote changed · A → B“You're all overthinking it. One of these told me what it was before my thumb moved. The other asked me to figure it out.”
That's the verdict: A wins hearts. B wins the two seconds that decide — and on a feed, the two seconds are the market.
of this panel chose Creative B — after the argument, not before it.
Consensus isn't universal.
One creative wins the room. The other wins rooms you care about. Re-cut the same panel and the map redraws itself.
Launch B broadly. Run A for luxury and gifting. That's two campaigns from one test — found before a single dollar of media.
Somewhere in your ad account, a losing creative is spending money.
Run a SimulationLaunch with evidence, not optimism.
Verdict, evidence, voices, and the three fixes that matter — assembled from up to a thousand individual decisions. This one's a walkthrough — read a real one.
One number does the selling. The checks do the proving. The warmth lost to the clarity.
“It says 10g and proves it three times. I don't need a mood — I need a reason.”
What we won't tell you.
A prediction tool earns trust by knowing its limits. These are ours, in writing.
Don't take our word — read a real reportA fake CTR.
No panel — synthetic or human — can predict click-through. Anyone selling you one is guessing with extra steps. We show you which creative people prefer and the reasoning behind it. The media plan stays your craft.
More confidence than the data earned.
Every report grades itself in the open: how much of your creatives the panel actually saw, how varied its reasoning was, and how much the verdict deserves your trust. Three scores, printed on the report.
A decisive winner that isn't one.
When two creatives are genuinely close, the report says so out loud — and tells you a 95% verdict on near-identical ads would be a lie. It recommends a small real-traffic test instead of a confident shrug.
Every campaign teaches the next one.
Every report you run joins your library. Past verdicts become context for the next ones — your account learns your audience, and the hundredth test is sharper than the first.
Fair questions.
Are the panelists real people?
No — and we won't pretend otherwise. Each panelist is a synthetic consumer built from your targeting brief, with an occupation, a budget, a buying style, and priorities that differ from their neighbor's. Treat the verdict as a disciplined read on how your audience reasons, not as a survey of humans.
How is this different from pasting my ads into ChatGPT?
A chat model gives you one articulate opinion. Splitroom runs a pipeline: both creatives are measured on identical dimensions before any judgment happens, hundreds of panelists with genuinely different priorities decide independently, and every claim in the report traces back to something scored on both ads. Then the report grades its own confidence — a chat window never will.
Will it predict my CTR or ROAS?
Never. Anyone selling you a synthetic CTR is guessing with extra steps. We tell you which creative your audience prefers, why, which segments disagree — and how much to trust that verdict.
Do you train your AI on my creatives?
Today, no — nothing you upload trains any model. Your history only sharpens your own runs: past verdicts become context for your future ones. If that ever changes, it will be opt-in, aggregated, and anonymized — and you'll hear it from us before it happens.
If I select several age ranges or income levels, how does the panel split between them?
Not evenly, and not by a hidden quota — by plausibility. Your filters define who is allowed on the panel; within them, the Audience Engine seats people the way the real market skews. Select middle and luxury income in India and you'll get far more middle, because that's the audience that actually exists. Each panelist is generated as a whole person — age, city, occupation, and budget that make sense together — rather than slotted into demographic cells, which is how you get panels without 19-year-old rural CEOs. The trade-off is honest: the mix mirrors reality rather than your intent, so a thin segment stays thin. Rerunning the same brand and brief reuses the same panel, so your comparisons stay apples to apples.
What do I need to run a simulation?
Two creatives and a description of your audience — country, age, income, platform, buying behaviour. The same brief you'd hand a media buyer. The verdict arrives in minutes.
Can I test more than two creatives?
A simulation is a head-to-head: two creatives, one verdict, and the reasons. Testing your top contenders bracket-style takes a few runs and still finishes before your morning standup.
What does it cost?
We're in early access. Your first simulations are free — enough to judge us the way the panel judges ads.
You spent weeks creating your campaign.
Spend two minutes testing it.
Free in early access · No card required