Predict how your buyers react — before you spend.

Audience simulation is a method for predicting how a buyer audience will react to an ad creative without running it as a live campaign. Kettio generates synthetic respondent reactions from a chosen audience profile, embeds each reaction against calibrated anchor reference texts, and returns a 1–10 score with a conviction label. The result is a directional read on creative performance you can use to decide which creatives are worth your media budget.

Kettio’s SSR (Silicon Survey Ranking) engine has been validated against the University of Washington ad-perception survey at ρ=0.775 (n=160 paired creatives, text-embedding-3-large) and against the CreativeRanking real-CTR panel at 70.3% within-product pairwise. We publish the benchmarks because we measure them — competitors that claim prediction rarely do.

No. Kettio scores creatives without any ad account, pixel, or historical campaign data. Upload an image or paste a URL and the simulation runs against a chosen audience. This is what makes pre-launch scoring possible — there’s nothing to integrate before you can use it.

Kettio ships with four pre-built personas (Skeptical Millennial Browser, Loyal Boomer Ready to Buy, Gen Z Impulse Scroller, Cautious Parent Researcher) and lets you build custom audiences from your real buyer profile — demographics, motivations, objections, ad skepticism. The custom-audience layer is what turns a generic AI scorer into a brand-specific one.

AdCreative.ai outputs a confidence percentage from a single proprietary model with no published validation against real ad performance. Kettio publishes ρ against UW (0.775) and within-product pairwise on CreativeRanking (70.3%) so you can audit the prediction quality. Every score also ships with a rationale — the persona reactions in plain English — so you can see why a score landed where it did.

Yes. You can upload a mock-up, a competitor’s ad you’re drawing inspiration from, or even early concept boards. Simulation only needs an image and an audience — it doesn’t care whether the creative is final, draft, or borrowed.

Most runs return a calibrated score in 15–30 seconds. Two language models (Flash + Haiku) generate persona reactions in parallel, then we embed and score against anchor reference texts using text-embedding-3-large.

JPG, PNG, and WebP creatives for Meta feed, Meta stories/reels, TikTok in-feed, Google Display, LinkedIn single image, Amazon Sponsored Brands, Reddit Promoted Posts, and Pinterest Standard Pins. Each platform unlocks objective-specific scoring (Sales, Engagement, Awareness, Traffic, and more) because what reads as a winning ad on Meta doesn’t always read the same way on Google Display.

No — it’s a filter for it. A live A/B test on Meta costs $200–500 of spend and roughly two weeks before you have statistical significance. Audience simulation gives you a 30-second directional read so you can kill the obvious losers and run your live test on creatives that actually have a chance. The two methods are complementary: simulation eliminates wasted spend; live testing confirms the survivors.