Cortexa · for marketers and creators
We predicted 66.7% of ad winners on a within-channel holdout.
Upload two ads — or a baseline plus a candidate video — and Cortexa tells you which one will perform best, before you spend on it. One model. Real numbers. No vibes.
Method · v64a_PROD · KuaiRand-disjoint val · 84-pair channel holdout
What we actually measured
66.7%
Held-out within-channel pair accuracy on 84 channel-disjoint comparisons. The industry baseline a marketer would otherwise use — coin-flip plus seasoned intuition — averages around 51% in independently audited studies.
The Cortexa model is built on a 4-pillar harmonic Winner Score — Hook, Hold, Algorithmic Fit, and Brand Lift — calibrated against real downstream conversion data. We don't optimize for engagement alone; we optimize for the joint distribution of attention, retention, distribution, and lift.
Every result you see on Cortexa explicitly separates what the brain-pattern model can predict from what it can't. The load-bearing UI element on Content results is a three-segment bar attributing performance to brain-pattern advantage, algorithmic chance, and unknown noise — not a gauge dialed up to look confident.
Two products
For marketers and for creators.
Cortexa Ads
Compare creatives before you spend on them.
Upload 2 to 5 ads. Cortexa returns a predicted winner with a Hook / Hold / Algorithmic Fit / Brand Lift breakdown and translates the technical metrics into language a growth team actually uses.
Open Ads →
Cortexa Create
Predict winners within a channel.
Set a baseline video for your channel. Drop in a candidate. Cortexa returns a within-channel win probability with the honest breakdown of what's predictable and what isn't.
Open Content →
How it works
Three steps.
One model.
No retraining per channel.
- 01
Upload your creatives.
2–5 ad variants, or a baseline plus a candidate video. Drag, drop, done. We sign the URLs server-side; the files never leave your account.
- 02
We run the brain-pattern model.
The same 4-pillar harmonic Winner Score we use in research — Hook, Hold, Algorithmic Fit, Brand Lift — applied to your specific demographic cell.
- 03
You get a verdict, in plain English.
A predicted winner, the metric translation table, the rationale, and an honest breakdown of what the model can and cannot explain.
Pricing
Start free. Upgrade when the model has paid for itself.
Free tier is a hard cap — five ad comparisons, three content predictions, no daily reset. You decide when it's worth more, and you only ever buy what you'll actually use.
- Free
5 ad comparisons · 3 content predictions · lifetime cap, never resets
$0 - ProRecommended
100 ad · 50 content per month · 3-ad comparisons
$20 / mo - Max
1,000 ad · 500 content per month · 5-ad comparisons
$100 / mo - Team
Pooled usage · shared library · centralized billing
$30 / seat - Enterprise
Unlimited · private deployment · SLA
Contact