Upload your creative so your agent can inspect where human attention is predicted to go, what gets ignored, and how attention is distributed across the ad.
Your usual Claude or Codex.
Give your agent attention data.
Recheck. Compare. A/B test.
GO BEYOND.
Make the important parts easier to notice. Give people a clearer next step.
The intention
Tap an element on the phone to see its predicted attention share.
Lower HigherOura UI example. Independent analysis, not an endorsement.
A real screen. A reason to make it clearer.
Share of predicted attention across the full image.
Choose an element to see its measured boundary on the image.
The bright area around “Learn more” spans several UI elements. Inside the button’s exact outline, the share is 3.33% of the full photo, with 7.44× the photo’s average predicted density.
Each percentage is the model’s predicted attention inside a manually defined rectangle, divided by the total across the full 1,200 × 1,200 photo. It is not a percentage of people, clicks, understanding, or a prediction of what someone sees first.
The 16 non-overlapping regions account for 60.68%. The remaining 39.32% lies outside them, including scenery, gaps and the photo background. Values are rounded to two decimals, not claims of human measurement precision. The white ring marks the model’s peak.
Larger regions can collect more attention. Density relative to the image average adds size context, not click probability. This is one prediction of a photographed screen, not a usability study of the app.
The story behind attentionproof
Dmitry Baranau Founder, attentionproof
I design UI and create visuals with my Codex agent. Some looked great. Others missed the point, so I kept reviewing and re-explaining.
Meta’s TRIBE v2 research demo sparked a question: could my agent inspect predicted visual attention in the images we make?
With 8+ years in engineering, I built our own heatmap model. I use it in client work now. It saves me hours of back-and-forth.
Thoughtful visual decisions, without a whole agency team. That’s my secret sauce.
Compare monthly plans
Start with the price of a 100-analysis alternative. Then compare what each plan includes.
EyeQuant’s US$149 price comes from a one-user app view supplied September 25, 2026; its public page asks visitors to contact sales. Tools and usage units differ. Apple confirms your local AttentionProof price.
Swipe or scroll sideways to compare all three.
Your first 5 checks are free. 30 checks each month on Starter. 100 checks each month on Pro.
Per-image figures divide monthly price by the full stated allowance; they are not pay-as-you-go prices. AttentionProof shows US reference prices; Apple confirms your local price before purchase. EyeQuant prices are from the one-user monthly app view supplied September 25, 2026; its public site currently uses a different plan catalogue and asks visitors to contact sales for prices. Attention Insight charges in EUR; its approximate USD figures use the September 24, 2026 ECB reference rate of €1 = US$1.1367, not a live payment exchange rate. Its credits can also pay for URLs, AI recommendations or video, and longer images can use multiple credits. An analysis or credit is not necessarily equivalent to one AttentionProof check. *Unlimited plans have usage terms or fair-use limits. Prices and terms can change. Neither company is affiliated with AttentionProof.
Your iPhone. Your AI. One balance.
Choose your plan in the AttentionProof app. Use the same Apple Account to bring your checks to the web, Claude and Codex.
Continue to iPhone appEach ad or variant you analyze uses one check. Unused paid checks do not expire.
Your next creative direction starts here.
Try your UI or any image. No card or automatic upgrade.
US reference prices. Apple shows your local price, taxes and renewal terms before purchase. Monthly subscriptions renew automatically until canceled in Apple subscription settings. Free checks are a one-time allowance per account. Failed analyses return the check. Your agent and image-generation costs are separate.
One UI, three predictions
Here is the same saved Linear homepage through three attention tools. Look at the headline, the product preview, and the Sign up action in the top-right corner. Which prediction looks most useful for the design decision you would make?
The headline and product preview lead. The header actions, including Sign up, have much less visual emphasis.
Saved prediction overlay · Open image
The headline and product preview draw heat. The New Loops link and Sign up action are visible in the prediction too.
Saved overlay with vendor watermark · Open image
In the saved Engaged view, the headline carries the visible heat. The Sign up action is comparatively quiet.
Saved Engaged view, browser bar removed · Open image
The Sign up action appears most pronounced in the saved Attention Insight overlay. EyeQuant and AttentionProof give it less visible emphasis. That is a difference in predictions, not a measured accuracy result.
MIT/Tübingen benchmark: AttentionProof has no published MIT300 score or rank yet. We will show the official result here if it is listed. View the benchmark
These are saved outputs of the same visible design, not a controlled accuracy test. We cannot verify identical uploaded bytes, crops, settings, model versions, or run dates for all three. EyeQuant shows its Engaged mode. Heatmap color scales also differ, so color intensity cannot be compared numerically across vendors. Deciding which prediction is accurate requires human eye-tracking ground truth under matched conditions.
Our MIT/Tübingen benchmark submission is pending an official published result. No ranking is claimed today.
Linear UI is an independent example, not an endorsement by Linear or either competitor. The EyeQuant display omits the private browser address bar; prediction pixels are unchanged.
Product Agent MCP attention data Creative direction
Attention heatmaps for A/B tests
Keep every creative, AttentionProof heatmap, hypothesis and campaign result tied to the same version ID. Use the prediction to shape the next variant; use Meta or your testing software to measure the outcome.
Saved in your chosen project folder, with a version ID for each creative.
Add the campaign results to the same version IDs. Campaign data is not imported automatically.
Use the prediction. Keep testing the outcome. A heatmap predicts visual attention. It does not measure gaze, clicks or sales.
Before you start
AttentionProof predicts visual attention and returns the heatmap, overlay and attention data. Claude or Codex interprets the result, records the reasoning and uses its available image tools to create revisions. If your agent cannot generate images, it can prepare an editing brief instead.
No. A heatmap shows what is likely to draw attention. Use it to choose a specific revision, then run both ads in your A/B testing software to measure clicks, purchases, cost per acquisition and return on ad spend. AttentionProof does not run campaigns or import their results automatically.
The ad workflow asks your agent to save the creatives, heatmaps and version notes in a project folder you choose. Your AttentionProof account keeps usage metadata, not a web library of uploaded ads. Keep the version IDs when you add results from your ad platform.
Use AttentionProof to compare predicted attention, then run both versions in the advertising or A/B testing software you already use. Your ad platform measures clicks, purchases and campaign cost.
Start with a PNG or JPEG of your UI, ad, slide or photo. Choose your image here, then sign in to run the attention check. No agent setup is needed. New accounts include 5 free checks; each completed image analysis uses one. You can connect Claude or Codex later.
5 free checks. Start with your UI, or any image.
