Heatcast: Predictive Eye Tracking & Website Heatmaps

See where people will look, before anyone visits. Heatcast predicts visual attention on any web page and shows you the result as a heatmap, a predicted scanpath, and a full-page report. It needs no tracking script, no analytics install, and no traffic at all, because it predicts attention instead of recording visitors. That means you can test a page before it is published, test a competitor, or test a design nobody has ever seen. WHAT YOU GET • Attention heatmap: warm areas are where attention is predicted to concentrate, cool areas are where it probably will not. • Predicted scanpath: not just how much attention, but the order. See which element is likely noticed first, and what gets skipped entirely. • Scroll-depth analysis: where your calls to action sit relative to how far down the page attention actually reaches. • Full-page report: a stitched heatmap of the whole page plus a ranked, filterable table of the highest-scoring elements. Export it as a standalone HTML file or save it as a PDF to send to a client. • Reversible CSS preview: try a design change in place and see how the prediction responds, then undo it. EVERYTHING RUNS ON YOUR DEVICE The attention model and the face-cue detector are bundled inside the extension and run locally on your machine. Screenshots, page content, predictions, and exported reports never leave your device. The extension is configured to block its own analysis code from contacting outside servers. No account. No login. No telemetry. The prediction runs on your own device, and no page content is sent anywhere to produce it. HOW THE PREDICTION IS BUILT Heatcast uses FastSal, a saliency model tested against eye-tracking datasets it was not tuned on, to produce the base heatmap. A calibration layer learned from the UEyes webpage eye-tracking dataset corrects for how people read interfaces rather than photographs. A behavioural layer then turns that static field into a predicted sequence, inhibiting recently inspected regions the way real attention moves on. On-device face detection can slightly favour a visible face that the base model already supports. It reports anonymous counts and shapes only. It does not identify anyone and does not infer emotion, age, gender, or any personal characteristic, and it stores no face images or landmarks. WHO IT IS FOR Designers checking whether the hero actually leads. Conversion and CRO people asking why a button gets ignored. SEOs and content strategists checking what a page emphasises. Agencies who need a visual, client-friendly artifact to justify a layout recommendation. WHAT THIS IS NOT • It is not recorded eye tracking. No camera is used and no visitor is watched. It is a forecast of where attention is likely to go. • It does not read your analytics. The prediction comes from the page itself, not from visitor data. • It is not a conversion prediction. Attention is not intent. A hot area means noticed, not persuaded. Predictions are a design question worth investigating, not proof. Confirm important changes with analytics, A/B testing, or user research. ONE LIMITATION WORTH KNOWING Chrome blocks every extension from running on the Chrome Web Store, on internal chrome:// pages, and on other extensions' pages. That is a browser rule, not a Heatcast bug. Any normal web page works. More detail, and the full method: https://gridlok.co/tools/heatcast/