Tell human form fills from bot submissions
Measure how a visitor fills a form, describe it in a few sentences, and let Jev decide person or bot before accepting the submission.
The use case
Site owners fighting spam sign-ups and fake leads dislike puzzles that annoy real visitors. In this build the browser records how the form is filled, such as typing rhythm, pauses and pointer movement, and the server writes those numbers up as a short text description. Jev answers whether this looks like a person. Clear bots are rejected, unsure cases get a second check such as email confirmation, and the site owner reviews blocked counts weekly. Respect privacy rules about what behaviour you record and tell visitors.
Example workflow
- Add a small script that records form interaction timings
- Summarise the measurements as short text on the server
- Ask Jev a Noul question on whether a person filled the form
- Accept, reject, or send unsure cases to email confirmation
- Review blocked and passed counts weekly
Tools
- Jev (TypeSafe AI)
- Cloudflare Workers
- Next.js
Risks and guardrails
- Accessibility tools and autofill can look bot-like
- Behaviour tracking needs privacy notice and consent
- Attackers adapt once they learn the signals
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- A CAPTCHA rebuilt with Jev by Jarek Ceborski on Made with Jev: madewithjev.com/builds/captcha-rebuilt-with-jev
- A CAPTCHA rebuilt with Jev: original post or repo by Jarek Ceborski: x.com/jarekceborski/status/2103071312323158512