AI-written contact form spam is outreach, and sometimes a scam, that a language model writes for each business after looking at its website, then submits through its contact form. It gets past the old defences because it gives them nothing to catch: the wording is ordinary business English, the sender address is whatever was typed into the form, and challenge CAPTCHAs can be solved cheaply by paid services. What still gives it away is substance. It praises your work and asks for your time, but contains nothing that only the sender could know.
The short version
- Much of it is machine-written now. Researchers estimated that at least around 51% of spam in their dataset was generated using language models as of April 2025, and found it more grammatical and formal than the human-written kind.
- Keyword lists, sender blocklists and CAPTCHAs each miss it. A 2026 preprint found that commercial solving services achieve "near-perfect bypass" of challenge-based CAPTCHAs.
- The dependable test is what a message says. A real enquiry contains a detail from the sender's own situation and asks something you could answer; machine-written outreach opens with a compliment and asks for fifteen minutes.
- Some of it is dangerous. Fake copyright complaints sent through contact forms have been used to deliver malware since at least 2021, documented by Microsoft and seen again by Proofpoint in February 2024. Do not open the link.
- Any filter that judges the message is probabilistic and sometimes wrong both ways, so held mail should be kept rather than deleted, and a new filter should run in dry run before it holds anything.
Where to start
- Why your spam filter stopped catching contact form spam explains what changed, why the old tools miss it, and how to test a message-level filter safely.
- AI-written contact form spam: what it looks like, how to spot it covers the four patterns in machine-written sales approaches and what a real enquiry does instead.
- The contact form spam you get most, and how to recognise each is a field guide to the recurring types, with the tell for each.
- Fake legal threats through your contact form: the malware lure sets out how the copyright lure works and what to do instead of clicking.
- Fake form fills are training your Google Ads campaign explains how junk submissions that fire a conversion tag train the bidding, and what fixes it.
Terms used here
- AI SDR: an automated "sales development representative", software that writes each business a personalised message and sends it, aiming to book a meeting.
- Scraped noun: the specific detail in a machine-written compliment, taken from your homepage or latest post.
- Lure: a message built to get you to click, download or sign in, such as a fake copyright complaint with "evidence" behind a link.
- Honeypot: a form field hidden from people and left empty by them, which simple scripts fill in and so give themselves away.
- Offline conversion import: sending the outcome of a submission back to Google Ads days later, so the bidding learns from qualified results rather than from every form fill.
- Dry run: a trial period in which a filter holds nothing and only records what it would have held, and why.
Related on this site
- AI spam that reads like a real enquiry sets out the four things that separate a machine-written approach from somebody who needs you.
- Contact form spam started suddenly explains why a quiet form starts receiving spam all at once.
- How to stop contact form spam without a CAPTCHA starts with the defences no real visitor has to pass.
- How it works describes how Humainbox holds machine-written spam for review and forwards the rest, starting in dry run.
- How filtering works collects the articles on judging a message by what it says.