AI-powered spam protection for contact forms

Where it started

If a website has a contact form, sooner or later spam starts showing up. The platform, the stack, the protections in place. None of it seems to make much difference.

We saw this across the sites we managed over the years. Client websites, landing pages, personal projects. The volume varied, but the pattern was the same: automated submissions filling inboxes with irrelevant pitches, phishing attempts, SEO offers, and outreach that had nothing to do with the business.

We tried different approaches along the way. Some helped reduce the volume, others added friction that ended up discouraging the people we actually wanted to hear from. A few worked well for a while and then gradually lost effectiveness as the patterns shifted. The messages that still got through were often the hardest to tell apart from something real.

The pattern we noticed

After years of working through this across different sites, something became clear. The spam had a pattern. Not a technical one, but a linguistic one.

Automated spam tends to read like a template. Generic greetings, placeholder-style names, vague references to "your website" without mentioning anything specific about it. Language that looks plausible at a glance but rarely shows any sign the person visited the site.

Legitimate messages are different. They mention specific things, ask about a particular service, include typos and informal phrasing. They sound like someone with a question and a few minutes to write it down.

The distinction comes down to context. A filter that understands what the business does has a much easier time telling whether a message is relevant.

What we built

That insight is where FormFend started taking shape. Instead of scanning for known spam patterns, it begins by learning about the business from the website itself. If the site describes what the business does, what services it offers, who it serves, FormFend picks that up and uses it as a reference point for evaluating what comes through the contact form.

When a submission comes in, AI evaluates it against that understanding. Does this message read like something a person interested in this business would write? Does it reference anything specific? Does the language feel human or templated?

The person filling out the form never sees any of this happening. No puzzles, no checkboxes, no extra steps. The form works the way it always did. The analysis runs in the background, and legitimate messages get delivered as expected.

What we did

A system that keeps learning

What makes this more than a filter is the feedback loop. Every submission is visible to the site owner. If something gets flagged incorrectly, whether a legitimate message gets caught or a spam message slips through, the owner can correct it. Those corrections feed back into the analysis, and over time the system becomes more accurate for that specific site.

The more it processes and the more feedback it receives, the better it gets at telling relevant messages from irrelevant ones.

From our problem to a product

FormFend started as something we built for ourselves. A way to stop spending time sorting through spam across the sites we managed. Once it was running and proving effective, turning it into a product felt like a natural next step.

You connect it, it learns what the business does, and it starts filtering. The simplest integration requires no code changes. For teams that want more control, there is a full API.

We are still building it out, but the core has been working well and the approach continues to hold up.