What counts as form spam and what does not
Before installing a filter, separate enquiries you want to answer from unsolicited submissions. A short message, spelling mistakes or another language do not prove spam. Consider what the person is asking and whether it relates to your business.
Real form spam has recognisable marks: messages full of links, texts with no context or copied from templates, repeated names and emails, fields filled with random strings. Cold agency pitches for SEO, backlinks, development or design also belong here, even when a person wrote them.
A legitimate but unusual enquiry is different: a potential customer writing from a phone in a hurry, someone who struggles to explain their need, a visitor from an unrelated sector. They sound strange but bring a concrete question or a sender who survives a quick search.
Dismissing as spam every message that does not match your ideal customer is a known source of lost opportunities. Before activating any filter, sort a handful of recent messages by hand. That sample will be your reference every time you tune a rule.
Where the spam reaching your form comes from
Your form can receive automated submissions, sales pitches written by people and genuine enquiries that resemble either. Language, timing and writing quality cannot reliably distinguish them.
Bots can run JavaScript and handle cookies too. A simple filter may stop some abuse without solving all of it. OWASP describes automated application abuse as a range of behaviours, rather than a single kind of bot.
Also check whether the form sends a copy to the address the visitor supplies. If that copy includes unrestricted message text, someone could use it to send messages to third parties. Bounces are a reason to investigate, not proof of that cause. A Reply-To address different from your own can be entirely normal for a contact form.
Before filtering, measure what is coming in
Review a sample of recent messages and note volume, repetition, genuine enquiries and possible rejections. You do not need to wait weeks before acting on active abuse or a form sending messages to third parties.
- Content: distinguish unsolicited pitches, repeated messages and enquiries that need an answer.
- Frequency: look for bursts without treating a particular time of day as proof of automation.
- Delivery: not receiving a bounce does not confirm that an address exists or received the message.
- Logs: if your team reviews IP addresses or browsers, use a defined purpose, restricted access and limited retention. Do not store full messages or sensitive data unnecessarily.
Use the sample to compare results before and after a filter. Leave uncertain messages for review rather than automatically discarding them.
Diagnosis: how much it interferes and with whom
With the sample collected, translate volume into real cost. Five odd messages a week is not the same as ten a day. The decision to add friction should rest on the time it costs and the risk it adds, not on how the inbox feels.
A practical set of criteria to decide whether the current level is tolerable:
| Situation | Reasonable decision | What to check first |
|---|---|---|
| A few odd submissions a week, no dangerous links | Tolerate and read the inbox with judgement. Adding friction costs more than the saving. | That nobody in the team marks a real email as spam by mistake. |
| Sustained daily flow burying real enquiries | Raise the first invisible layer (honeypot and timing). | That the invisible layer does not harm screen readers or older browsers. |
| Messages with active links clicked by accident | Intervene with priority. The risk is someone on the team clicking one. | Short internal reminder about not opening links from unknown senders, and server mail filtering. |
| Submissions spoofing the sender or producing bounces | Review how the form sends email. The server may be misconfigured. | Outgoing-mail authentication (SPF, DKIM) with whoever maintains your website. |
The operational rule that works best: start with the layer that disturbs legitimate visitors the least and only step up if the sample justifies it. A visible captcha installed on day one is, in many cases, a disproportionate barrier for a volume of noise you could have absorbed.
First invisible layer: honeypot and timing checks
A honeypot is a decoy field that people do not need to complete. Your team should exclude it from keyboard navigation and assistive technology, and check that autofill does not trigger it. Validation belongs on the server; some bots avoid these fields, so they are not a guarantee.
Submission timing can provide another signal, but a fast submission does not prove a bot: someone may use autofill or return to a prepared form. Avoid rejecting enquiries solely because they take fewer than a fixed number of seconds.
Test changes with a keyboard, a phone and autofill. The MDN guide to form validation distinguishes helpful visitor feedback from checks the server still needs to perform.
Hypothetical example: after adding a decoy field, a business compares rejected submissions with its normal enquiries. If genuine messages appear, it adjusts the filter before making it stricter. No complaints does not prove nobody was excluded.
Captchas: when they fit and when they add friction
If you need a CAPTCHA, assess effectiveness, accessibility and data handling. Notice or consent requirements depend on the service and configuration; a cookie checkbox does not resolve every case. Review this with whoever handles your website privacy.
- reCAPTCHA v3. Invisible. According to the official reCAPTCHA v3 documentation, Google returns a score between 0.0 and 1.0 and leaves the site to decide what to do with each submission. Requires picking and maintaining a threshold, and sends visitor data to Google.
- reCAPTCHA v2. Shows the «I'm not a robot» checkbox and, sometimes, a visual challenge. More friction, less threshold dependency.
- hCaptcha. Visible checkbox, different privacy approach and pricing. Behaves similarly to reCAPTCHA v2 for the visitor.
To choose, lay them next to the kind of visitor you expect:
| Situation | What usually fits | What to review before installing it |
|---|---|---|
| Medium volume, digital-fluent audience | reCAPTCHA v3 with a conservative threshold and manual review for low scores. | Review the data processed, legal basis and information or consent required for the service and its configuration. |
| High volume and resistant bots | Add visible reCAPTCHA v2 or hCaptcha as a targeted reinforcement. | That the challenge works on mobile with a keyboard and screen readers; prepare a clear fallback message. |
| Older audience, less digital fluency | Probably no visible captcha. Reinforce honeypot, timing and server filters instead. | That the form explains in plain language what happens after sending. |
| Medical, social or support forms | Avoid challenges that could exclude vulnerable people. | A clearly visible alternative: direct phone or email alongside the form. |
A visible challenge can backfire with the audience your business serves. Test the form with a non-technical colleague's phone; if they take longer or give up, the captcha is not helping you.
Server-side reinforcements: rate limiting, WAF and reputation
Not all the work has to live in the form. Several useful layers sit on the server and do not require touching the HTML. Effective against simple bots and automated attacks.
- Rate limiting by IP and session. Rejecting more than a set number of submissions per minute from the same address cuts the obvious bots. Whoever maintains your website should pick the threshold with the sample at hand, so offices sharing a public IP are not blocked.
- Log review. Read web server logs during the worst spam days. Compare request frequency at the address processing the form with accepted and rejected messages. An old browser identifier or a shared IP address alone is not enough to justify a block.
- WAF or hosting rules. Confirm whether an application firewall is available and what abuse it detects. Not every plan includes contact-form spam filtering. Review blocks before enabling broad rules; the HTTPS, WAF and CDN guide distinguishes their roles.
- Outgoing mail authentication. SPF, DKIM and DMARC help recipients assess mail sent on behalf of your domain. They do not prevent abuse of an authorised form: recipient, content and frequency controls are still needed.
- Sender reputation. For manual agency pitches, an internal list of domains already marked as cold outreach saves the team time.
Avoid broad country or IP-range blocks without assessing who could be excluded. A rule should address observed abuse, rather than a location or browser alone.
Check you are not losing real enquiries
A silent filter is worse than a form with noise. If a layer rejects legitimate enquiries without you noticing, you lose customers without warning. This step is the one most often skipped and the one most worth turning into a routine.
- Test the form under different conditions. From a phone on cellular, from a network outside your office and from a freshly installed browser without your cookies. If any fails to send or the captcha does not load, a real visitor will fail too.
- Review the mailbox spam folder. Not everything flagged as spam is. Spend a few minutes a week reading it with the same care as the main inbox. General guidance in web.dev's form best practices is a reminder not to sacrifice real users to add antispam layers.
- Keep a visible alternative channel. A phone number or a direct email on the contact page gives visitors an exit if the form fails.
- Keep a log of rejected submissions. If your form records why it discarded a message (honeypot, timing, low score, rate limiting), reading the log shows which filter cuts too much. Adjust the rule according to what it measures: lowering a minimum time is different from lowering a minimum trust score.
- If nothing arrives, first rule out delivery. Sometimes the problem is not noise but that form emails are not reaching the inbox. The article on WordPress form emails that do not arrive helps isolate whether delivery is healthy before blaming the filter.
Antispam layers are not install-and-forget. Review them every few months with an updated sample and whatever has changed: new channels, languages, forms.
If you want us to walk you through it
If spam is making it harder to respond to customers, tell us which form you use, roughly how much spam it receives and what you have tried. We can help review the causes and agree an improvement that also considers genuine visitors.
For a broader review, see what our security analysis covers. If genuine enquiries are missing too, start with checking form email delivery: filtering spam and delivering messages are different issues.