Skip to content
MethodInsights

Form spam without captcha theatre

You have stopped spam when the bot fails quietly and the human does not have to prove it is a human.

Spam on the contact form is stopped on the server with a honeypot, timing and a cap — not with a puzzle in front of every human. Captcha is a last resort, not the first field.

15 min read
Stylised process diagram of a contact form with a hidden field and a server-side gate — not a puzzle in front of the human.

Method & web

If you are searching for contact form spam, or for honeypot vs captcha, you are typically in one place: the inbox is full of 'SEO offers' and prescription links, and someone has said you need reCAPTCHA on it. The portal post is when the form should become a state the customer can see. The rate-limiting post is the cap on the expensive call. The public-site security post is headers and admin on www. The cookie and GDPR posts are consent and data flows. This one is the bot in the POST. A puzzle in front of every human is not a filter. It is a tax you have put on the people you want to hear from.

A form that works is a POST you own. Spam is the POST you did not ask for. You stop it where you can recognise it without punishing the one you want: a field a human does not see, a time window a human does not beat, a cap per IP and per identity, a secret in the request that a scripted dump does not have. Captcha is what you turn on when that layer has been broken — not what you meet everyone with because the plugin sat at the top of a list.

TL;DR

The main takeaways

  • Spam is stopped on the server. A hidden field, timing and a cap beat a puzzle in front of every honest human.
  • A honeypot is a field humans do not see and bots often fill. It is boring. That is the point.
  • Captcha is a tax on the people you want to hear from, and a third-party script you have given your form. Use it when the boring layer is broken.
  • The rate-limiting post is the cap on the expensive call. This is the cap on this POST — plus owning the 429.
  • The cookie and GDPR posts are consent. The portal post is when the form should become self-service. This is the bot in the inbox.
  • You have stopped spam when the bot fails quietly and a human can send without proving it is a human.

Why this matters at all

A contact form is a promise that a human can write, and that you read it. The promise rots when the inbox is a feed of bots, and when the honestly interested human has to click a grid first. You have not made the form safer. You have made it more expensive for the person you wanted, and still open to the scripted POST that never saw the grid — because it hit the endpoint directly.

Teams do it because the first wave of spam is met with 'put captcha on, then it goes away'. It goes away in the UI. It comes back on the network. Three months later you have a reCAPTCHA badge, a cookie problem you did not have, an accessibility break, and still rows in the CRM from 'Dear Sir'. You rotate the keys. You do not touch the POST.

The damage is concrete. A human with a password manager, another language, or a browser without third-party cookies does not get through. An accessibility requirement you otherwise meet falls on the grid. An endpoint without the same check still accepts the dump. You notice when the good enquiry did not arrive, or when sales ask why the inbox is a newsletter you did not subscribe to.

We write this because 'we have set captcha' has become a check you tick when the badge is visible. An honest setup is boring: the server knows a hidden field, a timestamp, a cap and a secret. The human sees a form. The bot sees a trap it often walks into. Captcha is the door you open when the trap no longer bites. Without that layer you have theatre. You do not have a filter.

What a filter on the form actually is — and what a badge is not

A filter is a decision on the server: this POST is one you want, or it is not. The signal should be something you can defend without requiring a human to solve a puzzle. A honeypot field hidden with CSS or `tabindex` and `autocomplete=off` is filled by many bots and no humans. A timestamp you set when the form was shown exposes the POST that arrived in the same millisecond. A cap per IP and per form beats the wave that cannot be bothered to be clever. A signed token in a hidden field that you check on POST beats the dump that hit the URL without having seen the page.

It is not the same as 'we have a captcha plugin'. The plugin is a third-party script, a cookie, a challenge, and an API call you have to remember on the server. If you only show the widget and do not verify the token behind it, you have UI. You do not have a filter. If you do verify, you have still given the form's fate to a vendor, a cookie banner and an accessibility compromise. That can be right when the boring layer is broken. It is wrong as the first field.

It is also not the same as rate limiting in general. The rate-limiting post is the cap on login, search, webhooks — the expensive call. This is the cap on the contact POST specifically, plus the signals that are specific to a form: honeypot, timing, token. A global cap without those signals also hits the human who tries three times because the network dropped. A form cap without a global cap on login is a different hole. The two posts are neighbours. They are not the same job.

And it is not 'hide the form behind a login, then there is no spam'. The portal post is when a form should become a state the customer can see. Spam on a public contact form is a different problem. You can have a portal and still an open 'write to us'. That open POST still has to be owned. Login is not a honeypot. A honeypot is not consent. The cookie post is the banner. This is the bot.

What typically goes wrong when someone says captcha is set

The first failure is putting the widget in front of the human and forgetting the server. Bots POST the endpoint. Humans fight the grid. You have inverted the filter. The direction is the reverse: the server rejects what looks automatic. The human sees a form. Captcha comes on when you have evidence that the boring layer no longer bites — not because a list said 'best practice'.

The second failure is a honeypot humans can hit. A field you only hid with `display:none` but left in the tab order. A label a screen reader still reads as 'website', which someone politely fills in. A name like `email2` that a password manager fills. The honeypot must be out of the way for humans: `aria-hidden`, `tabindex=-1`, `autocomplete=off`, a name bots love (`website`, `url`, `company_url`) and humans do not expect to see. The trap that catches your customers is not a trap. It is a bad form.

The third failure is trusting `Referer`, JS-only submit, or 'hide the button until load'. The clever bot runs JS. The dumb one hits the endpoint. You have filtered the dumb ones in the UI and invited the clever ones to the POST. A minimum of JS can be a signal. It is not the filter. The filter is the server that knows token, time and honeypot — also when JS failed to load, so you can reject honestly instead of hanging.

The fourth failure is beating spam by moving it. You close the mail and open a webhook, a CRM, a Slack. The bot follows because the endpoint is still open. Or you set a cap so hard that a human on a shared IP (office, mobile network) hits 429. The rate-limiting post is owning that 429. This is letting the cap sit on the form, with an honest 'try in a bit', and letting the honeypot rejection be quiet — not a 500 you log as your fault.

Boring server-side vs. a puzzle in front of the human

The boring filter

The server knows the trap. The human sees a form. The bot often walks into it.

  • Honeypot out of tab, screen reader and password manager — bots fill it anyway
  • Timestamp and signed token from render to POST — the dump does not have them
  • A cap per IP and per form, with an honest 429 you own
  • Is boring to maintain, and that is the whole point — the human should not have to prove anything

Captcha theatre

A badge, a cookie, a grid, and an endpoint that still accepts the dump.

  • Widget in the UI, no token check in the handler — signs, not a filter
  • Third-party script, cookie and accessibility break as 'security'
  • Humans with other languages, managers and strict browsers drop off
  • Is a hope you call a spam filter. The inbox is still a newsletter you did not subscribe to
form-honeypot.htmlhtml
<!-- Visible to bots, hidden from humans. Checked on the server. -->
<form method="post" action="/api/contact">
  <label>
    Message
    <textarea name="message" required></textarea>
  </label>
  <div aria-hidden="true" style="position:absolute;left:-10000px;">
    <label>
      Website
      <input type="text" name="website" tabindex="-1" autocomplete="off" />
    </label>
  </div>
  <input type="hidden" name="form_started_at" value="SERVER_RENDERED_TIMESTAMP" />
  <input type="hidden" name="form_token" value="SERVER_SIGNED_TOKEN" />
  <button type="submit">Send</button>
</form>

<!-- Server: reject if website is filled, if the time is unrealistic, if the token is invalid, if the cap is hit. -->
<!-- Captcha only once that layer is broken — and then verify the token there, not only in the widget. -->

How to decide whether the form may meet a captcha at all

Without this order you get another badge you call a filter.

  1. 01

    Find the POST the mail is actually sent from — and put the trap there

    Not in the button. Not in a client-only check. In the handler. If you have two endpoints (Next.js route, WordPress admin-post, a Zapier URL), you have two traps you owe. A hidden field in HTML without a check behind it is decoration. The portal post is the state. This is the open POST.

  2. 02

    Set honeypot, time and token before you touch a third-party script

    A field bots fill. A time window a human can hit and a scripted dump cannot. A signed value you issued when the page was shown. That catches the boring wave. It does not cost a human a grid. If you skip it and go to captcha, you have paid the tax without trying the cheap gate.

  3. 03

    Put a cap on this form specifically, and own the 429

    Per IP, per identity, per day — what you can defend. The rate-limiting post is the expensive call in general. This is the contact POST. Show an honest 'try in a bit' to humans. Be quiet toward honeypot hits so you do not train the bot. A cap that hits a whole office is too tight. A cap that never hits is not set.

  4. 04

    Turn captcha on only when the boring layer is broken — and verify it

    Evidence: honeypot is passed, time is passed, token is passed, the cap holds, and the inbox is still full. Then you may put a challenge on. Verify the token on the server. Read the cookie and GDPR posts before you drag a third-party script in. The EAA post is accessibility: a grid is a break you have to be able to defend. Invisible mode is still a script and a call. It is not 'no captcha'.

  5. 05

    Give the filter an interval, not only a go-live

    When you add a field, you add it to the honeypot check if it is the trap. When you change endpoint, you move the check with it. When a wave comes, you tighten the cap — not the badge first. When a human complains about the grid, it is a signal that you turned theatre on too early. The security post is headers. This is the habit in the handler.

“You have stopped spam when the bot fails quietly and the human does not have to prove it is a human.”

Honeypot, cap, captcha and the open inbox — the same word, four jobs

Honeypot is the job most often skipped because it 'feels too simple'. It is the job that most often removes the boring wave. A field bots love, humans do not see, and the server refuses. If the trap catches humans, you have built it wrong — not 'honeypot does not work'. Fix the field. Do not jump to the grid.

The cap is the job you forget until one IP posts a hundred times. The rate-limiting post is login and webhooks. This is the form. A cap without a honeypot also hits the impatient human. A honeypot without a cap lets the clever bot try for the rest of the night. The two belong together. The 429 is yours. A quiet 200 that 'succeeds' toward the bot can be right on honeypot — not on a human you owe an answer.

Captcha is the job you turn on when you have evidence. It is also the job that drags cookies, an accessibility break and a vendor call into a page that was otherwise first-party. The cookie post is the banner. GDPR is the data flow. First-party analytics is what you own after a yes. A captcha is a no to 'we keep it first-party', unless you have chosen that with open eyes. Invisible is still that call.

The open inbox is the job you call 'just mail' when you mean you do not have an endpoint. A `mailto:` is spam in another way. A Zapier URL in the client is an endpoint you have published. A WordPress `admin-post` without a nonce is an open door. Find the POST. Put the trap there. The portal post is when the inbox starts lying as workflow. This is when the inbox starts lying as a feed.

What you should typically do in the situation you are in

The left is what you typically say. The right is the next step we typically recommend.

  • Situation

    The inbox is full, so you put reCAPTCHA on the button and not in the handler

    Next step

    Verify the token on the server, or take the widget off and set honeypot, time and cap first

  • Situation

    The honeypot catches real humans with a password manager or a screen reader

    Next step

    The field is in the way. aria-hidden, tabindex -1, autocomplete off, a name humans do not expect. Not a grid on top

  • Situation

    Spam still arrives, and you have only filtered in React state

    Next step

    Bots hit the endpoint. The trap lives in the handler, or you do not have a filter

  • Situation

    One IP posts relentlessly, and you are considering turning the whole form off

    Next step

    Cap that POST, honest 429. The rate-limiting post is the expensive call. This is the contact field

  • Situation

    You 'just' want captcha because everyone else has the badge

    Next step

    A badge is theatre. Try the boring layer. Turn the challenge on when you have evidence, and read the cookie post first

Checklist before you call it a filter

If you are missing several of these, the sentence is captcha is set — the setup is a badge you are hoping for.

  • You can point at the handler that sends the mail, and at a no that lives there — not only in the UI
  • The honeypot is out of the way for humans and is checked on the server
  • Time and a signed token from render to POST reject the dump that never saw the page
  • A cap per IP/form with a 429 you own — a quiet no to honeypot hits
  • Captcha only after evidence, with server verification, and with the cookie and accessibility reading done
  • The portal post is workflow. Rate-limit is the expensive call. Cookie is consent. This is the bot in the POST

Questions we get again and again

  • Does honeypot still work, or are bots too clever?

    The boring wave is not clever. It fills fields named website and url. It POSTs in the same millisecond the page could be shown. Honeypot plus time plus token removes that wave. The clever bot is what you turn captcha on for — when you have evidence that the boring trap has been passed. Jumping to the grid first is paying the tax for a wave a hidden field would have taken.

  • Is this the same as rate limiting, GDPR, or turning the form into a portal?

    No. The rate-limiting post is the cap on the expensive call — login, search, webhooks. GDPR and the cookie banner are consent and data flows. The portal post is when a message should become a state the customer can see. The security post is headers and admin. This is the bot in the open POST. You can have an honest banner and still an inbox full of prescription links. You can have a portal and still a public contact form that has to be owned. They are five jobs. One does not replace the others.

  • May we use reCAPTCHA or hCaptcha at all?

    Yes, when the boring layer is broken, and when you verify the token on the server. Read the cookie and GDPR posts first: you are dragging in a third-party script and often a cookie. Read accessibility: a grid is a break you have to be able to defend. Invisible mode is still a call and a script. It is a tool. It is not the first field. A badge without a server check is a sign.

  • Should we answer 200 on spam so the bot does not learn from the rejection?

    On honeypot hits a quiet no — the same response as success outwardly, no mail inwardly — can be right, so you do not train the bot. On a human you owe an answer, it is a lying UI. Split the two. The cap 429 should be honest toward humans. The honeypot trap may be quiet. A 500 on spam is your monitoring you have given to the bot.

  • What about WordPress plugins versus a Next.js API route?

    The job is the same: the trap lives where the mail is sent. A plugin that only renders a widget is UI. An API route that does not check the honeypot is open. Nonce, token, time, cap — in the handler. Next.js, WordPress, a form SaaS: the pen is different. If the pen does not write a no on the server, you do not have a filter. You have a button.

If the inbox is a feed, and the human has to prove it is a human

Let us separate a boring filter from a badge you are hoping for.

Honeypot, time and a cap beat a puzzle in front of the people you want to hear from.

Why we wrote this

This is how we think — and it's what we build.

Our insights are about the work we actually do. If this hit something you're working on, there's a concrete service that lines up.