How to Protect Your Website from Bot Signups and Fake Email Confirmations
Explore my NLP research and published research.
I woke up to a flooded inbox again. If that sounds familiar, you are not alone. In my earlier post, The Truth About Bot Clicks: Are AI Agents Bypassing Your Email Verification?, I described a sudden wave of corporate-looking subscribers and explained why automated security gateways and AI agents were most likely the ones clicking the confirmation links.
That post was about diagnosis. This one is about the cure. Below are the concrete defenses I added to my own site afterwards, and you can apply most of them to any website with a newsletter, a signup form, or a double opt-in flow.
A quick privacy note: I am deliberately not publishing any of the email addresses or company names involved. The people whose addresses were used are almost certainly unaware that their inbox security tools clicked the links on their behalf. This is about hardening systems, not naming individuals.
First, understand what you are up against
The flood was not one attack. It was two stages working together:
- Stage 1 - the form flood. A botnet submitted hundreds of signup requests, mixing random free-mail addresses with targeted corporate addresses. Every request came from a different IP but carried the same fake referrer and stale browser user agents.
- Stage 2 - the automatic confirmation. Mail security gateways at the receiving organizations scanned the confirmation emails and followed the links automatically. The bots could never have completed double opt-in on their own; the defenders did it for them.
The numbers made the pattern obvious. In one recent window on my site, 145 free-mail signups produced a single confirmation, a rate below one percent. Meanwhile, 40 corporate-looking signups produced 12 confirmations, a rate of thirty percent. Every confirmed corporate domain I checked used Microsoft 365 / Outlook mail protection. That is not a marketing miracle. It is automated URL scanning.
The lesson is simple: your signup form and your confirmation flow are two separate attack surfaces. You need to defend both.
1. Make confirmation a two-step, POST-based action
This is the single highest-value fix, and it takes very little effort. A bare link in an email is a GET request, and link scanners, preview generators, and AI agents follow GET requests automatically.
So stop letting a GET request activate an account. Instead:
- The emailed link opens a confirmation page (a GET that changes nothing).
- The page contains a button that submits a POST request, and only that POST sets the subscriber to confirmed and active.
Automated scanners follow links but do not fill in and submit forms. This one change alone neutralizes the entire class of auto-clicking gateways. It also protects real readers from accidentally subscribing just by previewing a link.
2. Plant a hidden canary link
Add an invisible link to every confirmation email, hidden with CSS such as display:none or a zero-height container. Human readers will never see it, but automated scanners that follow every link in the message will.
Point that link at an endpoint that records a timestamp and returns a harmless 1x1 transparent image. If the canary is clicked before the subscriber confirms, you know the email was processed by a machine. You can then flag the sign-up automatically, or require an extra step before it is treated as genuine.
Think of it as a tripwire: it does not block traffic, but it tells you exactly which confirmations cannot be trusted.
3. Add a privacy-friendly CAPTCHA
Cloudflare Turnstile is a good choice because it is free, does not track visitors with cookies, and usually verifies people with an invisible check rather than a puzzle. Put the widget on every form that can create an account or send an email.
One practical tip: make verification fail open when your keys are not configured. That way your signup form keeps working during development or if you have not set up Turnstile yet, and you simply flip it on in production by adding the site key and secret key as environment variables.
4. Rate-limit by the real client IP
If your site is behind Cloudflare or another CDN, REMOTE_ADDR will show you the CDN's IP, not your visitor's. Use the CF-Connecting-IP header (or your provider's equivalent) so every bot in the network gets counted separately.
Store counters in the database or a shared cache, not in a single process's memory. If you run several application workers, an in-memory counter will be bypassed as requests are spread across them. A simple table keyed by a hash of the IP, with a window of a few attempts per ten minutes, is enough to stop the bulk flooding.
5. Record click metadata and flag fast confirmations
When a confirmation finally happens, capture the time between sending the email and the click, along with the IP address and user agent. Then apply a simple rule: a human rarely opens an email and confirms it within a minute of delivery. Anything faster is suspicious.
In my case, the suspicious confirmations landed 18 to 50 seconds after delivery, from modern browser user agents with no referrer, while the original signup requests came from stale browser versions with a fake referrer. Two completely different client profiles in the same flow. That mismatch is a strong signal.
Store the reason for each flag, such as "confirmed 22 seconds after delivery" or "canary link was followed", so you can review decisions later rather than relying on an opaque score.
6. Segment and hold flagged subscribers
Do not delete flagged records outright. Keep them, but exclude them from broadcast emails until they prove they are real. A flagged subscriber who later opens a message, clicks a non-canary link, or replies can be cleared manually or by a rule you define.
This protects two things at once: your delivery reputation, because you are no longer mailing addresses that nobody asked to receive messages, and your analytics, because fake confirmations no longer inflate your active subscriber count.
7. Rotate tokens and expire stale links
Every time you send a fresh confirmation email, generate a new confirmation token and a new canary token, and reset the metadata from the previous attempt. This prevents an old link that a scanner cached from being reused later.
Give confirmation links a short lifetime, such as 72 hours, and run a cleanup task that deletes unconfirmed records past that window. Fewer pending records means a smaller attack surface and a cleaner database.
8. Watch your logs for patterns, not people
Your access logs are the best early-warning system you have. Look for aggregate signals: many POST requests with the same referrer, a cluster of outdated user agents, or confirmation clicks with no referrer at all. In my investigation, every abusive signup pointed to the same blog post as its referrer, which made the campaign easy to spot.
When you investigate, work with counts, domains, and user-agent families rather than individual addresses. You usually need the pattern, not the person, and keeping personal data out of your analysis is both safer and cleaner.
Layered defense beats a single fix
None of these controls is perfect on its own. CAPTCHA can be solved, rate limits can be spread across a botnet, and not every mail gateway follows the canary. But stacked together they turn an expensive problem into a cheap one: the bots fail at the form, the scanners fail at the confirmation page, and anything that slips through is flagged before it reaches your mailing list.
The whole set took a single afternoon to implement and cost nothing beyond existing infrastructure. If you run a site with a double opt-in flow, I would start with step one today, add the canary and rate limiting this week, and switch on Turnstile when you have a moment to configure the keys.
Bot traffic is not going away, and AI agents will only get better at clicking links. But you do not have to fight fire with fire. A few thoughtful layers are enough to keep your subscriber list honest and your inbox quiet.
If you missed the original analysis of what was happening, read The Truth About Bot Clicks: Are AI Agents Bypassing Your Email Verification? first. It explains the "why" behind the "how" above.
Related Articles
Displaying the BIR Registration Seal Badge
The Bureau of Internal Revenue requires the Registration Seal Badge under Revenue Memorandum Circular (RMC) No. 38-2026, which was issued on April 29, 2026. This circular mandates all persons engaged …
Read More →We May Be Looking for the Skills Gap in the Wrong Place
Studies and congressional reviews in the Philippines have pointed to a persistent concern: many K–12 Senior High School graduates enter tertiary education needing additional support in foundational knowledge and skills. …
Read More →AI Just Designed 16 New Viruses. Should We Be Worried?
Scientists just used artificial intelligence to design 16 new viruses! These viruses do not exist in nature. The AI invented them from scratch. You can read the latest reports from …
Read More →Subscribe to Updates
Get notified about new blog posts, AI insights, and digital transformation strategies.
We respect your privacy. Unsubscribe at any time.