Is CAPTCHA enough to stop free trial abuse?
CAPTCHA filters crude bots, but the person repeating your trial is human. What it catches, what it misses, and how it compares with other signup checks.
Say a batch of fake signups ate through your free trials last month, so you added a CAPTCHA to the signup form. It took an afternoon. You closed the ticket on a Friday, told the team it was handled and went back to the roadmap.
Six weeks later, on a Tuesday at 10:26 a.m., you open the monthly numbers and trial costs are back where they started. You scroll the signup list looking for bots and can't find any. What you find is 40 accounts created on a Thursday three weeks ago, between 1:05 p.m. and 5:20 p.m., each with a different email address and a solved CAPTCHA. You only notice them because whoever made them typed the same display name into all 40.
You can tell yourself that person was never going to pay list price. The next number is harder to wave away. Say each of those trials burned $38 in model and compute costs (every figure in this scenario is illustrative, so swap in your own). That's 40 × $38 = $1,520 you actually paid.
Now the Wednesday after that Thursday makes sense. At 4:47 p.m. your shared model quota ran out and paying customers got errors for 40 minutes. It looked like a traffic spike. You raised the quota, answered messages until late and refunded two customers $99 each. A third customer, who pays for 12 seats at $1,188 a month, wrote, "Is this going to happen again?" You told them no, and you believed it when you wrote it. Add that $198 in refunds to the $1,520 and one afternoon has cost you $1,718 in cash.
If you spend $6,000 a month, that's $200 a day, and $18,000 in the bank was 90 days of runway. It's now $16,282, or about 81 days. Those 40 accounts were also in the growth number you sent your investors.
Meanwhile every honest person who signed up in those six weeks did an extra step to protect you from someone it didn't stop. So you're back to a decision you thought you'd made. Cut the trial to seven days and real users lose the time they needed. Ask for a card and watch signups fall. Change nothing and pay for the next afternoon. The ticket you closed six weeks ago is open again, and this time you don't know what to write in it.
Why CAPTCHA doesn't stop free trial abuse
A CAPTCHA asks one question: is there a human here right now? The person abusing your trial is a human, so the correct answer is yes. It gives that same yes on the first trial and the fortieth, because it has no memory of who already took one. Nobody built it to have one.
"Is this a human?" and "Has this person already claimed this trial?" are different questions, and a yes to the first tells you nothing about the second. Your trial terms depend on the second. Most signup forms only ask the first, so a CAPTCHA can work exactly as advertised while your trial costs climb back up.
What the research says about bots and CAPTCHA
The other half of the comfort is that CAPTCHA at least stops the bots. In An Empirical Study & Evaluation of Modern CAPTCHAs (Searles et al., USENIX Security, 2023), 1,400 paid participants solved 14,000 CAPTCHAs. The authors set those human results beside bot results from earlier published attacks and found that "bots can outperform humans, both in terms of solving time and accuracy."
Two types show the gap clearly. On distorted text, people took 9 to 15.3 seconds and agreed on the answer 50 to 84% of the time, which the study used as its stand-in for accuracy. The attacks it cites report under 1 second and up to 99.8% accuracy. On the one-click checkbox kind, people took 3.1 to 4.9 seconds, against a reported 1.4 seconds for a bot.
How to read this chart
Slow and fast are the two ends of the human range the study measured. The bot bars come from attacks published between 2014 and 2020 against the CAPTCHAs of their day, so this compares separate papers and isn't a head-to-head race. The text bot took under 1 second and is drawn at 1.
Those numbers also show what a repeat claim costs the person making it. If your CAPTCHA is the one-click kind, 40 solves at the slow end of that range take 40 × 4.9 = 196 seconds. A little over three minutes of CAPTCHA bought $3,960 of product.
What CAPTCHA is actually good for
Keep your CAPTCHA, as long as you know what it's for. It's cheap to add and it filters crude scripts that hammer a form with no real browser behind them. On contact forms, comment boxes and anything else a spam wave can flood, that relief is real. It also buys you time in the first hour of a signup attack.
The cost lands on the people you wanted to let in:
- People who use screen readers or can't make out distorted images. A visual puzzle can lock them out, and audio fallbacks are often hard to use too.
- People on privacy tools. Privacy browsers, ad blockers and VPNs can bring harder challenges, or a loop that never ends.
- AI agents working for a real customer. An agent that signs up on someone's behalf can't always get through a puzzle, and that customer is worth having.
- Everyone else. Each honest signup now has one more step.
In the same study, participants asked to solve CAPTCHAs while creating accounts were 120% more likely to abandon than participants simply asked to solve CAPTCHAs. They were paid study workers and not your visitors, and the authors note possible confounding factors, so this isn't a measured loss in signup conversion. It's a reason to measure yours.
Say 3,000 people a month start your signup form and the CAPTCHA step loses 3% of them, an illustrative rate. That's 90 people. If 5% of people who finish signing up go on to pay $99 a month, you've given up 90 × 0.05 × $99, about $446 of new monthly revenue, every month.
CAPTCHA vs other free trial abuse checks
After CAPTCHA, most teams reach for more basics. None of them was built to know who already claimed.
| Check | What it catches | What it misses | Who it wrongly blocks | What it costs to keep working |
|---|---|---|---|---|
| CAPTCHA | Crude scripts | Repeat humans and capable bots | Screen reader users, privacy tool users, AI agents | Watching drop-off |
| Email verification | Mistyped and made-up addresses | Anyone with a second real inbox | People whose mail lands in spam | Support when codes go missing |
| Unique emails, Gmail dots and plus addressing | The same address twice, or one Gmail inbox written another way (Gmail ignores dots and anything after a plus) | A different address, which is free | People who sort their mail with plus addresses | A rule for each mail provider |
| Disposable domain blocks | Known throwaway domains | New domains, which appear constantly | People on small mail providers | A list someone updates forever |
| Phone verification | People without a second number | Anyone with one | People without a mobile number | A fee for every text you send |
| Card requirement | People unwilling to enter a card | Anyone with another card | Students and anyone who wants to try first | Signups you never see |
| One per IP | Repeat signups from one connection | The same person on another connection | Offices, schools and homes that share an address | Exceptions and support tickets |
| VPN blocks | Known VPN ranges | Anyone not on one | Privacy-minded customers and remote staff | Another list that goes stale |
| Rate limits and usage caps | Bursts from one source, and runaway use on one account | Slow signups like 40 in an afternoon, and new accounts with new caps | Launch days and your heaviest real users | Retuning limits after every incident |
| Shorter trials | Total free usage per claim | Repeat claims, which simply come sooner | People who need longer to see the value | Lower trial conversion, which you'll need to test |
| Manual review | Patterns a person can spot | Whatever arrives at 3 a.m. | Real people left waiting | Someone's hours, every week |
Each row is a rule somebody on your team now owns. The disposable list you updated on Monday is behind again by Friday. The one-per-IP rule you shipped before a long weekend turns away a whole design school on Tuesday, and you hear about it from a professor. Stack all eleven and the forty-first trial can still pass every one, while each honest signup goes through all of them.
Three things are worth doing with the CAPTCHA you already have:
Keep it where crude scripts actually hit you. On a page nobody's attacking, it only costs you honest visitors.
Measure the step. Compare signup completion with and without it for a few weeks, and work out what each free user costs you.
Count repeat trials separately from bots. They're different problems, so a drop in one tells you nothing about the other.
Is this a human, or is this claim a repeat?
Everything in that table looks at the request in front of it: one address, one connection, one person clicking a box. A repeat claim is about history. Spotting one has to reach past the request in front of you, across new accounts and new email addresses.
Whether a claim looks like a repeat is the question AbuseZero answers. CAPTCHAs focus on whether someone is human. AbuseZero evaluates whether someone is actually eligible to claim your trials, credits and other benefits, even if they're a real person. Its multi-layer AI risk analysis examines 140,000+ signals, detects linked claims and flags abuse that single checks miss, and the table above is a list of single checks. Signals are read together, never one at a time, and AbuseZero's proprietary algorithm, with AI that checks the claim again, was built for exactly this question. AbuseZero recognizes repeat devices across new accounts, even when email addresses keep changing. It detects repeated payment fingerprints across users, even when new accounts and emails hide the connection. It checks mailbox aliases and disposable domains. It blocks spammy bots, browser automation and malicious scripts while allowing genuine AI agents and assistants. Creating a fresh account isn't enough to erase previous activity.
You make one API call and get a clear allow or deny, using rules you control. Observe mode shows you what would be denied before anyone is blocked. You don't have to build your own abuse prevention system.
Frequently asked questions
Does CAPTCHA stop free trial abuse?
It stops some of it. A CAPTCHA blocks crude scripts, but a person repeating your trial with a new email can solve it every time.
Can bots solve CAPTCHA?
Some can. A 2023 USENIX Security study gathered published attacks that solved several common CAPTCHA types faster than the people in its own tests.
Should I remove CAPTCHA from my signup form?
Not automatically. Keep it if crude scripts are hitting the form, and measure how many honest people it costs you.
What's the difference between CAPTCHA and abuse prevention?
A CAPTCHA asks whether someone is human. AbuseZero evaluates whether a claim for a benefit should go through.
Protect your next trial
Start with Builder
Pay as you go. Protect your first trial today.
See pricing
Builder, Pro, Scale and Enterprise.
Talk to us about Enterprise
High-volume signups and custom terms.