How do I stop people abusing my free trial?
Cap the cost, not the person. Nearly every exploited free trial was protected by a limit attached to an account, and an account can always be created again — so the visitor who hits your ceiling simply starts over. What holds is a hard ceiling on what one trial can consume, a smaller daily ceiling underneath it so a single sitting cannot drain the whole allowance, and canonicalising email addresses to close the lazy routes. Treat whatever still gets through as a marketing cost with a known maximum, because past that point prevention costs more than the abuse does.
Decide which problem you actually have
Abuse hides two problems that want opposite fixes. One is cost: a free account consumes something you pay for — model calls, compute, storage, an email quota — and enough free accounts become a real bill. The other is lost revenue: someone keeps restarting the trial instead of subscribing. For most small products the first is genuine and the second is mostly imagined, because a person willing to juggle inboxes to avoid a small monthly fee was rarely going to pay it.
The distinction decides how much engineering is justified. Capping cost is metering, and metering is cheap and certain: you choose the most a stranger can ever spend, and you enforce it. Protecting revenue means establishing identity, which on the open internet is unreliable, never finished, and falls hardest on the honest — shared offices, family devices and privacy-conscious customers all look like duplicates. Build the metering first, then see whether you still care about the rest.
| Defence | What it actually stops | What it costs you |
|---|---|---|
| Canonicalising email addresses | One inbox spelled several ways — dots, plus-addressing | Almost nothing, as long as both spellings stay searchable |
| Blocking disposable-inbox domains | Throwaway addresses, some of the time | A list to maintain, plus the occasional real customer on an odd domain |
| Asking for a card up front | Nearly all casual restarting | A large share of your signups. This is a funnel decision, not a fraud one |
| A hard spend ceiling per trial | The bill itself, whoever the person turns out to be | Engineering once, then nothing |
| A daily ceiling under the total | One sitting draining the entire allowance | A real constraint on your most enthusiastic evaluators |
| Device or browser fingerprinting | Determined repeat signups, imperfectly | Privacy exposure, false matches, and permanent upkeep |
Read that as an order of work rather than a menu. The first row and the fourth are close to free and stop most of the damage. The last row is the one founders reach for first, and it belongs last.
The loophole is the word account
A limit that lives on the account is the standard design, and it is the one that gets walked around, because the account is the cheapest thing in the whole system to replace. So before you ship any ceiling, answer one more question: what does someone do in the first minute after they hit it?
We can answer that from our own logs. AgentCeres — the AI Growth Officer at agentceres.com — runs a card-less trial with a hard spend ceiling, and the ceiling does exactly what it was built to do. One customer was building a page with the agent and kept asking for corrections; each correction republished the page, and he reached a separate daily publishing limit. He was told the quota would reopen the next day. Instead he registered again with a second address and started the same work from the beginning. We know it was the same person because the first message he sent the new workspace was word for word the message he had sent the old one, sixty-six minutes earlier.
Nothing malfunctioned there. The limit was correct and the message was honest, and the outcome was still a second free workspace redoing the same work — plus a customer who had lost an hour, which is the more expensive half. A limit stated per account is a limit per account. If the answer to what happens next is that they sign up again, you are maintaining a speed bump and calling it a wall.
Cap the day, not just the trial
The change that helped us most was not a better identity check. It was a second, smaller ceiling: a maximum for one day, sitting underneath the maximum for the whole trial. An allowance with only a total can be spent in an afternoon, which is the pattern behind both the surprising bills and the disappointed evaluator who has nothing left on day two.
We measured before choosing the number, and the measurement is the part worth copying. Of 376 trial workspaces, 346 had never spent as much in a single day as the daily ceiling we were considering. The thirty that had were two groups: people genuinely building something on their first day, and every cluster of repeat signups coming from one device. So the cap would be invisible to the overwhelming majority of trials and would land on exactly two cohorts — one we want to serve better, one we do not.
That second group is the uncomfortable one, and it is worth saying plainly: a daily cap paces the person you most want to impress exactly as it paces the person recycling accounts, and no version of the rule can tell them apart at the moment it fires. We took the trade knowingly. The ceiling is generous enough that the measured population barely touches it, it resets without anyone doing anything, and the enthusiastic builder who does reach it has it back the next day.
- A ceiling for the whole trial, plus a smaller one for a single day. The day rolls over by itself, so there is no pause anyone has to remember to lift.
- Email addresses canonicalised at signup, so one inbox cannot quietly become several accounts.
- No device fingerprinting. A shared budget per device was designed, written down, and shelved.
- Which means a fresh account still gets a fresh daily allowance. That is a bounded cost we have measured, not an open one.
That last line is the whole point. With both ceilings in place you can multiply the worst case out — the daily figure times the length of the window — and decide whether the number frightens you. Ours does not, so the remaining leak is budgeted rather than fought. If yours does, the cheapest next move is almost never fingerprinting: it is lowering a ceiling, shortening the window, or asking for a card and accepting the smaller funnel that comes with it. Which of those fits depends on what the trial is for, and whether a trial is the right model at all is worth settling first.
FAQ
- Should I just require a credit card to stop trial abuse?
- It is the most effective single move available and also the most expensive, because it is not really a fraud control — it is a change to your funnel. A card up front removes most casual restarting and converts by default when the window closes, and it costs you a large share of signups, including people who would have bought after trying. Decide it as a growth question rather than a security one: free trial or freemium is where that trade-off actually lives.
- Is blocking disposable email domains worth the effort?
- It catches a slice and it is never finished, because the domain lists move faster than you will update them. The cheaper neighbour is canonicalising the addresses you already accept: treating dotted and plus-addressed spellings of one inbox as the same account closes the laziest route at almost no cost. Do that first, keep both the typed and the canonical form searchable so support can still find a customer by what they entered, and treat domain blocking as optional.
- When should a daily allowance reset?
- Whenever your usage records can actually prove, which may not be the answer you want. Ours resets at midnight UTC because the spend ledger buckets by UTC day and nothing finer, so a customer in São Paulo gets a fresh allowance at nine in the evening. That is a real compromise, and the way to stop it becoming a support thread is in the wording: tell people the allowance returns in about so many hours rather than naming a clock time, because a server-rendered page does not know what time zone the reader is in.
- How do I know a signup is the same person as a previous one?
- Usually you do not, with certainty, and you should be careful what you attach to a guess. Every signal available — a device, a network, a payment instrument, the timing of the signup — produces false matches in shared houses and offices. The most convincing evidence we have ever had was behavioural rather than technical: two workspaces whose first message was identical, minutes apart. That is good enough to understand a pattern and to size a cost. It is not good enough to lock someone out, which is another reason to spend the effort on the ceiling instead.
Want this done for you?
AgentCeres is a managed AI marketing team — specialists draft the work, you approve what ships. 14-day free trial, from $39/month.