What is credit farming and why do AI apps get hit first?
Credit farming is one person opening many accounts to collect free credits. What it costs, why AI apps get hit first and why the usual fixes break.
Say you run a small AI writing app. It brings in $3,000 a month, and every new signup gets 500 free credits. A credit costs you $0.004 in model spend, so an account that uses all 500 costs you $2. You've been happy with that deal, because two dollars is a cheap way to show someone what your product can do.
Now picture the second of the month. It's 6:40 a.m. and your phone is buzzing on the nightstand. Your model provider has emailed to say you've reached your $2,500 spend limit. Below that are messages from paying customers: the app is throwing errors. One of them has a client draft due at 9.
You open your dashboard, and for a moment it looks like the best night you've ever had. Signups hit a record, about 1,200 since midnight. Then you look at upgrades, and there are none.
The logs explain it. Just after midnight, a script started opening one account every 20 seconds, and every account spent all the credits it was given.
180 accounts an hour × 6 hours 40 minutes = about 1,200 accounts
1,200 accounts × 500 credits = 600,000 credits
600,000 credits × $0.004 = $2,400An illustrative night of credit farming:
That $2,400 is 80% of your monthly revenue, spent in under seven hours. You'd used $100 of your limit before last night, and the script took the rest. That leaves $600 of this month's revenue for your real customers' model use, your hosting and everything else. If you were already losing $1,200 a month, you just lost two months of runway while you were asleep.
None of it comes back. Your provider ran those requests, so it'll bill for them. You can delete all 1,200 accounts this morning and you'll still owe the $2,400.
By 7 a.m. you have to choose. Raise the limit and your customers are back online, but if the script is still running, you're paying for it at $360 an hour. Switch off free credits and the script gets nothing, and neither does the next real person who finds you. Do neither and the people who pay you keep hitting errors. Some will ask for a refund. Others will cancel and never tell you why.
You message a founder friend, who replies with two words: credit farming. So you open a new tab and type what is credit farming.
That morning was a made-up example. This one wasn't:
From our early beta
A team in our early beta woke up to a $6,000 OpenAI bill. Someone had created hundreds of free accounts and used up the complimentary AI credits on every one of them.
What is credit farming?
Credit farming is when one person or script opens many accounts to collect the free credits each new account gets, then spends them all. You meant the offer to be once per person. Farming turns it into once per account, and accounts cost nothing to make. It's multi-accounting aimed at the one part of your product that's paid for in cash.
Look at any single account in that loop and nothing seems wrong. It signed up, received what you promised and used it. The problem only shows up across accounts, and most of your software looks at one account at a time. That's why the obvious fix, one grant per email address, doesn't hold. To that rule, a second address is a brand-new customer.
How credit farming differs from trial farming
Trial farming is the same habit aimed at a trial: someone starts a new free or discounted trial each time the last one ends. They hurt in different ways.
| Credit farming | Trial farming | |
|---|---|---|
| What's collected | Free credits, tokens or generations | Repeated free or discounted trial periods |
| When it costs you | The moment the credits are spent | Over the trial, mostly as subscriptions you never get |
| Can you stop it partway? | Rarely, the credits are already gone | Yes, you can end a trial |
| What you're left with | A bill | A conversion rate you can't trust |
A trial that includes model use behaves like credit farming for as long as it lasts. For the trial side, see how to prevent free trial abuse.
Why AI apps get hit first
One more free user costs most software very little: a database row and some bandwidth. An AI app pays a model provider every time someone presses the button. At $0.004 a credit, your signup offer is $2 of real spend per account, and you've advertised it on your pricing page.
Those credits also go fast. A script can spend 500 in minutes, and once the model has answered, that money is gone. A farmed trial can at least be cancelled on day three. Farmed credits leave you only the invoice.
And what comes out is worth having. Farming a trial gets someone access to software for a while. Farming credits produces finished writing, images or code that they can use today or sell on. So the reward is real and a new account costs nothing. To put a number on your own offer, work out the cost of a free user in an AI app.
Signs of credit farming in your own numbers
You can spot it in numbers you already have.
- Signups are up and revenue isn't. A spike with no rise in upgrades deserves a look, especially in a week you didn't launch anything.
- Free credit spend is climbing, and free users start to cost more than paying ones.
- Credits are all used on day one. Real people try a few things and wander off. A batch of accounts that spend down to zero and never return looks different.
- Email addresses look alike. Read your newest signups. A long run of near-identical addresses is easy to spot.
None of these proves anything about one account, and a good press mention can look like a spike too. They tell you where to go and look. If it's happening right now, start with what to do in the first hour of a signup attack.
Why the usual fixes for credit farming keep breaking
After a morning like that, you start adding rules. They're all reasonable. Each one catches something, and each one comes with a bill of its own.
Email verification and unique emails
Verification proves someone can open that inbox right now. It stops typos and made-up addresses, and requiring unique emails stops the same address signing up twice. Neither tells you whether the person behind a new address has already had your credits. Both are cheap, so keep them, but expect to lose some real people when the confirmation email lands in spam.
Disposable domains, Gmail dots and plus addressing
Blocking disposable email domains catches the well-known throwaway services. New ones keep appearing, so someone has to update that list for as long as you run the app. Say you ship the block on a Friday. By Monday, signups can be arriving from a domain that wasn't on Friday's list. It also turns away people who use a privacy email service.
Gmail ignores dots and delivers plus-addressed mail to the same inbox, so sam.lee@gmail.com, samlee@gmail.com and samlee+app@gmail.com are one mailbox. Treating those as one address catches a casual repeat. Other providers follow different rules, though, so that rule needs upkeep.
CAPTCHA
A CAPTCHA filters out simple scripts, and that's worth having. It doesn't stop a person farming by hand, and paid solving services exist for the scripts it would have stopped. It also adds a hurdle at the moment you most want a stranger to keep going, and it's hardest on people using screen readers. More in is CAPTCHA enough to stop free trial abuse?
Phone numbers and cards
Asking for a phone number or card makes each extra account more work, and it does put off casual repeaters. Determined farmers can buy virtual numbers and prepaid cards. Real people who wanted to try before handing anything over leave, and someone without a credit card can't get in at all. You also pay for every text, and you never see the signups a card form cost you.
IP limits and VPN blocks
One signup per IP address stops a script running from a single connection. It misses anything that changes address, and it blocks the second person in an office, a dorm or a home, because they share one. Many mobile customers do too. VPN blocks catch the services on your list and turn away people who use a VPN for privacy or work, and that list needs keeping up to date or paying for.
Rate limits, usage caps and shorter trials
Rate limits work per account or per address. Each farmed account only used what you gave it, so a limit on one account has nothing to object to. A limit per address helps until the traffic moves, and it hits your busiest real users first.
Usage caps are the fix to keep. Give free users their own budget, apart from your paying customers', and a bad night has a ceiling that doesn't take your customers down with it. The farmer still spends everything up to the cap, and once it's full, your real new signups get nothing.
Shorter trials don't help when credits are spent in minutes. Smaller grants do cut the damage: 100 credits instead of 500 is $0.40 an account, or $480 across the same 1,200 accounts. They also give real people less reason to stay, and a script can just open more accounts.
Manual review
A person reading new signups can notice things a simple rule won't. At an illustrative 30 seconds per account, those 1,200 accounts are 10 hours of work, and the money was spent before you'd have started. You'll also close a few real accounts by mistake, and those customers rarely write in to tell you.
Stack all of this and you're maintaining lists and rules on top of the product you meant to build, and none of it can reliably tell you whether the person asking has already had the credits.
Evaluate the claim before the credits go out
Whether this claim looks like a repeat is the question AbuseZero exists to answer, and you don't have to build your own abuse prevention system to ask it. AbuseZero flags abuse that single checks miss. Its multi-layer AI risk analysis examines 140,000+ signals and detects linked claims. Those signals are read together, never one at a time. There's a proprietary algorithm too, and AI checks the claim again. AbuseZero checks mailbox aliases and disposable domains. It recognizes repeat devices across new accounts, even when email addresses keep changing, and detects repeated payment fingerprints across users, even when new accounts and emails hide the connection. You won't put that together over a weekend.
It's one API call. Before you grant the credits, you ask for an evaluation and get a clear allow or deny, using rules you control. If you'd like to see the effect first, Observe mode shows you what would be denied before anyone is blocked. The quickstart walks you through your first benefit.
Frequently asked questions
Can I get the money back after credit farming?
Usually not. Model spend is charged when the requests run, so deleting the accounts only stops any credits that are still unspent. Ask your provider anyway.
How do I stop credit farming?
Put free users on a spend cap of their own so a bad night has a limit. Then evaluate each credit claim before you grant it. AbuseZero returns allow or deny for each one. For repeat grants and top-ups, see how to stop repeat credit abuse.
Do small AI apps really get farmed?
Size doesn't change the reward. Your 500 free credits are worth the same as anyone else's, and a small team is less likely to be awake at 3 a.m. See why new SaaS founders wake up to surprise bills.
Protect your free credits before the next bill
Start with Builder
Pay as you go, from your first claim.
See pricing
Builder, Pro, Scale and Enterprise.
Talk to us about Enterprise
High-volume apps and custom terms.