Back to Blog

How to stop onboarding reward farming

Setup rewards pay out before anyone looks. What onboarding reward abuse costs per claim, which basic fixes break, and what to change before the next grant.

Say you run an AI app builder. Someone describes the app they want, your agent writes and deploys it, and every build step uses one credit. Your Pro plan is $50 a month for 1,000 credits, and each credit costs you about $0.03 in model usage. This is a scenario, and those numbers are assumptions for you to swap for your own.

New accounts get 50 credits, which won't finish much of anything. So you ship a setup checklist: verify your email, pick a template, build your first app, publish a preview, invite a colleague. Finish all five and you get 800 build credits. That's enough to finish something a first-time builder would pay to keep, and it's a lot of money for one claim:

  • 800 credits × $0.03 = $24 of model usage, paid by you
  • $24 ÷ $50 = 48% of what a Pro customer pays you in a month

You're careful about it, so you cap the program at 200 rewards a month. 200 × $24 = $4,800, which is what you can spare.

At 10:37 a.m. on the 3rd, a message comes in from someone who found you through a newsletter: "I did all five steps on my lunch break and now it says the reward has run out. Your email said 800 credits." You open the rewards table. The cap was reached at 5:26 p.m. on the 2nd, less than two days into the month. You click through the accounts. Verified emails, five steps out of five. You open ten of the published previews, and they're all the starter template with the title unchanged. By the end of the day you've counted 180 accounts like that.

This is onboarding reward abuse, and by the time you find it, it's over. Every one of those balances is already at zero:

  • 180 accounts × $24 = $4,320 in model usage, already billed
  • 180 accounts × 800 credits = 144,000 credits, spent in about two days

You could reverse the balances, except there's nothing in them. Taking credits back only works on credits nobody has used, and your model provider charged you when the builds ran.

Those builds also ran on the capacity your paying customers use. Your 400 Pro customers burn about 8,000 credits on a normal day. The farmed accounts added 72,000 a day (144,000 ÷ 2), nine times your usual load, so you hit your provider's rate limit and paying customers' builds started failing halfway through. Three of them asked for their month back, which is another $150. The $4,320 on its own is 21.6% of the $20,000 those 400 customers pay you each month.

Then there's the group you built this for. About 170 real people finish your checklist in a normal month, and only 20 of them got in before the money ran out. The other 150 did every step and got an apology banner. Say 30% of people who get the reward go on to Pro, and 10% of people who finish and get nothing do:

  • 150 people × (30% minus 10%) = 30 customers who never subscribe
  • 30 × $50 = $1,500 a month
  • $1,500 × 12 = $18,000 a year, from one month of signups

The cap resets on the 1st. You have until then to choose: kill the checklist that brings you your best customers, cut the reward and go back on what your emails promised, or leave it alone and hand over another $4,320.

Here's one claim from start to finish. Nothing in it asks whether the reward should go out at all.

Finishes step 5800 credits addedSpends all 800 credits800 build steps$24 on your billNothing left to take backNew accountYour serverModel providerNew accountYour serverModel provider
Finishesstep 5800 creditsaddedSpends all800 credits800 buildsteps$24 onyour billNothing left to take backNewaccountYourserverModelproviderNewaccountYourserverModelprovider

Why onboarding reward abuse costs so much per claim

A free trial gives someone access and then waits to see what they do. A setup reward pays a lump sum the moment an account looks its most engaged, and your activation chart counts that as a win. Those 180 accounts just gave you your best activation month ever.

The checklist itself is the weak part. Every step had to be something your code could confirm, which makes every step something code can complete. A step that takes a first-time builder twelve minutes costs a script next to nothing, so a reward designed to pay for effort goes first to the accounts that spend none.

Speed makes that worse, because credits can be spent as fast as your build queue accepts them. It's the same economics as credit farming, at $24 a claim instead of $1.50 in starter credits (50 × $0.03).

The obvious fix is a harder checklist with more steps. A script doesn't get tired of step eight, and people do. Say 384 real people start the checklist each month and 85% get through each step. Five steps leaves 170 of them (384 × 0.85, five times over). Eight steps leaves 105. You've turned away 65 real builders and not one scripted account.

Bars are real people who finish. The line is scripted accounts.

Finishers by checklist length, illustrative050100150200250345678AccountsSetup steps
Finishers by checklist length,illustrative0100200345678AccountsSetup steps

Basic fixes for onboarding reward abuse, and how each one breaks

Each of these catches something. Start with email. You already have verification, because it's step one of your checklist, and all it proves is that someone can open an inbox. A unique-email rule has a similar limit, since it compares text and one mailbox can have many spellings. Gmail ignores dots, so sam.lee@gmail.com and samlee@gmail.com are the same inbox, and mail to sam.lee+build@gmail.com lands there too. That's how email works.

Basic fixWhat it catchesWhat it missesWho it wrongly blocksWhat it costs to keep working
Email verification and unique emailsTypos, dead addresses, the same address twiceAny real second inboxPeople whose code lands in spamDeliverability work
Disposable-domain blocksKnown throwaway servicesNew domains and ordinary mailboxesPeople using privacy forwardingA list someone updates forever
Gmail dots and plus addressingOne inbox written several waysA second real mailboxSomeone who keeps a plus address for a second projectUpkeep as mail services change
CAPTCHA before the rewardSimple scriptsSome automation, and anyone working by handPeople using screen readersFees and lost finishers
Phone or card requirementCasual second accountsAnyone with another number or cardFirst-time builders who aren't sure yet, people without a reliable mobile numberA charge per text, fewer finishers
One reward per IP addressRepeats on one connectionAny other networkA no-code class on campus Wi-FiExceptions and tickets
VPN blocksKnown VPN rangesEverything off your listDevelopers on company VPNsAnother list to maintain
Rate limits on buildsBursts from one account180 accounts each under the limitA real builder on a fast streakTuning per route
Daily cap on reward creditsOne account burning it all at onceAccounts that can wait a few daysSomeone building all weekendTickets from real builders
Smaller rewardSome of the cost per claimThe empty poolEvery honest finisherFewer people reach Pro
Manual reviewWhat a person can spotAccounts that look completeWhoever looks odd to a tired reviewerHours every month and late rewards

Run the numbers on the smaller reward, since it's the tempting one. At 200 credits a claim costs $6, so a full month of 200 rewards costs $1,200 instead of $4,800. But 180 scripted accounts still get there first, 150 real builders still find an empty pool, and the 20 who got in have a reward too small to finish an app.

At 4 minutes a claim, manual review of 200 claims is 800 minutes, more than 13 hours a month, and what's in front of you is a verified email and five finished steps. You'd like to know whether you've already paid this person, and nothing on that screen says so.

Stack the rest and your checklist has grown a puzzle, a phone number and a card. Those are extra steps too, so go back to the chart.

What to change before the next reward goes out

Pay the reward in parts. Give 200 credits at step five and the other 600 over the next three weeks, as the account comes back to build. An account that takes the first part and vanishes costs $6 instead of $24, so 180 of them cost $1,080 instead of $4,320.

Give reward traffic its own budget and limit. Use a separate provider key or queue with a hard cap, so farmed builds can't touch paying customers' capacity.

Don't run it first come, first served. If the month's budget runs out, hide the offer from new starters. Never take it away from someone on step four.

Decide what a no looks like. Let people finish setup anyway, and tell them plainly what they didn't get and how to reach you.

Those steps shrink what a bad claim costs, but a new account arrives with no history in your tables and still collects the first 200 credits at step five. The question your checklist never asked is still open: should this claim go through?

Ask before the credits move

You don't have to build your own abuse prevention system to get that answer. Step five completes, your server makes one API call to AbuseZero, and a clear allow or deny comes back under rules you control.

Every row in that table is a single check, and that's where it runs out. With multi-layer AI risk analysis, AbuseZero examines 140,000+ signals, detects linked claims and flags abuse that single checks miss. The accounts that cruised through your five steps are exactly what it's for: it blocks spammy bots, browser automation and malicious scripts while allowing genuine AI agents and assistants. We built a proprietary algorithm for this, with AI that checks the claim again. Creating a fresh account isn't enough to erase previous activity. AbuseZero recognizes repeat devices across new accounts, even when email addresses keep changing, and it checks mailbox aliases, disposable domains and mail routing. It also sees beyond an IP address with VPN, hosting, network reputation, Tor and proxy intelligence.

Make the setup reward its own benefit with an eligibility of once, and start it in Observe mode to see what would be denied before anyone is blocked. The quickstart covers your first one.

Frequently asked questions

What is onboarding reward abuse?

It's someone collecting a reward for finishing setup more times than you meant to give it, usually across several accounts and often with scripts doing the steps.

Should I only pay setup rewards after someone upgrades?

Part of it, if you can. A reward that needs a payment costs a non-payer nothing, but it stops working on people who aren't sure yet, and they're who the checklist is for.

Does a second account always count as abuse?

No. Someone might build at work and at home, and an agency may open a fresh account for each client. Decide what you mean to reward and write it down. AbuseZero applies the eligibility you set when it evaluates each claim.

Keep your setup reward for real builders

Keep reading

Copying your prompt…

Copied the to your clipboard. Give it to your coding agent & it will handle the rest.