Referral incentives that work for freemium AI apps
credits, generation limits, early features and cosmetics: rewards a freemium AI app can afford, with accounting too simple to farm.
last updated 16 November 2026
a freemium AI app has no cash margin to give away, so pay referrals in product: credits, generation limits, early feature access and cosmetic identity. price them in inference cost, not sticker value, make the reward two-sided so the invite reads as a gift, and keep the accounting simple enough that farming it is not worth anyone’s evening.
Pay in product, not in money
cash rewards are the default because they are easy to reason about, and they are wrong for freemium twice over. first, you do not have the margin: a free user has paid you nothing, and a cash bounty on their friends is a bet your future conversion rate cannot always cover. second, cash reframes the recommendation as a commission, which is the fastest way to make users stop sending it. product rewards flip both problems. your marginal cost is inference, which is real but sits far below the value users assign to the same credits, and a product reward keeps the share inside the relationship: i am giving you more of a thing i like, not collecting a fee on your signup.
The four currencies
| currency | what it costs you | who it moves |
|---|---|---|
| credits or extra generations | inference only | heavy free users who hit the ceiling |
| raised limits for a period | inference, time-boxed | users mid-habit who feel the cap weekly |
| early feature access | nothing marginal | power users and your loudest fans |
| cosmetic identity | nothing | anywhere status is visible: profiles, badges, styles |
the table is a menu, not a stack. pick the one that maps to what your users already run out of, which you can read from paywall taps and support tickets. an app whose free tier never gets hit has no credit economy to reward in, and should look at features or cosmetics instead. and time-box whatever you pick: a permanent limit raise is a plan change you granted forever for one invite, while a month of it is a taste that converts.
Two-sided beats one-sided
give something to both the sender and the friend. the reason is not generosity, it is what the message says: a one-sided reward makes the invite an extraction, i earn when you sign up. a two-sided reward makes it a gift, we both get a month of pro if you try this. the receiving side also does mechanical work, because the friend’s reward is a reason to accept a link they would otherwise ignore, and acceptance is where most referral funnels actually die. weight the split however your economics allow, but the friend’s side should never be zero.
the friend’s reward has one extra design constraint the sender’s does not: it has to land immediately and visibly on first open. a credit grant buried in a settings page might as well not exist. show it in the first session, name who it came from, and let the friend spend it before your paywall says hello. the reward doing its job is the difference between a referred install and a referred user.
Keep the accounting too simple to farm
referral fraud is an economics problem before it is a detection problem, and simple rules remove the economics.
grant rewards on activation, never on signup. the rewarded event should be the friend reaching first real value, which a script cannot fake cheaply. cap total referral rewards per account, so the best case for a farmer is small and bounded. keep every reward non-convertible: nothing that turns into money, gift cards or anything resellable, because the moment your reward has street value you have created a job and someone will take it. and time-box grants so mistakes expire on their own.
with those four in place, detection can stay light: watch for clusters of redemptions from the same device or network and review the top handful of earners weekly by hand. that is usually enough at consumer scale, and it is a fraction of the cost of the fraud pipeline you would need if the rewards were money-shaped.
incentives are the last layer of the stack, and they only pay once the timing is right and there is real enthusiasm to amplify, which is the running argument of the product-led growth playbook. want your reward economy checked before it ships? the breakdown call is free. Let’s talk Growth.
this page is part of the Product-led growth & viral loops playbook.
More in this playbook
Why nobody used your referral program
the post-mortem of the bolted-on referral tab: missing motive, wrong timing, friction, rewards nobody wants, and what to fix before rebuilding.
Product-led growth & viral loopsShareable outputs: the AI-native viral loop
how to design outputs people want to post, make export native per platform, and why share rate times conversion beats invite programs.
Product-led growth & viral loopsWatermarks: growth vs user experience
a watermark converts viewers of shared outputs, but too loud and sharing stops. the subtle-but-legible rules and when to sell removal.
Product-led growth & viral loopsWhen to add a referral program: pre or post PMF
add referral after retention flattens, not after launch. the readiness checklist, and why referral amplifies love it cannot create.
Product-led growth & viral loopsK-factor math that isn't fantasy
the real k-factor formula, why k above 1 is almost mythical, and how sub-1 k still discounts your paid CAC if you measure honestly.
Product-led growth & viral loopsApple's rules and rejected referral programs
the reward mechanics Apple rejects under guideline 3.2.2, why referral rewards trip review, and designing incentives that pass.
Product-led growth & viral loopsOnboarding as a growth surface
time to first output is the retention lever. how removing onboarding steps compounds every channel, and why activation beats signups.
Product-led growth & viral loopsRetention before referral: the order of operations
loops multiply retention, they never replace it. the honest sequence of activation, habit, then loops, and how to read your curve.