Freemium vs ads, for AI apps.
They are not mutually exclusive. Most AI apps should run both.
The math of each model.
Start with a worked example. 100,000 monthly active users on a consumer AI app. The same user base under two models. The numbers below are rounded for clarity and use the middle of the ranges we see in production.
Freemium math.
Consumer AI apps convert 2–5% of monthly actives to paid, depending on the category and the design of the paywall. Take 3.5% as a mid-point. That is 3,500 paying users at $15/mo blended (some on $10 personal, some on $20 premium), for $52,500 monthly recurring revenue. Annualized, call it $630,000 on this user base. Monthly churn in the 5–7% range is standard for consumer AI, which means customer lifetime is 14–20 months. Blended LTV lands somewhere around $240 per paid user.
The other 96,500 users pay nothing and cost something to serve. Inference, storage, support, and infrastructure run $0.15–$0.50 per monthly active depending on model intensity. At $0.25 per non-paying monthly active, that is $24,125 in monthly cost carried by the paid tier. Net contribution from freemium alone: $28,375 per month, or $340,500 annualized. Real, but thinner than the headline revenue.
Ads math.
Surfaces run across all 100,000 monthly actives because disclosed placements do not require the user to pay. Assume an average user sees 8 eligible surface impressions per month and revenue per impression lands around $0.015 in a mature ad market, less in early days. At maturity: 100,000 × 8 × $0.015 = $12,000 per month, or $144,000 annualized. Cost to serve the same 100,000 users at $0.25 each is $25,000 per month, so the ad-only model loses money at 100,000 MAU unless impressions per user or RPM rise.
Side by side.
At 100,000 MAU, freemium contributes roughly $28K/month net and ads alone contribute roughly negative $13K/month net. Freemium wins on contribution margin. But the hybrid — freemium with Surfaces on the free tier — adds the ad revenue to the freemium base. The free tier moves from costing $24K/month to costing $12K/month after ad credit, which lifts total monthly contribution to about $40K. The paid tier is unchanged. The ads do the work of paying for free users.
When to combine them.
Hybrid is the default for consumer AI apps in 2026. The reason is structural, not strategic. Consumer conversion rates are low enough that a freemium-only model under-monetizes the 95–98% of users who never pay. Those users still cost money to serve. Ads on the free tier turn that cost into a smaller cost or a positive contribution, depending on scale and fill.
The canonical hybrid shape: free tier with Surfaces, paid tier without. Users who upgrade to paid buy the absence of sponsored content as much as they buy the product. Users who stay free get a product that is genuinely free rather than a trial that ends in a paywall. Both populations contribute to revenue through different mechanisms.
The math supports this pattern because the two revenue streams are complementary, not redundant. Subscription revenue follows the paid tier and scales with conversion rate times LTV. Ad revenue follows the free tier and scales with impressions times RPM. Growing the overall user base grows both lines. A product that is subscription- only loses the ad revenue of the free tier entirely. A product that is ad-only misses the subscription revenue of users who would have paid to skip ads.
Operationally, the hybrid needs one control: a flag inside the SDK that suppresses Surfaces for authenticated paid users. That is usually a single environment check inside the Surfaces call. The rest is billing and authentication plumbing you have already built if you are running a paid tier. For the pricing side of this setup, read AI subscription pricing, and for the developer-facing setup see monetize AI app.
The 5% that convert to paid subsidize the 95% who do not, but only partially. Ads on the free tier cover the rest. Without both revenue streams, most consumer AI apps run at a loss on the free base and can never justify scaling it. The hybrid makes the free base economic, which makes top-of-funnel investment economic.
Case studies.
Three pre-AI patterns that translate to AI apps. The mechanics are the same; only the product surface has changed.
Spotify: freemium with ads on the free tier.
Spotify built the largest consumer hybrid of the streaming era. Free users get music with ads. Paid users pay around $10/mo for no ads and offline listening. The company reports roughly 40% paid conversion at scale, but the first decade ran closer to 20–25%. Ad revenue on the free tier carried the early years and still represents a meaningful fraction of total revenue. The pattern translates cleanly to an AI app where free users get Surfaces and paid users do not.
Canva: freemium with high-converting paid tier.
Canva runs a thick free tier that stays useful indefinitely and a paid tier that adds premium assets, brand tools, and team features. Conversion runs higher than most freemium consumer products because the free tier produces work that users can build careers on, and the paid features protect that work. An AI app that produces professional output can run the same shape: free tier good enough to do real work, paid tier good enough to do paid work.
Duolingo: freemium with ads and a paid tier.
Duolingo runs all three revenue layers: a free tier with display ads, a paid tier (Super) without ads and with extra features, and occasional in-app purchases. The free tier is the majority of users and a large share of revenue through ads alone. The paid tier lifts the top. This is the closest published analog to what a consumer AI app should run: ads pay for the free base, subscription lifts the committed segment, and occasional upsells catch users who want specific add-ons.
Read these as patterns, not prescriptions. The mix of revenue lines and the exact price points will vary by category. The principle — combine the revenue streams that match the two populations using your product — is general.
When freemium alone beats hybrid.
Three cases where a paid-only model outperforms the hybrid, and a founder should not bother adding ads.
Enterprise-adjacent apps. If your users are employees using the product inside a company, ads hurt the sale. Procurement teams flag any tool that shows sponsored content to employees, and the cost savings on a per-seat basis do not justify the trust loss. Run paid-only and price against enterprise anchors.
Niche vertical apps. A product serving a small, high-willingness- to-pay audience (legal, medical, financial, specialty research) converts at rates where the paid tier is thick enough to carry the cost of the free tier directly. Ad revenue would add noise without adding meaningful margin, and advertiser fit inside a narrow vertical is hard.
Apps with very high LTV per paid user. If average revenue per paid user runs above $100/mo and retention holds beyond 18 months, a small paid base can subsidize a large free base without any ad revenue at all. The hybrid would still add dollars, but those dollars come with operational overhead that may not be worth the friction.
Even in these cases, the decision is not permanent. A vertical product that eventually serves a broader audience may want Surfaces on the expanded free tier later. An enterprise product that launches a consumer-facing companion can run ads on the companion. Freemium-only today is compatible with hybrid later.
When ads alone beats hybrid.
The mirror case. Apps where building a subscription tier costs more in product complexity than it returns in subscription revenue.
Occasional-use tools. If users come once a month for a specific task and leave, subscription conversion will be low regardless of how well the paywall is designed. The user is not around often enough to value ongoing access. An ads-only model monetizes the visits that happen without asking for commitment the user will not give.
Search-style wrappers. Products where the interaction is a single query and a single answer have limited surface for paid differentiation. Users compare to free alternatives immediately. Ads monetize this shape well because the single answer is also the single surface where a disclosed placement fits.
Companions with low payment intent. Entertainment-adjacent AI products — companions, games, casual conversation — see payment intent below the threshold that makes subscription meaningful. Ads fit this shape naturally because the user is engaged and the session is long enough to surface placements without feeling interruptive. For a direct comparison, see subscription vs ads for AI apps.
Frequently asked questions.
Is freemium or ads better for a brand-new AI app?
What conversion rate should I plan for on a freemium AI app?
Will running ads on my free tier hurt conversion to paid?
Can I build a serious business on ads alone?
How do I know the hybrid is working?
Advertising for AI agents, built to be disclosed.
Join the waitlist. We are onboarding developers and advertisers in the order they sign up.