Revenue Per Query for AI Apps — 2026 Model

Revenue Per Query for AI Apps: How to Model It Honestly in 2026

For AI apps running through independent ad networks in 2026, blended revenue per query sits in the $0.001–$0.015 range at current eCPMs of $15–$50, publisher shares in the 50–70% band, and fill rates of 10–30% on commercial-intent queries. Apps that serve only low-intent chat queries land at the bottom of the range; apps with dense commercial-intent traffic and a monetization-optimized ad load hit the top. RPQ beats DAU as the north-star monetization metric because it bakes in fill, CPM, and ad load in one figure.

Case Study ->
Case Study ->
Case Study ->
Case Study ->
Case Study ->
Case Study ->

Case Study ->

Revenue Per Query for AI Apps — 2026 Model | Thrad

Revenue per query (RPQ) is the one number that tells an AI-app founder whether the free tier can ever pay for itself. This piece walks through the honest math — advertiser CPM, publisher share, fill rate, ad load — and lands on a 2026 range publishers can actually plan against.

Date Published

Apr 20, 2026

Date Modified

Apr 20, 2026

Start monetizing your AI app in under an hour

With Thrad, publishers go from first API call to live ads in less than 60 minutes. With fewer than 10 lines of code required, Thrad makes it easy to unlock revenue from your conversational traffic the same day.

Monetize your AI

Revenue per query (RPQ) is the one number an AI-app founder should burn into their spreadsheet before modeling anything else. It tells you whether your free tier can ever pay for itself, how much GPU spend you can absorb, and whether advertising is a real business line or a rounding error. This guide walks through the honest 2026 math — no hand-waving, no inflated CPMs.

What is revenue per query for an AI app?

Revenue per query is the average ad revenue an AI app earns each time a user sends a prompt. The formula is simple: RPQ = eCPM x fill rate x ads per query / 1,000. Every knob that matters in an ad-funded AI business lives inside that equation — demand density, user intent mix, ad load, and auction clearing price.

Most AI-app founders first reach for DAU or MAU as their headline number. That is a growth metric, not a monetization metric. DAU tells you how many people you served; RPQ tells you what each served prompt was worth. Two apps with identical DAU and radically different RPQs have radically different P&Ls, and the gap compounds every month.

The practical reason RPQ matters in 2026 specifically: AI-app query volume is exploding while direct ad spend is lagging. The AI app sector reached an estimated $16.5 billion in revenue in 2025, a 180% year-on-year jump per Business of Apps reporting, but most of that was subscription revenue — not ads. Founders who model RPQ honestly can see how much runway they actually have before the demand side of the auction catches up to the supply side they are already generating.

How do you calculate RPQ step by step?

Start with the auction outcome and work back. eCPM x fill x ads_per_query / 1,000 = RPQ. A worked example: an AI app with a $25 eCPM on filled impressions, a 20% fill rate (one in five queries monetizes), and one ad unit per monetized query earns $25 x 0.20 x 1 / 1,000 = $0.005 per query. 10 million queries a day at $0.005 is $50,000 per day, roughly $18M per year. That is the shape of a real ad business — not a typo.

Input	Low-intent chat app	Mixed-intent productivity	Research / commerce
eCPM (filled)	$10–$20	$20–$40	$25–$50
Fill rate (blended)	5–15%	15–25%	25–35%
Ads per filled query	1	1	1–2
RPQ range	$0.0005–$0.003	$0.003–$0.010	$0.008–$0.015

These ranges assume a 50–70% publisher share after an independent ad network's take — which is where most sub-OpenAI-scale AI apps actually monetize. If you are early and shopping for supply-side plumbing, look at Thrad's ad platform for AI-app publishers before you build bespoke integrations; wiring an SSP into a conversational interface from scratch is a 3–6 engineer-quarter project that almost never outperforms a network built for the format.

A $25 eCPM at 20% fill earns the same RPQ as a $50 eCPM at 10% fill. Publishers who chase headline CPM numbers without moving fill usually end up worse off than publishers who accept a modest CPM and optimize their auction.

Why does CPM not tell the whole story?

CPM is the sticker price of a cleared impression. It says nothing about how often your inventory clears. AI-app inventory clears unevenly for two structural reasons: most prompts carry no commercial intent, and most demand sources are tuned for search or social — not conversational surfaces.

When ChatGPT's direct ad program launched in February 2026, per PPC Land reporting CPMs opened at roughly $60 and dropped to approximately $25 within nine weeks as the auction densified and the $250K minimum was cut to $50K. That is a textbook CPM-versus-fill story: the $60 number was an artificially sparse auction with few advertisers; as more bidders joined, clearing price normalized but total revenue went up because fill went up.

For independent AI apps, the lesson is the same in reverse. You will launch with a modest CPM and low fill. Your job is to expand the set of monetizable query classes (more fill) before you can pretend to push CPM. Run a publisher P&L against RPQ, not CPM — otherwise you will optimize the wrong lever and wonder why revenue is flat.

How does fill rate actually work inside an AI app?

Fill rate in an AI app means the percentage of prompts that produce a billable ad impression. On a conversational surface it is not a passive property of the inventory; it is shaped by three decisions the app makes on every turn.

Classifier confidence — Does the prompt have high enough commercial
intent to surface an ad? A prompt like "best running shoes for flat
feet" is a near-certain monetize. "Explain reinforcement learning
in 3 sentences" is not.
Ad-unit placement — Inline sponsored recommendation inside the
answer, sidecar card next to it, or a follow-up suggestion chip?
Each has different click propensity and different demand pools.
Auction competition — Does the demand-side have advertisers
bidding on this segment? A travel prompt in Q4 fills at near-100%;
a niche B2B prompt may fill at 5%.

Independent networks solve point 3 by aggregating advertiser demand across many AI-app publishers. The creative examples in Thrad's ad gallery show how the same prompt class clears very differently when the auction is thick versus thin — fill moves from single digits to 40%+ as advertiser count scales.

What is the right ad load for an AI app?

Ad load is the number of ad impressions per session or per N queries. It is the most abusable lever in the RPQ equation — doubling ad load doubles revenue in the short run and craters retention in the medium run. The 2026 working consensus for general-purpose AI chat apps is one ad per 6–10 queries on the free tier, with zero ads on paid tiers.

The constraints are different per surface:

Shopping-flow apps (product search, travel planning): ad load can be
higher because users expect commercial results. 1 ad per 2–3 queries
is typical without harm.
Companion / social AI (character chat, roleplay): ad load must be
lower — one ad per 15–20 queries or users churn. These apps typically
lean on subscriptions and IAPs first.
Productivity / work-task AI (email drafting, code help): near-zero
in-task ads; monetization happens on landing pages, upgrade prompts,
and partner referrals, not inside the chat.

Getting ad load wrong is the #1 RPQ-destroying decision founders make. Publishers who push past the retention-safe threshold to hit a quarterly number typically see 10–20% DAU decay within 60 days — which erases the RPQ gain and then some.

How do different monetization stages change RPQ?

RPQ is not a static number — it moves as an AI app matures through three identifiable stages.

Stage 1 — Bootstrapped / pre-network

No ad network integration. Revenue comes from direct sponsorships, affiliate routing, or first-party commerce. RPQ is functionally $0 on most queries with occasional high-value affiliate clicks that average out to $0.0005–$0.002 blended. Most AI apps spend 6–18 months here.

Stage 2 — Early network integration

Connected to one or more AI-native ad networks. Demand is thin, fill is 5–15%, and CPMs land at the low end of the $15–$30 range. Blended RPQ lands $0.002–$0.005. This is where most 2026 AI apps with 10K–1M DAU actually sit.

Stage 3 — Dense auction / optimized

Multi-network integration, mature classification on commercial intent, optimized ad placements. Fill on monetizable queries climbs past 30%, CPMs hit $30–$50, and RPQ lands $0.008–$0.015 blended. Only a handful of independent AI apps are here in 2026, but the gap between Stage 2 and Stage 3 is almost all operational — demand density plus better targeting — and does not require more users.

The economics reason to jump quickly to Stage 2 is that unit cost of serving a free query runs $0.001–$0.005 per Silicon Data's 2026 inference cost analysis. At Stage 1 you are eating that cost entirely; at Stage 2 you start closing the gap; at Stage 3 you are contribution positive.

Why does RPQ matter for AI app founders in 2026?

Because the subscription-only model is visibly under-delivering. Free tiers generate the query volume that drives product-led growth, but they also generate the compute bill that kills margins. Advertising is the only model that scales revenue with query volume the same way inference cost scales. RPQ is how you measure whether that scaling is working.

The sharpest frame comes from Vincent Schmalbach's chatbot monetization analysis: in search, Google serves ten ads per query at near-zero marginal cost. In an AI app, the cost of producing the answer is several cents of GPU time, and serving even one ad carries real latency and UX tax. If RPQ sits below marginal inference cost, the free tier is structurally loss-making and the business depends entirely on conversion to paid. RPQ above marginal inference cost means the free tier is a real business.

That single threshold — RPQ greater than or equal to marginal query cost — is the most important number in an AI-app P&L for 2026. Founders should model it weekly, not quarterly. Publishers evaluating how conversational inventory prices differently from search and display should skim Thrad's publisher infrastructure page for how the mechanics differ from a traditional SSP.

What are realistic RPQ targets by app type?

Directional 2026 ranges blended across filled and unfilled queries on independent networks:

App archetype	Commercial-intent share	Realistic RPQ range
AI companion / character	<10%	$0.0005–$0.002
General-purpose chat	10–20%	$0.002–$0.005
Productivity / writing	15–25%	$0.003–$0.007
Search-style / research	30–45%	$0.006–$0.012
Shopping / travel planning	50%+	$0.010–$0.020

The delta between archetypes is 20–40x. Founders who assume universal RPQ numbers from chatbot monetization blog posts miss this by an order of magnitude. Your app's intent mix is the single largest determinant of RPQ, and you cannot change it by changing ad networks — you change it by changing product surface.

Common misconceptions

"Higher CPM equals better publisher." False. Higher fill at lower
CPM usually wins. The ChatGPT direct deal's $60 to $25 CPM drop was a
good thing for the auction — it meant fill was rising.
"Our DAU is huge, so ad revenue will follow." Only if your DAU
generates commercial-intent queries. A 10M-DAU companion app and a
100K-DAU travel assistant can have the same absolute ad revenue.
"Network take rates are the problem." At Stage 2 they are not —
you cannot build the demand side in-house at your scale. 50–70%
publisher shares net of an AI-native network are standard and fair.
"RPQ scales linearly with users." Only until the auction densifies.
After that, eCPM rises as advertisers compete for the same prompt
classes — so RPQ scales superlinearly with users on monetizable
segments.

What comes next

RPQ will keep expanding for four reasons through 2027: advertiser adoption of AI-native formats rises, classifier quality improves (lifting fill), auction density rises (lifting CPMs), and new ad units (inline citations, follow-up chips, shopping cards) widen the set of monetizable query classes. Founders who stay disciplined on ad load while these tailwinds hit compound faster than founders who burn retention chasing short-term revenue.

How to get started

Instrument your app to log every prompt with a commercial-intent
score. You cannot optimize RPQ without knowing which queries are
in the monetizable subset.
Pick one ad-network integration — not five. Measure fill, CPM, and
blended RPQ for 30 days. Publishers evaluating supply-side plumbing
can start with Thrad's ad platform for AI-app publishers,
which is purpose-built for conversational inventory.
Set a hard ad-load ceiling (1 per 6–10 queries on free tier) and
resist the temptation to exceed it for a monthly number.
Re-plan the P&L around RPQ, not DAU. Forecast revenue as RPQ times
projected queries, and compare against projected inference cost.
If RPQ exceeds cost, the free tier is a business. If not, you have a
growth problem dressed up as a monetization problem.
Review creative examples in the format-level case studies before
picking inventory types, so your UX assumptions and your auction
assumptions are looking at the same thing.

RPQ is boring to work on — it does not make announcements, it does not close funding rounds. It is the number that quietly determines whether your AI app is a product or a subsidized science project. Build around it early and the rest of the business takes care of itself.