The AI ad pilot terms worth settling before launch are the revenue share base (gross or net), which events are counted, how invalid traffic is adjusted, what reporting you can verify, and when and how payouts arrive. The percentage is the least informative number in the contract. This is an educational checklist, not legal advice, so have counsel review anything you sign.
Published splits look comparable and are not. Dappier's AskAI offers another model for revenue sharing and platform-covered costs. ProRata's Gist says it shares 50% of its advertising and subscription revenue with partners, allocated by how much each partner's content contributes to answers. Perplexity's publisher program has been reported to pay up to 25% of ad revenue when cited content is followed by a served ad. Three programs, three different things being divided. Read the definitions, not the headline.
Why pilot terms matter more than the headline percentage
The percentage only has meaning once you know what it applies to. A 50% share of a small, heavily adjusted net figure can pay less than 30% of a clean gross one.
A pilot is also when terms are cheapest to negotiate. Neither side has volume, and you have no history to defend. Once an integration ships and users see ads, your leverage drops. Get the revenue base, counting rules and exit rights into writing first.
A pilot also exists to teach you something. If you can't reconstruct how a payout was calculated, you can't tell whether ads fit your app or the arrangement is quietly leaking value. Terms are what make the experiment readable. For the wider conversation to have before any code ships, see Before the SDK: the publisher conversation worth having.
Revenue share: what the percentage is a percentage of (gross, net, deductions and fees)
A revenue share percentage applies to whichever revenue figure the contract defines, so define it before you discuss the rate. Ask which of these is meant:
Gross advertiser spend: what the advertiser was billed for delivered activity.
Net revenue: gross minus listed deductions.
Platform receipts: only what the platform actually collected, which can lag or fall short of billings.
Then ask for every permitted deduction by name. Typical candidates are platform or network fees, payment processing, taxes, agency commissions, refunds and credits to advertisers, and invalid traffic adjustments. A clean clause lists them exhaustively and says anything not listed can't be deducted.
Ask three more questions. Is the share applied before or after fees? Can fees change mid-pilot, and with what notice? And if the platform covers costs such as inference or hosting on your behalf, is that cost netted out of your share or absorbed? The Dappier example is useful because it states who carries LLM and hosting cost, a line many agreements leave out.
On Jolts, revenue share, counting, fees and payout timing are agreed with each publisher before going live. There is no fixed rate, guaranteed earnings figure or test minimum to promise. Put the definitions on the table in the first call.
What counts: impressions, clicks, invalid traffic and counting rules
What counts is whatever the contract says is a billable event, so write down the event, the moment it is counted and who decides. Campaign models in this space are usually CPM (per thousand impressions) or CPC (per click). The counting rules underneath them differ more than the labels suggest.
For impressions, ask: is an impression counted when the ad is requested, delivered, rendered, or actually visible? In a chat interface, a sponsored card can be sent but scrolled past, or sent twice after a reconnect. See The chat reconnected. Did you count the ad twice? for why deduplication belongs in the rules.
For clicks, ask whether repeated clicks from one user within a window collapse into one, and whether accidental or automated clicks are filtered before billing.
For invalid traffic, ask for the definition (bots, self-clicks, test traffic, duplicate events), who detects it, and whether adjustments can reach back into periods you were already paid for. Clawbacks after payout are a legitimate thing to negotiate. Set a lookback limit, such as a fixed number of days, and require the adjustment to be itemised.
Also confirm whether CPA is on offer. On Jolts it isn't available yet, so a pilot rests on CPM or CPC counting.
Reporting and verification: what you should be able to see and check
You should be able to see enough reporting to recompute your own payout, and the contract should say so. If the numbers arrive only as a total, you're trusting rather than verifying.
Ask for these, at minimum:
Counted impressions and clicks by day, placement and campaign category.
Gross spend, each deduction and the net figure, line by line.
Invalid traffic removed, with the reason category.
The effective rate you earned per thousand impressions or per click.
Then compare against your own logs. Your app knows how many eligible moments occurred and how many ads rendered. A gap between your count and the platform's is the first thing a verification clause should address: who investigates, in what timeframe, and what happens to a disputed amount.
Audit rights are the stronger form of this. Even a light version helps: the right to request supporting records for a disputed period, with reasonable notice. Pilots rarely need a full third-party audit, but they do need a documented path to questions.
Reporting also depends on what data crosses the boundary. Match what you share to what the platform needs for ad matching and counting. AI ad request data: what publishers should share covers that side.
Payout terms: thresholds, timing, rollover, currency and disputes
Payout terms are five separate questions: when money is calculated, when it is paid, how much must accumulate first, in what currency, and what happens when you disagree. The published descriptions of chatbot ad programs typically don't state net-30 or net-60 timing or a minimum payout threshold, so confirm them in your own agreement.
Work through the list:
Payment timing: net 30 means payment is due 30 days after the invoice or statement date, net 60 after 60. Ask what the clock starts from: month end, report finalisation or advertiser payment. A net 30 that starts when the advertiser pays is effectively longer.
Minimum threshold: if earnings below a set amount aren't paid out, ask whether they roll over, whether they expire, and what happens to an unpaid balance when the pilot ends. A pilot that ends under the threshold can leave you with nothing.
Currency and conversion: which currency the share is calculated in, which exchange rate applies and who bears transfer fees.
Adjustments: whether later corrections are netted from future payouts or invoiced back.
Disputes: a window to raise them, a named contact, and whether undisputed amounts are still paid on time.
Keep these as written terms, not assumptions. On Jolts, payout timing is agreed before going live, alongside revenue share, counting and fees.
Pilot scope and exit terms: categories, frequency, exclusions, no-ad fallback and ending the test
Scope and exit terms protect your product, and they are commercial terms, not just policy. Revenue share means little if the pilot degrades the experience you are funding.
Agree these before launch:
Eligible categories and excluded topics, such as sensitive subjects you never want sponsored content near.
Frequency limits, so users don't see ads on every turn.
Separation from the organic answer. Sponsored content should be distinct and clearly labelled. See Sponsored offers in AI: keep ads separate from answers.
A no-ad fallback: when no suitable match exists, the user sees no ad rather than a poor one. Confirm that unfilled moments carry no penalty.
Rendering and shared data, written down so the integration scope isn't open-ended.
Then the exit. Ask how either side ends the pilot, with what notice, and whether you can pause immediately if an ad harms users. Ask what happens to earned but unpaid revenue on termination, and whether any exclusivity or minimum-volume commitment survives it. A fair pilot has neither. Also settle how success is judged, so ending or extending the test is a decision against criteria rather than a negotiation after the fact. Anything that must be unwound should be unwound by contract, not by goodwill.
A pilot terms checklist to bring to the first call (and how Jolts approaches it)
Bring one page to the first call and don't leave without answers to each line. Jolts connects advertisers with AI publishers through relevant, clearly labelled sponsored placements in AI conversations. It agrees revenue share, event counting, fees and payout timing with each publisher before going live. No fixed rate, guaranteed earnings or test minimum is promised, which is why the questions below matter.
Revenue base: gross, net or collected? Which deductions are allowed, listed exhaustively?
Fees: who sets them, and can they change mid-pilot?
Counted events: impression and click definitions, deduplication, CPM or CPC.
Invalid traffic: definition, detection, lookback limit, itemised adjustments.
Reporting: daily breakdowns, deductions and rate visible, plus a way to reconcile with your logs.
Verification: how discrepancies and record requests are handled.
Payout timing: net 30 or net 60, and what starts the clock.
Threshold and rollover: minimum, expiry, treatment at pilot end.
Currency, transfer fees, disputes and the dispute window.
Scope: categories, frequency, excluded topics, shared data, no-ad fallback.
Exit: notice, immediate pause, unpaid balances, no lock-in.
Platforms differ in what they will put in writing. For how to compare them on controls and terms, see AI chatbot ad networks: what publishers should compare. For the commercial context on funding a free tier, see What should pay for your AI app's free tier?. Jolts details for publishers are on its publisher page.
Put the definitions in writing before you discuss the rate
A revenue share percentage is only as good as the definitions beneath it, so settle the revenue base, counted events, invalid traffic rules, reporting and payout timing before launch. If you are weighing an ad pilot for your AI app, book a call with Jolts, bring the checklist above, and agree audience fit, placement scope and these terms before anything goes live.