Performance-based AEO pricing works when the invoice is triggered by pipeline events recorded in your CRM, such as sales-accepted opportunities created from AI-influenced sources. It breaks when the trigger is a citation count or a visit from a chat assistant. Those are leading indicators. They are worth tracking, and they are not revenue.
Some AEO agencies now publish outcome-based terms and month-to-month contracts, and that is pushing down on retainers. Buyers we talk to are asking for it. One told us the gap between two agency quotes was under €1,000, so the pricing model, not the number, was the real decision. That is a fair way to buy. But "performance" needs a definition before anyone signs.
Why citations and AI-referred traffic are not revenue
A citation is a named mention of your brand inside an AI answer. Whether you get one depends on the prompt set, the model and the day you check. The seller usually picks the prompt set, which means the seller also picks how easy the target is.
AI-referred traffic has its own problem. Google's documentation says traffic from AI features in Search is counted inside the overall Search Console performance data, so there is no clean AI-only line to pay against. Other assistants send visits you will need to attribute yourself, and Search Console help will not do that for you.
For instance, one AEO agency publishes a self-reported case study that moved a SaaS client from 550 to 2,300+ AI-referred trials in four weeks. I can't verify that figure, and it doesn't matter for this argument. What matters is that trials are not pipeline until someone checks fit, stage and conversion. A buyer told us they were once handed a TAM list of about 10,000 companies where fewer than 200 were relevant. That is the same failure: volume that looks like delivery.
If you want the wider case for why visibility metrics and pipeline drift apart, we wrote about the gap between SEO rankings and AI citations.
A trigger ladder for performance-based AEO pricing
So the idea is a ladder. I'll call it the trigger ladder: five rungs, from easiest to game to hardest to game.
| Rung | Trigger | What it proves | Failure mode |
|---|---|---|---|
| 1 | Citation share on a prompt set | Visibility | Seller picks the prompts |
| 2 | AI-referred sessions or trials | Interest | Wrong segment, bots, missing attribution |
| 3 | Sales-accepted meeting or lead | Fit | Definition drift over time |
| 4 | Opportunity created with a source | Pipeline | Long cycles, shared credit |
| 5 | Closed-won revenue | Revenue | Too slow, too many other causes |
My recommendation is to pay a success fee on rungs 3 and 4, and treat rungs 1 and 2 as diagnostics. Rung 5 is a fantasy for most contracts. A 120-day sales cycle and six other channels touching the deal mean no vendor can claim it, and you shouldn't want to pay for it as if they could. That is the core of performance-based AEO pricing done properly.
What to negotiate before month one with performance-based AEO pricing
The reason we settle terms up front is so that nobody argues about definitions after the first good quarter. These are the items I would put in writing:
- The trigger as CRM fields. Define "AI-influenced opportunity" using a source property plus a self-reported "how did you hear about us" option for AI assistants, and require the sales-accepted stage.
- A baseline. Measure opportunities from those sources for 90 days before the start and pay only on lift above it.
- The attribution window. Match it to your sales cycle. Month-to-month terms on a 120-day cycle mean someone gets paid for work they didn't do, or doesn't get paid for work they did.
- Record ownership. The data lives in your CRM, and the agency gets read access, not the pen.
- A cap and a floor. Uncapped success fees invite arguments, and a zero floor invites the cheapest possible work.
- Exit terms. Month-to-month cuts both ways, so agree what happens to in-flight opportunities when either side leaves.
Every one of these terms should exist before the first invoice, because performance-based AEO pricing without a written trigger is just a retainer with extra steps.
Where performance-based AEO pricing breaks down
For better or worse, AI attribution is leaky. A buyer asks an assistant for a shortlist, then types your brand into a browser. That visit lands as direct. Self-reported fields recover some of it, and I don't have a great answer for the rest.
The incentive problem is just as real. If the agency is paid on rung 2, it will chase branded and easy queries. If it is paid on rung 4 with no baseline, it will be paid for deals already in your funnel. Both cases look like performance and neither is.
Pricing context helps here. Published price points in this market run from about $2,500 a month for fractional advisory to $120,000 a year for full-service organic work. We broke down what that range buys in our look at program costs now that price floors are public. A success fee on top of a thin base can work. A success fee in place of any base fee usually means the seller is pricing in the risk that you will under-report.
What I would do with a retainer or month-to-month offer
My recommendation is a hybrid: a fixed work fee scoped in hours, plus a success fee on rungs 3 and 4 with a baseline. Fixed hours keep the work honest, and the success fee keeps the seller looking at the CRM instead of a dashboard.
That also answers the question of whether a tool can replace the agency. A visibility tool shows you the gap. Deciding which gaps govern citations and which ones touch pipeline is the job. If your last SEO engagement stalled, the reasons SEO isn't improving search performance are worth reading before you write any AEO contract.
Whichever model you choose, the first question for any seller is the same: which exact field in my CRM triggers payment? If the answer is a count of citations, you are buying reporting. If it is an opportunity record that sales accepted, you are buying answer engine optimization tied to demand generation.
We embed rather than advise, so we agree these definitions with you first and build the CRM fields ourselves. It is time and materials, plus a success component where the attribution is clean enough to support it. Worse for us when the data is messy, better for you.