What ROI should you expect from AEO?

ROI is incremental contribution minus program cost, divided by program cost.

Each term needs a definition before it's useful. Contribution is what a customer pays you minus the direct cost of serving that customer. Incremental means customers you wouldn't have won without the program. No attribution record shows that by itself, so the number of incremental customers enters this example as an estimate. Program cost is everything you spend because the program exists, which includes the agency's fee, any tools and your own team's hours.

The arithmetic is easy. Being strict about the inputs is the hard part, because contribution is easy to overcount and cost is easy to undercount.

Start with a listed price. Lean Labs' AEO page lists an off-site program at $5,000 per month for 10 months, or $50,000 in listed payments. The same page has an on-site package at $9,000 one time with the first year included, and a combined program at $9,000 per month for 10 months, or $90,000. We compare offers like these in how much answer engine optimization services cost.

We'll use the $50,000 off-site program for the example. The fee is only part of the cost. Someone on your team has to sit for interviews, review drafts and approve what gets published, and some teams add a monitoring tool. Suppose that time and those tools come to $15,000 over the same period. Added to the $50,000 fee, that makes a program cost of $65,000, and it's the figure the return has to be measured against. We look at the time side in how much internal time AEO requires from a marketing team.

Suppose a new customer signs a $20,000 first-year contract, and serving that customer costs you $7,000 in the first year. Each customer contributes $13,000 in that first year. Divide the $65,000 program cost by $13,000 and you get five. The program has to produce five customers you wouldn't otherwise have won before it has paid for itself.

Every row uses the same program and the same cost. The return moves only with the number of incremental customers, so that count is the assumption to test.

Now change one input. Suppose your product sells for much less, and a customer contributes $1,300 in the first year. Break-even becomes 50 customers. The agency's work could be identical in both cases, and it has to produce ten times as many customers in the second.

This is why a single ROI figure for AEO would mislead you. The same cost has to be recovered from a different number of customers in each business. We go through that fit question in what kinds of companies benefit most from AEO, and the budgeting side in how an early-stage startup should budget for AEO.

A program report can show three numbers that look like the return and aren't the one the formula needs.

Unweighted pipeline is the stated value of deals that are still open. It counts each one at full value, including the deals that won't close. Revenue is larger than contribution, because the cost of serving the customer hasn't been taken out. Attributed revenue is a third thing. A deal attached to a contact whose recorded source is an AI referral tells you how a visit was recorded. That isn't a full account of how the buyer came to you, and it doesn't tell you whether the buyer would have found you another way. It can also miss a buyer the program did reach, if that buyer arrived by a different route.

Lean Labs' service page lays its measures out in three groups, which is a useful way to keep them apart. Visibility covers mentions, citations and how many tracked prompts you appear in. Demand is the visits assistants refer and the conversions they assist, and revenue is contacts, deals and pipeline. Only the last group is close to the formula, and it still records sources and deals. It doesn't record which customers were incremental.

So the incremental count stays an assumption, and the way to handle an assumption is to vary it. We'd run the return under a narrow assumption and a wide one. Under the narrow one, you count only customers whose first recorded visit came from an assistant, or who told your sales team an assistant pointed them to you. Under the wide one, you count everyone in the narrow group, plus closed deals where an AI referral shows up later in the history. Neither count proves those customers were incremental. Together they show how far the result moves with the assumption. If the program covers its cost only under the wide one, you know which assumption the case rests on.

Some of the effect won't be labeled as AI in your analytics. GA4's AI Assistant channel leaves out Google's AI Overviews and AI Mode, and GA4 counts those visits as Organic Search. Search Console's generative AI report shows impressions for those features, which count links shown and not visits. A buyer who read an answer and came to you later by another route can arrive with no AI source recorded, though your sales team may still learn it by asking. We'd leave whatever isn't recorded out of the return. Any multiplier you applied to cover it would be a guess, and a guess in the contribution figure makes the whole result hard to defend. The reports themselves are covered in how to track traffic and referrals from AI answer engines.

Cost and return arrive at different times. The fee is paid monthly across the term. A deal can close months after the work that led to it, because a page has to be published and found before a buyer visits, and your sales cycle comes after that. Buyers don't all follow that path. Still, a return calculated partway through can count fees for work whose deals haven't closed yet.

So fix the period before you start. Suppose your sales cycle runs about three months. Counting customers acquired through month 13 gives deals that began late in a 10-month term the time to close.

Then say what you're counting for each of those customers. The example above assigns each one a first year of contribution. That's a value for the group of customers acquired in the period. It isn't a count of what was collected by month 13. For a customer acquired in month 12, the first year extends well beyond month 13. Either measure works if you choose it in advance and keep to it. We look at the earlier stages in how long before AEO generates qualified leads.

An agency or a tool may show you a projected ROI. Google's guidance on third-party SEO and AEO services says such tools have no access to its internal ranking data and can't guarantee performance, and that any predictions are their own. A projection is still useful when you can see its inputs. Ask which contribution figure, which customer count and which costs it assumes, then replace them with yours.

Write your break-even count and both counting assumptions down before the first invoice. At the end of the period you'll be reading results against rules you set before any results came in.