The return on GEO is hard to pin down, and it’s better to say so than to pretend otherwise. The objection usually takes two forms. Attribution: some people influenced by an assistant will arrive by another route, so AI won’t get the credit. And outcome: a mention rate is a new number, and finance doesn’t know how it turns into sales. Both are fair. Both are more manageable than they look.
What you can measure directly
More than most teams expect, as long as each number comes with its limits.
- How often answers name youMeasured on prompts that don’t name you, by assistant, with a likely range and the number of prompts behind it.
- Who is named insteadYour rate next to your competitors’ on the same prompts.
- What the answers get wrongPrices and stock that contradict your catalogue, and the pages they came from. These have a direct cost and a clear fix.
- Visits from AI assistantsYour analytics can show visits from ChatGPT, Perplexity and other assistants, and how often they buy.
What you can’t measure directly is the link between the two: how many sales a rise in mention rate produces. That part has to be estimated, and the estimate should show its assumptions.
The attribution problem is real, and familiar
Someone who hears about you from an assistant might later search for your name, or type your address in. That sale won’t be credited to AI. But that’s true of PR, word of mouth and brand advertising too, and teams already accept it there.
A more useful question than “can we attribute every sale?” is “what do the AI visits we can see already tell us?”. If your analytics show visits from assistants, how often they buy and what they spend, you have a floor to work from.
Making the case to finance
Build the case from your own numbers, and show the assumption that links them. The calculator does the sum: it takes your AI visits, how often they buy, your average order value and your mention rate, and assumes visits rise and fall in proportion to how often you’re named. It’s a rough guide, and it says so.
It gets clearer as you measure
Before you start, you have nothing to compare with. After a first measurement, you know where you’re named, where you aren’t and what answers lean on instead. After acting on that and measuring again, you can see whether the ranges have moved apart, and what has happened to AI visits in your analytics over the same period.
That doesn’t prove your work caused a change, because assistants and the web change too. But it turns a guess into a measured question you can keep checking, and that’s usually enough to decide whether to continue.