The Economics of GEO: Budget, Attribution, and ROI

Why attribution is genuinely hard

Start with the honest version of the problem, because most explanations of GEO measurement skip it. In paid search or paid social, a person sees an ad, clicks it, lands on a page with a UTM parameter, and that session gets stitched to a signup and eventually to revenue. Even organic SEO, for all its own attribution headaches, gives you a referrer header and a landing page. You know someone came from Google, you know what they searched, and you can watch the session.

GEO gives you none of that, structurally, not as a temporary gap in your tooling. A founder asks ChatGPT or Claude or Perplexity a question like "what's a good tool for finding where to promote my SaaS product," the model reads across whatever it has ingested or retrieved, forms an answer, and — if you are lucky — mentions your product by name inside that answer. The founder reads the recommendation, maybe opens a new tab, types your domain directly, and signs up. Nothing in that path touched a link you control. There is no referral URL because there was no referral link. There is no UTM parameter because nobody clicked through your marketing. There is no pixel fire because the "impression" happened inside a chat window that is not yours and was never going to be yours. Your analytics will show this person as direct traffic or, worse, invisible entirely if they searched your brand name on Google first and arrived via what looks like branded organic.

It gets harder still because the recommendation itself is probabilistic and context-dependent. The same model can answer the same question differently across sessions, across users, across model versions released weeks apart. You cannot query "how many times was I recommended this month" the way you can query ad impressions, because that number does not exist anywhere, not even inside the AI labs in a form you could subpoena. What you have instead is sampling: you can ask models questions yourself and see what they say, but that is a proxy for the true distribution of answers, not a census of it. Anyone who tells you they have a precise measurement of "AI recommendation share" is describing a sample, dressed up as a metric. That does not mean the effort is unmeasurable or not worth doing — it means the measurement problem needs to be solved with different tools than the ones marketing teams are used to, and pretending otherwise just produces false confidence dressed up as a dashboard.

Sizing a budget: time and attention, not just dollars

For most startups and indie teams, the honest budgeting unit for GEO is not dollars, it is founder or team hours, because that is the scarce resource being allocated. GEO competes directly with content marketing, community engagement, cold outreach, and paid acquisition for the same finite pool of attention a small team has each week. A founder who spends six hours writing forum replies and comparison content is not spending those six hours on something else that also builds the business. Framing GEO as a line item with a dollar figure, the way you might frame a Google Ads budget, tends to understate the real cost for small teams and overstate it for teams that could actually afford to hire for it.

A more useful question than "how much should we spend on GEO" is "how does GEO rank against the other things we could do with the next five hours." Early on, before positioning is settled and before you have any presence in the places your buyers look for advice, GEO work overlaps heavily with foundational content and community work you would want to do anyway — writing clearly about what you do, answering real questions honestly in public, being present where your category gets discussed. That overlap is a gift: it means the marginal cost of adding a GEO lens to work you were already doing is low. The cost rises once you are past that stage and are deciding whether to spend additional hours specifically chasing AI-answer visibility versus, say, running a paid acquisition test or writing a technical deep-dive aimed at humans and search engines both.

Larger teams with real marketing headcount can and should think about this differently — allocating a fraction of a content or brand budget, assigning a specific person or a slice of an agency retainer to it, and treating it the way they'd treat organic social or PR: a real cost center with an expected, if fuzzy, payoff. But for a one-person or two-person founding team, the realistic framing is time-boxed and modest. A few focused hours a week, consistently, sustained over months, will do more than an intense one-week sprint followed by nothing. This is one of the places a tool like Wally earns its keep for small teams specifically — it can research where the conversations are happening and draft the replies, landing pages, and campaign material so the founder's limited hours go to judgment calls and approvals rather than the mechanical work of finding and writing everything from scratch.

Proxy metrics that function as leading indicators

Since there is no clean conversion path to measure, the discipline shifts from tracking a funnel to watching a set of proxies that, together, tell you whether you are becoming a more consistent presence in the answers models give. No single one of these is proof of anything on its own. Taken together and watched over months, they are the closest thing to a real signal you will get.

• Consistency of favorable mentions in periodic spot-checks. Regularly ask the major models the questions your buyers would plausibly ask — category comparisons, "best tool for X," "alternatives to Y" — and log whether you show up, how you are described, and whether the description is accurate. One good mention proves little. The trend line across weeks and months, across different phrasings of the same underlying question, is the actual signal.
• Direct and branded search volume over time. When someone hears about you from an AI answer and later searches your company name on Google, or types your domain straight into the address bar, that shows up as branded search or direct traffic — not as anything attributable to GEO, but it moves when awareness moves. A slow, sustained rise in branded search alongside consistent AI mentions is a reasonable, if imperfect, corroborating signal.
• Referral traffic from AI-product domains, on the occasions it does appear. Some AI products do pass a referrer when a user clicks a link inside an answer or a search-style result. It is a minority of the actual influence GEO has, since most exposure happens inside pure conversation with no link at all, but when you do see referrals from these domains in your analytics, they are a rare piece of hard data worth tracking as its own line.
• Unprompted customer mentions. Sales calls, onboarding surveys, and support conversations where someone volunteers "I heard about you from ChatGPT" or "Claude recommended you" are disproportionately valuable precisely because nobody asked them to say it. Start logging these deliberately — a shared note or a tag in your CRM — because they are easy to forget individually and easy to underrate collectively. A handful a month, sustained over a quarter, is a real pattern, not noise.
• Qualitative shifts in how you are described. Watch not just whether you're mentioned but what language models use to describe you. Are they characterizing your product the way you'd characterize it yourself, or is there drift, staleness, or a competitor's framing leaking into your description. This tells you as much about content and positioning gaps as it does about visibility.

When to increase or cut spend

The instinct carried over from paid channels is to look for a spike — a campaign that runs and produces a visible bump you can point to. GEO does not reward that pattern of thinking, and treating it like a campaign channel will make you conclude, wrongly, that it doesn't work. What you're actually looking for is compounding: is the same core positioning getting corroborated more often, by more sources, in more of your spot-checks, over a period of months. A single new mention on one forum thread is not compounding. The same accurate description of your product showing up independently across a review site, a Reddit thread, your own site, and a comparison page — and then that consistency starting to show up reliably in model answers — is compounding, and it's the thing worth reinforcing.

Increase investment when you see that consistency building and when the marginal hour is clearly going toward filling a real gap — a question you know buyers ask that you still don't answer anywhere, a comparison page that doesn't exist yet, a community where your category gets discussed and you have no presence. Increasing spend chasing a single good result, the way you might scale a winning ad, is usually a mistake, because a single result in this channel is much more likely to be noise than in a channel with real sample sizes and statistical power.

Pull back, or at least hold steady rather than escalate, when the spot-checks have been flat for a couple of months despite consistent effort, when the team's underlying positioning is still unsettled or actively changing, or when the hours going into GEO are visibly starving something else that has clearer evidence behind it, like a paid channel with a real CAC or a content effort already showing organic search growth. It is also fair to scale back deliberately during periods where you need to conserve founder attention for product work — GEO's returns are patient enough that a few quiet months rarely cost you much, which is different from a paid channel where pausing spend can cause an immediate, measurable drop.

Reporting this honestly to a boss, co-founder, or investor

The worst way to report on GEO is to manufacture false precision because the audience wants a number. Do not present an ROI figure, a CAC, or a conversion rate for this channel unless you can actually trace the underlying data, and you almost never can. If you present a made-up number to satisfy a board slide, you will either get held to it later when it turns out to be uninferrable, or you will train yourself to keep manufacturing numbers to match the story you told last quarter. Neither is good practice, and both erode trust faster than admitting the measurement is inherently fuzzy.

The more defensible framing is to describe GEO the way you'd describe a reputation or community-building investment: a long-horizon bet on being accurately and favorably represented wherever your buyers are forming opinions, including inside AI conversations you'll never directly observe. Report the proxy signals as a set, not a single headline metric — spot-check consistency, branded search trend, the count of unprompted "I heard about you from an AI" mentions, notable wins like a strong new mention on a high-traffic community or review site. Presented together, over a consistent cadence, these tell a coherent story even without a bottom-line dollar figure attached.

Be equally clear about cost, since that side of the ledger is much easier to state honestly: this many hours a week, this much of a person's time, roughly this dollar-equivalent if you want to translate it. Pair that plainly stated cost with the plainly stated, necessarily qualitative signal of return, and say directly that you are optimizing for compounding presence over quarters, not a payback period measured in weeks. Investors and co-founders who have run any kind of brand, community, or PR effort will recognize this pattern immediately, because it is the same one — the discomfort usually comes from not having named it that way from the start.