Measuring GEO ROI: The Question That Blocks Every Budget
70.6%. That is the share of traffic generated by AI chatbots that arrives on a site with no origin trace at all. Google Analytics dumps it all into the "Direct" bucket, the same one used for a visitor who typed the address from memory.
70.6%. That is the share of traffic generated by AI chatbots that arrives on a site with no origin trace at all. No "referred from ChatGPT", no "referred from Perplexity". Nothing. Google Analytics dumps it all into the "Direct" bucket, the same one used for a visitor who typed the site address from memory. A GEO budget (optimizing visibility in chatbot answers from ChatGPT, Claude, or Gemini, as opposed to classic Google search ranking) can therefore produce real results and still remain invisible in the dashboard used to justify it.
That is the heart of the problem. A marketing team invests in GEO, gets citations in AI answers, sometimes sees traffic rise, but cannot answer the simplest question a CFO will ask: how much does it return. Not because GEO returns nothing. Because measurement tools were built for a world where people clicked blue links, not for a world where they get a direct answer inside a conversation window.
Why the click is no longer enough
Return on investment (the calculation that compares what an action cost to what it returned) has always relied on a chain of trackable clicks. A user clicks a Google result, lands on a page, fills a form or buys a product. Each step leaves a trace, each trace feeds a dashboard.
That chain breaks with AI chatbots, for three precise technical reasons. The mobile apps for ChatGPT or Claude do not pass origin information when they open a link. Many users copy-paste an address straight from the chatbot answer instead of clicking, which erases any provenance trail. And ChatGPT Plus, like Google's AI mode, deliberately adds a technical instruction that prevents destination sites from knowing where the visitor came from.
As a result, a site can see its "Direct" traffic double without anyone on the team understanding why. The answer is almost always the same: part of that supposedly originless traffic actually comes from AI chatbots. And this ghost traffic is not anecdotal. It converts at 10.2%, versus 2.46% for non-AI traffic classified elsewhere. A visitor from a chatbot buys or fills a form four times more often than an average visitor, but nobody knows because nobody sees it.
Before trying to calculate ROI, the first task is therefore to repair measurement itself. A channel grouping that manually recognizes known AI domains (chatgpt.com, perplexity.ai, claude.ai and the like) recovers part of the identifiable traffic. For the rest, the traffic that arrives with no trace at all, only server-log analysis, more tedious but more reliable than classic analytics tools, can rebuild a faithful picture.
The metrics that replace the click
Once measurement is fixed, a deeper problem remains. Even when well measured, the click tells only a small part of the GEO story. A majority of Google searches now end with no click at all, and that share rises sharply whenever an AI-generated summary appears at the top of the results page. A user who asks ChatGPT a question and gets a satisfying answer, citing the brand along the way, often has no reason to click anywhere. The brand was still seen, named, recommended. That value does not land in any clicks column.
Substitution metrics are then needed, ones that measure presence rather than passage. The first is citation frequency: on a sample of queries representative of the sector, in how many answers does the brand appear, and at what rank in the source list. The second is share of voice, which compares that frequency to competitors on the same queries. The third is the sentiment attached to the citation: an answer that mentions the brand in positive terms does not weigh the same as a neutral mention in the middle of a list, or worse, a mention paired with a caveat about quality or price. The fourth is the nature of the sources the chatbot goes looking for to build its answer, because a brand absent from the pages the AI consults to form an opinion has no chance of being cited, whatever its budget.
These four metrics do not replace the financial calculation. They build the base without which that calculation means nothing. A GEO budget that moves citation frequency and share of voice forward, month after month, produces an effect that will eventually translate into traffic and sales, even if the exact path between the two stays blurry. Conversely, a budget that moves none of these four numbers will never produce a return, no matter how good the published content is.
What a chatbot visitor is really worth
The most counterintuitive point in this whole topic is that AI traffic, when measured correctly, converts clearly better than classic traffic. Several studies published between late 2025 and 2026 agree on an order of magnitude: between four and five times the average conversion rate of classic organic search, with gaps ranging from 1.3x on impulsive online purchases up to more than twenty times in some B2B sectors where the buying decision is long and deliberate.
| Traffic origin | Average conversion rate |
|---|---|
| Classic Google search | 1.76% |
| Claude | 5.0% |
| Perplexity | 10.5% |
| ChatGPT | 15.9% |
The logic behind these numbers is simple once you see it. A user who clicks a Google result is still comparing, opens several tabs, hesitates. A user who already queried a chatbot has gotten a synthesis, asked follow-up questions, eliminated some options. When they finally click through to a site, it is often because they are ready to act. The chatbot did part of the qualification work upstream.
This data changes how a GEO budget must be read. A lower volume of AI traffic than classic Google volume is not necessarily a failure, if that rarer traffic converts four times better. The right question is not "how many visitors did GEO bring", it is "how many sales or leads did this traffic, even reduced, produce". It is a complete reversal of how classic search ranking learned to think for twenty years, when volume was king.
Building a formula that holds up
Without waiting for a perfect tool that does not exist yet, a company can build a reasonable GEO ROI measure on four pillars.
The first is a starting snapshot, taken before any investment: current citation level, share of voice versus competitors, dominant sentiment. Without that baseline, there is no way to know whether the efforts produce an effect.
The second is regular tracking of those same visibility metrics, month after month, to objectify progress rather than relying on a feeling.
The third is reconstituting traffic that actually came from chatbots, via a correctly configured channel grouping and, when possible, server-log analysis to recover what escapes standard tools.
The fourth is linking that reconstituted traffic to the conversions it produces, keeping in mind that its conversion rate will likely be higher than other channels, not lower.
Once these four building blocks are in place, the calculation becomes: the value of sales or leads attributable to that traffic, minus what GEO cost, all relative to the starting investment. It is not exact science. It is a considerable improvement over the total absence of measurement, which remains the norm today in most companies that are already investing in this topic.
The temptation, faced with this complexity, is to wait for analytics tools to catch up before acting. That is the inverse error of spending without ever measuring. Companies that start today tracking their citation frequency and share of voice are building a history. Those that wait will start in a year with no baseline at all, at the exact moment their competitors already have twelve months of it.
This is exactly what Vurto does: track citation frequency, share of voice, and brand sentiment across ChatGPT, Gemini, Perplexity, and Claude, month after month, to give marketing teams the measurement base that most GEO budgets still lack.