Calculating the ROI of GEO: When the effort pays off

Why GEO ROI works differently from SEO ROI

With classic search engine optimization the chain is simple: ranking, click, session, conversion. Every step is countable, and at the end there is revenue you can attribute to a source. GEO, that is Generative Engine Optimization, breaks this chain. A user asks ChatGPT, Perplexity or Google's AI overview for a recommendation and gets a finished answer. Often they don't click anything. The value is created anyway: your brand was named, categorized, perhaps recommended.

That doesn't mean GEO has no measurable return. It means you have to switch the metric. Instead of clicks you count how often and how prominently you appear in AI answers, in what context, and with what tone. This visibility has value because it shapes purchasing decisions long before anyone opens your website. A tax advisor, a mechanical engineering firm and a dental practice all face the same problem and the same opportunity here.

The honest way to deal with this: you won't get a cent-precise ROI like the one you know from performance marketing. What you get is a robust range model that shows you whether GEO is a footnote or a lever in your industry. For many providers that is entirely enough to make a sound budget decision.

The basic formula and its components

The formula stays plain: ROI equals (value achieved minus costs) divided by costs, expressed as a percentage. The art lies in the two numbers you plug in. The cost side is still the easy part, because here you have invoices and working time. The value side you have to model, because it is made up of several indirect effects that rarely appear on a single invoice.

On the cost side this includes: time to analyze your visibility in AI systems, creating and revising content, technical adjustments such as structured data, ongoing monitoring and, where applicable, tool or agency costs. Calculate with full costs, meaning your own working time at a realistic hourly rate. Whoever sets their own time to zero is cheating themselves and makes poor decisions.

On the value side you work your way from the mention to revenue. How many relevant questions do people in your category ask AI assistants? In what share of those are you named? How many of these mentions lead to a visit, an inquiry, a deal? And what is a deal worth on average? You may estimate each of these figures, as long as you document the assumption openly.

Translating visibility into a monetary value

The core of every GEO calculation is translating mentions into euros. A clean route runs via the so-called share of voice: the share with which you appear in AI answers to relevant questions. Measure it by defining a fixed list of typical user questions and running them regularly through several AI systems. If you are named in 20 of 100 questions, your share of voice is 20 percent.

You multiply this share by the estimated question volume and a conservative conversion assumption. Trades example: assume that in your catchment area 500 people a month ask AI assistants a question where a recommendation like yours comes up. At 20 percent share of voice you are named in 100 answers. If only 5 percent of those become a real inquiry and every second inquiry leads to a job worth 800 euros, that is around 2,000 euros of revenue impact per month.

The important thing is humility in the assumptions. Set the conversion rates rather too low, because AI recommendations don't replace your entire sales effort. Calculate two scenarios, a pessimistic and a realistic one, and make your decision at the lower end. If GEO pays off even in the pessimistic case, the matter is clear. If only the optimistic scenario shows black numbers, caution is advised.

Naming the attribution problem honestly

The biggest weakness of every GEO calculation is attribution, that is assigning a result to its cause. When a new customer calls because an AI named your firm to them, that shows up in no analytics tool. The customer appears to come directly. Without countermeasures you systematically underestimate the GEO contribution and wrongly credit it to other channels.

The simplest remedy costs nothing: ask new customers how they found you, and explicitly add the answer AI assistant or ChatGPT to your form or your initial conversation. After a few months you have a data basis that grounds your model assumptions. In addition, you can see in the server log whether referral traffic comes from Perplexity, ChatGPT or Copilot, even if this only captures part of the effect.

Accept that some blur remains. That is exactly why you calculate in ranges and not with a single number. A GEO ROI of 40 to 180 percent is a more honest and more useful statement than a feigned 87 percent. Decision-makers who are familiar with investment risks respect this openness far more than a false precision that shatters at the first critical question.

A worked example

Take a mid-sized B2B software provider. It invests in GEO for one quarter: 30 hours of internal working time at 80 euros makes 2,400 euros, plus 3,500 euros for content and technical implementation, plus 600 euros for tools. Total costs: 6,500 euros. Its offering requires explanation; an average new customer brings 9,000 euros of contribution margin over the lifetime. Just a few deals change the calculation significantly.

Its monitoring shows that after the quarter it is named in 35 percent of relevant expert questions in AI answers, up from 8 percent before. It conservatively estimates the monthly question volume at 300. From the additional mentions it derives, via low conversion assumptions, two additional deals per quarter that it attributes to AI visibility. That yields 18,000 euros of contribution margin against 6,500 euros of costs.

The ROI thus stands at around 177 percent in the realistic scenario. In the pessimistic case, with only one attributable deal, it is still 38 percent. Both figures are positive, so the effort clearly pays off here. For a provider with a 200-euro deal value and little AI-relevant demand, the same calculation would come out negative. That is exactly why industry-specific modeling is decisive.

When GEO doesn't pay off

Honesty also means naming the cases in which you are better off leaving it alone or investing only minimally. GEO pays off poorly when your category is barely queried in AI assistants, when your margin per deal is very low, or when your target group decides in a distinctly local and personal way, for example with pure walk-in customers who do no prior research.

There is also a timing problem. Whoever has no clean foundation yet, meaning no structured content, no clear facts about their own offering on the web, burns money by jumping straight to sophisticated GEO tactics. In such cases the first sensible investment is the foundation, not the optimization. The ROI of the groundwork is often higher than that of any fine-tuning.

How to build your own calculation

You don't need an expensive tool to get started. A spreadsheet with clearly documented assumptions is enough for the first decision. What matters is that every number has an origin and that you separate optimistic and pessimistic. Update the model quarterly with real data from customer conversations and monitoring, then it becomes more accurate with every round.

The following process has proven itself across industries. It forces you to ask the right questions in the right order and prevents you from getting lost in detail metrics before the basic calculation stands.

• Define 30 to 100 realistic user questions for your category and measure your share of voice across several AI systems.
• Estimate the monthly question volume conservatively, if necessary via search volume as an approximation.
• Set low conversion rates from mention to inquiry and from inquiry to deal.
• Set the average value of a deal as contribution margin over the customer lifetime, not as revenue.
• Capture all costs including your own working time at a realistic hourly rate.
• Calculate two scenarios and decide at the pessimistic end.
• Add an origin question at first contact to ground the attribution over time.

Separating fixed and running costs cleanly

Many ROI calculations tip over because they throw one-time setup and permanent operation into the same pot. Separate the two from the start. Fixed costs include the initial analysis of your visibility, setting up the content structure and one-time technical adjustments. This sum occurs once and, for calculation purposes, spreads across the entire runtime. If you charge it in full every month, your ROI looks artificially poor in the first weeks.

The running costs are what recurs every month: maintaining the content, observing the mentions in the answer systems and regular refinement. Calculate these items separately and honestly. A common mistake is not pricing in your own working time. Even if you write it yourself instead of paying, every hour costs you something. Set a realistic hourly rate, otherwise you end up comparing apples with oranges and significantly overestimate your return.

Over what period you may even measure

GEO doesn't work instantly. Answer systems need time to take up new or revised content and factor it into their answers. Whoever calculates the ROI after two weeks is mainly measuring noise. A robust measurement period usually begins only after several months, because visibility builds up in stages and only then reaches a stable level.

Therefore set an evaluation horizon before you start and stick to it. Three checkpoints make sense: a baseline measurement before the start, an interim measurement after about three months and a main evaluation after six to twelve months. This way you recognize whether the curve is rising, stagnating or falling again.

It is important that you keep the same measurement approach across all points. If you change the questions, the systems or the counting method in between, your values are no longer comparable. Document your setup cleanly once and freeze it for the runtime.

Common reasoning errors in the ROI calculation

The first reasoning error is comparison against zero. Without GEO you would not be at zero visibility; you would presumably have received part of the mentions anyway. Subtract this baseline, otherwise you credit yourself with successes that would have arisen even without effort. Always calculate with the increase over your starting position, not with the absolute final value.

The second reasoning error is the inflated revenue value per mention. It is tempting to set a high value, because the result then looks better. Stay conservative and calculate with the lower bound of your estimate. If the ROI is positive even under cautious assumptions, you have a robust result. If it only holds with optimistic figures, the calculation is not sustainable.

The third reasoning error is reducing everything to a single metric. A positive ROI says nothing about the variance. Supplement your calculation with a rough range, so that you know how strongly your result tips when one assumption is off.