Calculating the ROI of GEO: a break-even sum you can do with your own numbers

Why the return is hard to read from analytics

Most of what an assistant does for a business leaves no trail. A buyer asks for an electrician, reads three names, and phones one of them; nothing in that business's analytics records the answer that sent the call. Referral traffic from the assistants counts only the buyers who clicked through, which is why zero-click search is the normal case rather than the exception. An ROI built on traffic alone measures the smallest part of the effect and calls it the whole.

So the calculation has to be built from the other end: from what the work costs, from what a customer is worth, and from a count of new customers who say an assistant sent them. The first is known, the second is in your own books, and the third takes a question at intake and a few months of answers.

Start from the cost, which you know

If you hire the work out, the cost is a price. Ours are the visibility check at $0; the AI Visibility Audit at $490, one-time, credited in full against optimization if you go ahead; GEO optimization from $1,900 for a fixed scope the audit sets; and monitoring at $290 a month, which you can cancel in any month. A first year that includes the audit, the optimization and monitoring costs the optimization price (the audit being credited) plus twelve months of monitoring.

If you do the work yourself, the cost is hours: the afternoons it takes to fix access, identity and markup, the days it takes to write answer pages, and the morning a month the panel takes to read. Count them at what an hour of your time is worth to the business, not at zero. A fix that costs nothing but the owner's evenings still has a price.

The break-even sum

Divide the cost by what one new customer is worth to you. The answer is the number of extra customers the work has to bring in to pay for itself. Written with brackets for your own figures:

Break-even customers = [cost of the work for the period]
                      / [margin one new customer brings in that period]

Example, with your numbers in the brackets:
[cost] / [margin per customer] = [customers needed]

Two choices decide whether the sum means anything. Use margin, not revenue: a job billed at a price that leaves little after materials and labor pays back far less than its invoice. And match the period: if a customer typically returns for years, count the margin you can expect in the first year for a first-year cost, and say so, rather than counting a lifetime against a single month's bill.

Then read the result as a plausibility test. A break-even of a handful of customers a year, for a business that wins dozens a month, is a low bar; the question becomes whether the fixes are right, not whether they are affordable. A break-even larger than the business's whole market is a clear no, whatever anyone promises.

The numbers only you have

The fourth row is the one to start this week, before any work begins, because it needs months of answers to mean anything. Add "an AI assistant (ChatGPT, Claude, Gemini, Perplexity)" to the question on your intake form, booking page or new-customer call about how people found you, and keep the answers. The fifth row is the panel described in setting up prompt monitoring.

The calculator's model, and its weakest input

Our ROI calculator runs a simple model of the opposite question, what being left out costs: customers lost equals potential customers, times the share who ask an assistant, times the share of answers that leave you out; revenue lost equals customers lost times the value of an order. Every input is yours, and the page says so.

The weak input is the share who ask an assistant. Until your intake question has a few months of answers, it is a guess, and the output inherits the guess. Run the calculator twice, with a low guess and a high one, and treat the space between the two results as the honest answer. The other input, how often the answers leave you out, comes from the panel, and it is the one the work is meant to change.

What can be counted, and what cannot

• Counted: new customers who say an assistant sent them, from the intake question.
• Counted: how often you are named, recommended and first, from the same panel each month, as the seven GEO KPIs set out.
• Counted, as a floor: visits referred by the assistants, where your analytics can see them.
• Not counted: the buyer who read your name, did not mention it, and booked anyway.
• Not counted: the buyer who would have found you without any assistant at all.

The last two cut in opposite directions, which is why no honest calculation claims precision. What it can show is direction over time: a named rate that rises on the same questions, beside an intake count that rises with it, is as close to proof as this channel allows. GEO reporting for management shows how to put both on one page.

When GEO does not pay

• When the break-even is larger than the market. A business with a thin margin in a small town may need more customers than the town holds.
• When the business is full and staying full. The work can still be worth doing for accuracy, so that answers state the right hours and prices, but not for growth.
• When your buyers do not ask assistants. The intake question shows it once it has a few months of answers; believe it.
• When the site cannot be changed. A page builder that will not let you edit text or add markup caps what any fix can reach.
• When the fixes are small enough to do yourself. Then the cost is a few afternoons, the break-even is low, and paying someone else is the part that does not pay.