Can You Measure GEO ROI When AI Sends So Little Trackable Traffic?

A homeowner asks ChatGPT which companies can diagnose drainage problems around a house. They see your company name, search for it on Google, read a few reviews, and call the next morning. Your website analytics might show a Google visit. Your phone log might just show a call. If the homeowner tells your scheduler where they first saw your name, you finally have a record of the AI recommendation.

That’s a measurement problem for generative engine optimization, or GEO, the work of improving how your business shows up in AI-generated answers. A buyer can get value from a recommendation without clicking a link. And even if someone does click, that visit may be days or weeks removed from a signed deal.

You can still measure what the work contributes. Begin with the evidence you can collect, follow it into your sales process, and be clear about how much credit it supports.

Check the visits you can actually see

In Google Analytics 4, open the Traffic acquisition report and look at Session source or Session source / medium. OpenAI’s publisher guidance says ChatGPT adds utm_source=chatgpt.com to referral URLs from ChatGPT search results. Look for that source and review the landing pages in a report or exploration that supports the combination. If a guide to diagnosing drainage problems gets those visits, you have a page worth examining more closely. Check what the page helps a reader do and whether there’s a clear way to request an assessment.

For a Colorado Springs SEO campaign, I’d compare those observations with the performance of the local service pages and the inquiries coming into the business. Google says traffic from its AI features is included in the overall Web search performance data in Search Console. The aggregate Web total on its own cannot tell you how much of a change came from an AI feature. Watch it for changes in search demand and page performance, and use lead records to understand who called and why.

A session gives you a visit to investigate. A phone button click gives you an action on the site. To know whether either turned into a lead, check for an actual call or a completed form. To know whether the lead paid off, follow it through to a booked and completed job.

Ask customers where the journey started

Train whoever handles new inquiries to ask, “Where did you first hear about us?” Ask it as an open question. If someone says “online,” a short follow-up such as “Do you remember the site, app, or search you used?” can help. Some people will remember the product name. Others will only remember that they saw an AI-generated list. Record the detail the customer gives you and leave the rest unknown.

A second question can capture what prompted the call. Someone may first see you in an AI answer but decide to contact you after reading a review that describes a job like theirs. Both details are worth recording. They tell you how the buyer discovered you and what gave them enough confidence to take the next step.

Put these answers on the same lead record you use for calls, estimates, and jobs. If a customer submits a form after clicking a trackable ChatGPT link and later says they found you through ChatGPT, you have two observations about one lead. Give the lead one sales record. Keep both pieces of evidence attached to it.

Use prompt tracking to see whether the work changed visibility

Start with the questions customers ask before they buy. A drainage company might test “Who can assess water pooling next to a house in Colorado Springs?” and “Which local contractors can diagnose yard drainage problems before proposing a fix?” Choose a small group you can repeat and record the exact wording, platform, date, and location context.

Save whether the answer named your business, recommended it for the job, or linked a page. Open citations that appear relevant. If an answer names your company for an assessment but the linked page describes installation only, check whether your public information gives prospects a clear account of what you actually offer.

Use these observations to decide what to improve. If a buyer asks about assessment and your page jumps straight to installation, explain what happens on the first visit and what customers receive. Then retest. If you appear in six of twelve answers after appearing in three of twelve in the prior round, report that change for the prompts you sampled. You still need intake and sales data to understand whether more qualified customers are finding you in the market.

Calculate an observed return using contribution

Count dollar value only after you know what happened to the leads. Use the revenue from completed work, subtract the direct cost of delivering those jobs, and compare that contribution with the cost of the GEO work you’re evaluating. Include staff or agency time and tools you can reasonably allocate. If the same work also serves your broader SEO program, record how you allocated the cost so someone can reproduce your calculation next month.

Here’s a hypothetical example. A contractor spends $1,500 on a GEO project. During the period it reviews, two completed jobs have documented AI involvement. One originated from a trackable AI referral that the contractor connected to a quote request. The other customer said they first heard about the company in an AI answer and later called after looking it up. Together, those jobs generated $7,000 in revenue and cost $4,500 to deliver, leaving $2,500 in contribution before the GEO expense.

Against $1,500 in GEO cost, that leaves $1,000 in observed net contribution. Divide $1,000 by $1,500 and you get about 67% observed ROI if you credit both jobs. If the trackable job alone contributed $1,600, counting only that job produces about 7% observed ROI. Put the crediting rule next to the number. The two figures let the owner see what changes when the self-reported job is included.

Attribution tracks the role AI appears to have played in those sales. To estimate how many sales the work added, you’d need a stronger comparison. For instance, you might update one group of service pages before a comparable group, check how both performed beforehand, and monitor qualified leads after the rollout. Differences in demand, seasonality, pricing, and other marketing activity still matter. With a handful of jobs, give the owner the underlying counts and costs so they can judge how much confidence to put in the percentage.

Build a monthly report you can make decisions from

Gather the figures in one place. I’d want to see the monitored prompt results, identifiable AI visits, unique qualified leads with documented AI involvement, completed jobs from those leads, job contribution, and the amount spent. On each relevant lead, keep the source of the evidence and any other touchpoints the customer mentioned. Add a line for current estimates that haven’t closed yet, then review them when the normal sales cycle has had time to run.

Decide what you’ll do with the information before you look at the next report. If visibility rises and the leads are a bad fit, review which jobs AI answers associate with the business and clarify the scope on your site. If only a few buyers mention AI but those leads become profitable jobs, keep tracking that service and look for other buyer questions with similar intent. If most leads have an unknown source, improve intake before drawing a strong conclusion from one month’s numbers.

Start with one service, a handful of buyer questions, and a source field your team will actually fill in. Follow the inquiries to completed work. That will tell you more about whether GEO is paying off than any referral chart on its own.

Explore More