GEO ROI Benchmarks: 2026 Report

Marketing teams have been buying generative engine optimization for about two years now, largely on faith. The pitch is easy to believe and hard to check: earn a place in the recommendations people get from ChatGPT, Perplexity, and Google's AI Overviews, and the leads will follow. Vendors report citations and share of voice. Finance departments ask a different question, which is how long the program takes to pay for itself and what each lead ends up costing.

This study was built to answer that second question. The sample is 143 GEO engagements at companies in the United States and Canada, some run in-house and some by agencies, followed across ten quarters that begin with the first programs to launch in early 2024 and close with data through June of this year. It spans seven industries and every size band from ten-person firms to enterprises above 5,000 employees. The one condition for inclusion was revenue attribution already in place before the work began, which is what made it possible to follow a lead from its first AI-referred visit through to booked results instead of stopping at a citation count.

The headline number is payback. The median engagement crossed the point where cumulative value of AI-sourced leads exceeded cumulative program spend at 9.6 months, roughly a quarter sooner than the traditional search campaigns the same companies were running alongside them. The spread behind that median is wide, running from 6.8 months at the fast end to 13.6 at the slow one, and most of the useful detail is in what explains the difference. The sections below break the economics down by industry, company size, deliverable type, and the platform that produced the lead.

GEO ROI Benchmarks by Industry

Of all the ways this dataset can be cut, industry produced the widest variance, which is why it comes first. Two companies can spend the same amount on the same deliverables and sit five months apart on payback, largely because one sells a $9,000 contract and the other sells a $400,000 one. The table below reports the four figures a budget owner tends to want in the same view, for the seven industries with at least eight engagements in the sample.

The spread tracks deal size and sales cycle length more closely than it tracks competitive difficulty. B2B SaaS cleared payback in 6.8 months at a $312 cost per MQL, helped by short evaluation cycles and buyers who ask AI assistants for shortlists by name. Education sits at the other end at 13.6 months while producing the cheapest leads in the sample at $221, since enrollment cycles defer the revenue those leads eventually generate. Legal and financial services carry the two highest acquisition costs of the group and still reach payback inside eleven months, a function of contract value rather than program efficiency.

Cost per AI-Sourced MQL, Q1 2024 to Q2 2026

The figures above are where programs landed by the middle of 2026. They look considerably better than the same programs looked two years earlier, and that trajectory is worth seeing on its own, because it is the reason a GEO budget approved in 2024 on optimistic assumptions has generally aged well.

Acquisition costs fell in nearly every quarter of the study as the assistants matured and marketing teams worked out which content earns citations. The table below tracks the quarterly median cost per AI-sourced MQL in the three industries with continuous data across the full period, and the chart beneath it plots the same three series.

• Cost per AI-sourced MQL fell 51.9% in B2B SaaS across the period, from $648 in Q1 2024 to $312 in Q2 2026.
• Financial services posted the only sustained increase in the series, climbing to $961 in Q2 2024 before declining in eight of the nine quarters that followed.
• The gap between the most and least expensive industry narrowed from $351 in Q1 2024 to $212 in Q2 2026, so the cost advantage of an easier category is compressing.

GEO ROI Benchmarks by Company Size

Industry sets the ceiling on what a GEO program can return. Company size largely determines how quickly a team gets there, though not in the way headcount alone would predict. Budget is only part of the story, since the larger organizations in the sample also carried longer review cycles and more stakeholders per published page. The table below groups engagements into five employee bands and pairs median monthly investment with the three outcomes that spend is judged against.

Payback is fastest in the 1,001 to 5,000 employee band at 7.6 months, and the curve turns back upward above that. Enterprises in the sample spent a median of $26,700 a month and waited 8.9 months, with the delay traceable to approval cycles that slowed publication rather than to weaker content performance. Cost per MQL rises steadily with company size, from $247 to $563, while MQL to SQL conversion improves from 21.4% to a peak of 34.9%. Larger programs pay more per lead and waste fewer of them.

Which Deliverables Produce the Return

The sections above describe what a GEO program returns. This one describes where the return comes from, which is the question most teams get wrong when they plan a content calendar. Every AI-sourced lead in the study was traced back to the page that earned the citation, so the five common deliverable types can be compared on outcomes rather than on how much each costs to produce. The table below covers the formats that appeared across most engagements in the sample.

• Original research and metrics reports earned 34.6% of all AI citations in the sample, the largest share of any deliverable type, at the lowest cost per MQL of the five.
• Comparison and best-of articles produced the fastest first lead at 3.6 months, ahead of every other format, and accounted for a further 27.1% of citations.
• Technical and schema work accounted for 7.3% of citations at $612 per MQL, the weakest return of the categories tracked, though it remains a prerequisite for the pages that do earn citations.

GEO ROI by AI Platform

The last cut in the study was the one that produced the biggest surprise. Marketing teams generally plan GEO around wherever the volume is, on the assumption that a lead is a lead. In this sample, the share of leads a platform sent and the quality of those leads turned out to be close to unrelated. The table below breaks AI-sourced leads down by the platform that produced them, with conversion measured from the first AI-referred visit through to marketing qualified lead.

ChatGPT produced 46.3% of AI-sourced leads in the sample, more than the next three platforms combined, with economics near the middle of the group at $281 per MQL. The more useful numbers sit at the bottom of the volume column. Claude accounted for 4.3% of leads at a 2.88% conversion rate, the highest in the study, and Perplexity was close behind at 2.61%, which suggests both are reached later in a buying process than their traffic share implies. Google AI Overviews inverted the pattern, delivering the second-largest share of leads at the lowest conversion rate and the second-highest cost per MQL, consistent with a surface that answers most questions without sending the reader anywhere. For a team allocating effort, the practical reading is that volume and quality need separate targets.

How This Study Was Conducted

Spend was normalized to trailing twelve-month totals in 2026 dollars so that programs starting at different points in the period could be compared, and engagements with fewer than six months of continuous data were excluded. Payback is measured against the value of those results rather than closed revenue, since roughly a third of the sample had deal cycles longer than the study window. Lead attribution relies on each company's own CRM, which understates AI-sourced volume in the common case where a buyer first encounters a brand inside an AI assistant and later arrives through a branded search, so the cost per MQL figures here should be read as conservative. All figures are medians unless stated otherwise.

Sources

• Aggarwal, Pranjal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan, and Ameet Deshpande. GEO: Generative Engine Optimization. arXiv, November 16, 2023. Princeton, New Jersey.
• Chapekis, Athena, and Anna Lieb. Google Users Are Less Likely to Click on Links When an AI Summary Appears in the Results. Pew Research Center, July 22, 2025. Washington, D.C.
• Crestodina, Andy. AI Traffic Conversion Rates: New Research From 97 B2B Websites and 29M Visits. Orbit Media Studios, July 22, 2026. Chicago, Illinois.
• Santiago, Erica. 2025 CPL and CAC Benchmarks. HubSpot, August 4, 2026. Cambridge, Massachusetts.