AI Search ROI Is Unprovable. Fund It Anyway.

If you are waiting for clean attribution before you fund AI search, you have already lost the quarter. The honest position on ai search roi is that you cannot prove it to last-click standards, and you do not need to: Forrester found that 94% of business buyers now use AI search when buying, rating it above vendor websites, product experts and sales reps. That figure was published on 6 October 2026 in an Ahrefs analysis by Louise Linehan, citing Forrester's research on zero-click B2B buying.

What that means for your number: your buyers are forming a shortlist inside ChatGPT, Gemini, Perplexity, Copilot and Google AI Overviews, and almost none of that shows up in your CRM as an AI touchpoint. If your board reviews channel spend on sourced pipeline alone, AI search looks like a zero. It is not a zero. It is unmeasured, and those are different problems with different fixes.

What is ROI in search engine marketing, and why AI search breaks the formula

ROI in search engine marketing is simple arithmetic: the value you get back, minus what you spent, divided by what you spent. Ahrefs sets out AI search ROI the same way: (AI revenue minus AI cost) divided by AI cost, times 100. Above 0% means the work paid for itself.

The formula is fine. The inputs are the problem. Someone can get a complete answer inside ChatGPT and never visit your site, so there is no session to attribute.

When they do click, the referrer is often stripped and it lands in your analytics as direct traffic. And buyers who act on an AI recommendation rarely tell you that is what happened unless you ask them.

Where AI search differs from organic search roi

Classic organic search roi has a trail: impressions, clicks, sessions, conversions. You can argue about the model, but the data exists. AI search mostly does not produce that trail, which is why b2b organic search roi and AI search ROI should be reported side by side rather than merged into one line.

What you can and cannot track: organic search versus AI search, as at October 2026

Signal	Organic search	AI search (ChatGPT, Gemini, Perplexity, Copilot, AI Overviews)
Impression data	Available in Search Console	No publisher-side impression reporting
Click data	Available	Partial, often stripped to direct traffic
Zero-click influence	Growing	The default behaviour
Best available proxy	Rankings and click share	Citation share and self-reported attribution

How to measure ai search roi without last-click: three proxies that hold up

This is the practical answer to how to track roi from ai search optimisation when the platforms will not hand you the data. Run three proxies in parallel and read them together. One on its own is noise. Three moving in the same direction is evidence.

One: share of answer. Track how often your brand is cited across a fixed set of buying-intent prompts, per engine, measured monthly. Own it in-house, SEO or GEO lead, and freeze the prompt list so quarter-on-quarter comparison means something.

Two: self-reported attribution. Add an open "how did you hear about us" field to your demo and trial forms. It is the cheapest measurement project you will run this year, it takes a sprint, and it is the only place a buyer will tell you ChatGPT sent them.

Three: branded search and direct traffic lift. If AI assistants are recommending you, demand shows up as people searching your name or typing your URL. Baseline it before you start the work, then watch the delta.

What AI search work actually costs, and how to split it

Ahrefs offers a clean test for separating pure AI costs from shared ones: would you still pay for this if you stopped AI search work tomorrow? If the answer is no, count all of it. If yes, count a share.

Their worked example is useful because it is unfussy. A content team of three costing $54,000 a quarter, spending roughly a third of their time on AI search work, gives you $18,000 of AI search cost. Pair that with $45,000 of attributed AI revenue and you have the 150% return Ahrefs uses to illustrate the formula.

Label	Value
AI revenue (US$)	45000
AI cost (US$)	18000
Net return (US$)	27000

Do not spend a fortnight apportioning every licence to the cent. A ballpark that you calculate the same way every quarter beats a precise number you only ever produce once. Consistency is what makes the trend readable.

What we would not do yet

We would not build a custom multi-touch attribution model for AI assistants in 2026. The referral data is too patchy to support it, and the engineering time buys you a model that is confidently wrong.

We would also not report AI search as a sourced-pipeline line in the board pack. Report it as influenced demand with the proxies attached, and say plainly which parts are estimated. Leadership does not need a perfect number. They need a sensible estimate they can compare over time.

If you want the measurement plumbing designed properly, that sits in our measurement and attribution advisory. The visibility side, including citation tracking and content built to be quoted, sits under organic search: SEO, AEO and GEO. We have also written about how brand visibility in AI search behaves like a staircase rather than a switch, which is the pattern these proxies tend to reveal.

What to do about ai search roi this quarter

Start your ai search roi baseline this month, because every week without one makes the next budget conversation harder. Pick 30 buying-intent prompts, measure citation share across ChatGPT, Gemini, Perplexity, Copilot and AI Overviews, ship the self-reported attribution field, and record branded search volume today. Then do nothing clever for 90 days and read the trend. That is a defensible roi search marketing case, and it is the one we would put in front of a board.

Cover photo by Igor Omilaev on Unsplash