How to measure AI search ROI when there's no click to track.

AI-generated answers in ChatGPT, Claude, and Gemini typically don't link out to the sources they draw from. When a buyer gets an answer that mentions your brand favourably and then visits your website directly, that visit appears in your analytics as direct traffic, not as "ChatGPT referral." When they come back days later after doing more research, it's even harder to connect. The dark funnel problem that has existed in B2B for years (buyers who form opinions before they ever identify themselves) is amplified by AI search because the AI answer is itself a high-quality, synthesised response that may fully satisfy the buyer's question without triggering a click at all.

Perplexity and Google AI Overviews do link out, and their referral traffic is attributable. But the volume is lower than you'd expect from the influence they actually exert, because many queries are answered without a click. You're measuring the fraction of AI-influenced buyers who clicked, not the full population of AI-influenced buyers.

Accepting this constraint is the starting point for useful measurement. You're not going to get click-level attribution for AI search in the same way you get it for paid search. What you can get is a set of correlated signals that, together, give you confidence that AI search investment is driving business outcomes.

The leading indicator. Build a prompt set of 50–100 buyer-intent queries for your category. Run it monthly across the engines you care about. Track your citation rate: the percentage of prompts where your brand appears in the answer. This is an output metric: it tells you whether your brand's visibility in AI-mediated research conversations is improving.

Citation share is the closest equivalent to keyword position tracking in traditional SEO: it doesn't tell you about revenue, but it tells you about the upstream visibility that influences revenue. I covered the mechanics of building this in detail in why citation share is the new keyword ranking.

Use citation share as the primary input to your GEO programme review. If it's moving in the right direction, your work is moving the visibility needle. If it's flat while you're investing in interventions, something isn't working and you need to dig into which interventions and which engines.

Branded search (people searching specifically for your company name in Google) is the best proxy for brand awareness growth that is reliably measurable in traditional analytics. AI search creates brand awareness by naming your company in answers to category research queries. If your citation share is growing, and buyers are hearing your name in AI-generated answers, some fraction of them will search for you directly. Branded search volume in Google Search Console is the downstream signal of that AI-awareness funnel.

The attribution chain: AI answer mentions your brand → buyer remembers the name → buyer searches your brand directly → you see an increase in branded Google search impressions. This chain is not instant. Expect a 4–8 week lag between a meaningful increase in citation share and a measurable lift in branded search. But it is real, and it is measurable.

Isolate branded search volume from non-branded in your GSC data. Track the trend monthly alongside citation share. Look for the correlation over 3–6 months. You won't have a clean causal proof. Other things affect branded search too, but a consistent co-movement between citation share improvement and branded search growth is a strong signal that AI search is doing what it's supposed to do.

The most direct measurement signal for AI search impact is asking your buyers. Add a "how did you first hear about us?" question to your demo request form, your trial signup, or your first sales call. Include "AI assistant / ChatGPT / Perplexity" as an explicit option alongside the standard sources (Google search, LinkedIn, referral, etc.).

This is not a novel measurement idea. "How did you hear about us?" has been a standard tool in marketing attribution for decades. What's new is adding AI assistants as an explicit option, because buyers who found you through an AI answer won't self-report that if the option isn't there. Without the explicit option, those responses collapse into "organic" or "word of mouth" and the AI signal is lost.

The data from these surveys is imperfect: buyers don't always remember or accurately report how they first heard about something, but it gives you directional signal about what proportion of your pipeline has AI search in the attribution chain. Even a rough number is more useful than no number for making investment decisions.

Perplexity passes referral data, and Google AI Overviews clicks appear in Search Console. These are the attributable slice of your AI search traffic: the buyers who clicked through from an AI answer to your site. Monitor both, track conversion rates from each, and compare to your other organic sources.

The referral traffic from these sources tends to be small in absolute volume but high in intent. A buyer who clicked through from a Perplexity answer that specifically recommended you has already cleared a research hurdle that a first-time organic visitor hasn't. That conversion rate difference is a useful argument for AI search investment that leadership can understand.

Don't overweight these numbers as the primary success metric. They represent only the fraction of AI-influenced buyers who clicked, not all of them. But track them as a fourth data point that contributes to the overall picture.

The reporting cadence that works: monthly citation share update (operational), quarterly aggregated report combining all four signals (strategic). The monthly update keeps the team honest about whether interventions are working. The quarterly report is the one you bring to leadership, because 90 days is long enough to see meaningful trends and short enough to course-correct before budget cycles close.

The quarterly report structure I use: citation share trend (with competitive benchmarks where available) → branded search volume trend → pipeline survey results → referral traffic from attributable AI sources → interpretation: what changed, what drove it, what to do next. Five slides or five minutes of talking points. Not more than that. The goal is a clear, defensible position on whether the programme is working, not a comprehensive attribution exercise.

One honest caveat: in the early months of a GEO programme, you won't have enough data to make strong claims. The right thing to say is "citation share is up 8 percentage points, branded search is flat, and we won't have meaningful pipeline signal for another quarter." That's a more credible position than inflating early results. Leadership teams that have been burned by overpromised SEO ROI in the past respond better to honest baselines than to optimistic projections.