Can you actually measure ROI on AI marketing?
Standard analytics sees a fraction of it. Here is what last-click attribution misses, what the hidden traffic is doing, and what you can honestly track.
Partly. A clean last-click ROI figure is out of reach, because most of the influence — a long research journey, a name remembered, branded search later — never leaves a tracked footprint; by some estimates last-click captures only 10–20% of the real return. What can be measured today are leading indicators: how often AI names you, the referral traffic that does carry a label, and branded-demand lift over time. The honest reading of AI-marketing ROI is directional, the way brand and PR have always been measured.
What does last-click attribution actually miss?
The instrument most teams use to judge marketing was built for a journey that no longer happens.
In 2025, Pew Research measured something narrow and telling: when an AI summary sits at the top of a Google results page, people click a link 8% of the time, against 15% when there is no summary. By early 2026, roughly two-thirds of searches ended with no click at all.
Last-click attribution assigns the entire credit for a sale to the final tracked touch before it. That logic held when a buyer found a link, clicked it, and converted in something close to one session. It holds less well when the decisive moment is an AI answer the buyer reads and never clicks.
The gap this opens is wide. One GEO measurement framework estimates that standard last-click attribution captures only 10–20% of the real return from AI visibility; the remainder sits in influenced pipeline, branded-search lift, and research that happened weeks before any session your analytics could see. A brand named consistently across a buyer’s long research stretch may never produce a single attributable visit, and still have shaped the outcome.
It is worth being precise about what is failing, because it is not the channel and not, strictly, the analytics. The tool is doing what it was designed to do: credit the last tracked step. The trouble is that for an AI-influenced purchase, the step that mattered most was never tracked — it sat inside a generated answer, on a surface the marketer does not own and cannot tag. The number that comes back is not wrong so much as incomplete, and the size of what it omits is the whole question.
Why doesn’t analytics show the AI traffic?
A large share of AI-driven visits arrive without the one piece of information attribution depends on.
When a model recommends a brand inside an answer, the visitor who follows up often does so later and by another route — typing the name into a browser, or searching for it directly. By some analyses, around 70% of AI-referred visits carry no referrer header, so they land in analytics as “direct” or simply unattributed.
This is usually described as dark traffic: real visits the dashboard cannot label. It is easy to read as a rounding error. The data suggests it is closer to the opposite.
In a study of more than 500 high-value topics published in 2025, Semrush found that a visitor arriving from an AI source converts at roughly 4.4 times the rate of one arriving from traditional organic search. The plausible reason is mundane: by the time someone acts on an AI recommendation, the comparison is already done. The part of the traffic the dashboard undercounts, in other words, tends to be the part that converts best.
What can you honestly track today?
Roughly four signals, ordered from most trustworthy to least.
What can be measured today — and how far to trust it
| Signal | What it tracks | How far to trust it |
|---|---|---|
| Citation rate & share of voice | How often AI names you, versus competitors, across the questions buyers ask | High — measured directly; the earliest clean signal |
| AI referral traffic | Labelled sessions from AI sources in your analytics | Low — most arrive with no referrer, hidden in ‘direct’ |
| Conversion of AI-referred visitors | How the labelled AI traffic performs once on site | Medium — small samples, but they tend to convert well |
| Branded-demand lift | Branded search, ‘how did you hear about us?’, influenced pipeline | Medium — directional; needs correlation over a quarter |
The order is worth pausing on, because it inverts the usual hierarchy. The signal a finance team trusts least — visibility, a soft-sounding word — is the most reliable one here, because it is measured directly: run a fixed set of category questions across the engines each month and record who gets named. The signal a finance team trusts most — tracked traffic tied to revenue — is the least reliable for this channel, because the footprints are missing for the reasons above.
None of these four is a clean ROI figure, and presenting them as one would be dishonest. Together they describe whether the channel is working: named more often, drawing better-converting visits, lifting branded demand over a quarter.
What is the single most useful number?
If only one signal can be tracked, track your share of the answer.
Citation rate and share of voice — how often AI names you, against competitors, across the questions buyers actually ask — behave much like “share of search” in classic brand measurement. They tend to move before revenue does, which makes them a leading indicator rather than a lagging one.
The number is measurable now, it is inherently competitive, and it does not depend on a referrer header surviving the journey. When share of the answer climbs while a competitor’s slips, sales tend to follow, even where no single one can be traced back to a citation.
How is this different from measuring SEO ROI?
Reach for the SEO instrument and the channel reads as worthless when it isn’t.
SEO ROI rests on tracked clicks and rankings, and that foundation is sound when most journeys produce a click. AI visibility often influences a decision without producing a click at all, so a click-based measure undercounts it badly — not because the return is small, but because the meter is reading the wrong quantity.
The closer analogue is brand or PR. A well-placed feature or a strong brand campaign rarely yields a clean last-click number either, and demanding one tends to make a team undervalue the work rather than measure it. The same discipline applies here: leading indicators, plus branded-demand lift, judged over quarters rather than days.
That shift in instrument matters more than it sounds, because the wrong meter does not simply under-report — it quietly steers the budget. A team that judges AI visibility on tracked clicks alone will tend to conclude the channel is marginal, defund it, and cede the answer to whichever competitor was patient enough to measure differently. The error is not in the spending; it is in the scoreboard the spending is judged against.
So what does an honest scoreboard look like?
Fewer numbers, read over a longer window, with the soft ones up top.
- Track share of the answer every month. Run a fixed set of buyer questions across the engines and record who gets named. This is the earliest, cleanest signal you have.
- Watch branded search against direct traffic. When direct climbs faster than branded search, the likeliest explanation is AI-driven visits your analytics cannot label.
- Add a “how did you hear about us?” line. Self-reported attribution recovers some of what the missing referrer threw away — cheap, and reasonably accurate at scale.
- Correlate rather than attribute. Tie rising share of voice to rising branded demand over a quarter, and stop hunting for a last-click line that will not exist.
Read that way, the scoreboard is a rising share of voice alongside rising branded demand — directional, compounding, and honest about its own limits. It is the kind of measurement a coordinated approach is built to produce. For where to put the budget behind it, our H2-2026 budget guide picks up the operational side.
Frequently asked questions
Can you measure ROI on AI marketing in 2026?
Partly. A clean last-click ROI figure is out of reach, because most of the impact — a long research journey, a name remembered, branded search later — never leaves a tracked footprint; by some estimates last-click captures only 10–20% of the real return. What can be measured are leading indicators — citation rate, share of voice, branded-demand lift — read over time.
Why doesn’t Google Analytics show my AI traffic?
Because a large share of AI-referred visits arrive with no referrer header — by some analyses around 70% — so they land in analytics as ‘direct’ or unattributed. Many visitors never click a tracked link at all; they read the answer, remember the name, and return later by another route.
Does AI-driven traffic convert better?
The available data suggests it does. A 2025 Semrush study of more than 500 topics found that visitors arriving from an AI source convert at roughly 4.4 times the rate of those from traditional organic search — likely because the comparison is already done by the time they act.
What is the single most useful metric for AI marketing?
Share of the answer — citation rate and share of voice across the questions buyers ask. It is measured directly, it is competitive, and it tends to move before revenue does, which makes it the most reliable early signal that the work is landing.
How is this different from measuring SEO ROI?
SEO ROI rests on tracked clicks and rankings. AI visibility often influences a decision without producing a click, so click-based attribution undercounts it. It is measured more like brand or PR: leading indicators plus branded-demand lift, judged over quarters.
- Pew Research — people click less when an AI summary appears (8% vs 15%, 2025)
- Search Engine Land — Google zero-click searches reach roughly two-thirds (2026)
- Semrush — AI search visitors convert ~4.4x better than traditional organic (2025)
- Superlines — how to measure the ROI of AI search optimization (GEO ROI framework, 2026)
- Discovered Labs — AI Overviews traffic impact & pipeline attribution (~70% no-referrer)
- Discovered Labs — GEO metrics: what KPIs matter and how to track them (2026)
- Semrush — the most-cited domains in AI answers (2025)
Start with what you can measure
Whatever your budget, the first move is the same: see where you stand. White Wood runs a free AI-visibility report that shows exactly where AI names you — and where it names someone else — across every engine. No strings.