Attribution Can't See GEO: How to Measure Narrative Authority Instead

TL;DR

"How do we prove it works?" is the question almost every brand asks us once we start working on its AI visibility. It deserves a better answer than "look at the AI referral traffic".

GEO is a brand channel. Its decisive moment happens when an engine names you, usually without a click. Judging it with click-based attribution means condemning it in advance.

At PingPrime we measure narrative authority on four layers: Presence (are you named?), Narrative (are you described correctly?), Recall (do buyers come looking for you?) and Revenue (does it reach the pipeline?). Each has a blind spot; together they tell a story you can defend.

Proof is built over time, with a fixed prompt panel, a baseline taken before any work starts, and a control market or language to compare against.

The question every brand ends up asking

When a brand starts working on its AI visibility, the first conversation is about presence: where does the brand appear in ChatGPT, Perplexity or Google's AI Overviews, and where do competitors appear instead? The second conversation, almost always, is about proof. A CMO has to defend the budget. A CFO wants to know what it returns. And the usual dashboards have very little to say.

What I observe systematically is the same reflex: teams open their analytics, look for an "AI" line in the acquisition report, find a sliver of referral traffic, and conclude that GEO is marginal. That conclusion is wrong, and it is wrong for a structural reason.

We have argued that GEO is a brand channel disguised as a performance channel. Measurement is where that disguise becomes dangerous. A brand channel judged with a performance instrument will always look like a bad investment.

Why the click cannot judge a brand channel

Look at how a buyer now reaches a vendor through an AI engine.

Before: search → click → convert

Now: prompt → synthesize → direct visit

The buyer asks a question. The engine returns a synthesis with two or three names. The buyer closes the tab, talks it over internally, and a week later types one of those names into the browser. Analytics records a direct visit. The moment that put the brand on the shortlist appears nowhere.

And that moment often happens even before the answer is written. As we showed in our analysis of fan-out queries, engines increasingly write brand names into their own background searches before reading a single page. The shortlist forms in a place no tracking pixel can reach.

The scale of the blind spot is starting to be documented. In a case published by Graphite, the automation platform n8n compared two readings of the same conversions: GA4 last-touch data credited AI answer engines with about 1%, while a post-conversion "how did you hear about us?" survey credited them with about 9%. A gap of roughly ten times. Kevin Indig cites the case in his latest Growth Memo, and his diagnosis matches ours: attribution was built to allocate advertising budgets with cookies and clicks. It was never designed to measure brand channels, and GEO is one.

What we measure instead: the four layers of narrative authority

At PingPrime, we don't try to find the one number that proves GEO. It doesn't exist yet, and pretending otherwise would undermine the very credibility we help clients build. Instead, we read four layers together. Each one answers a different question, and each one has a blind spot the others cover.

PingPrime measurement framework.

The reading rule is simple. When the four layers move in the same direction, confidence rises. When they diverge, the divergence itself is the diagnosis: strong presence with weak recall usually means the brand is named but not memorable; strong recall with weak presence means the market knows you, but the engines haven't caught up yet.

Independent work points the same way. Indig calls the approach triangulation, combining an exposure metric, a behavioral signal and a business outcome, and reports a medium-strong correlation between AI share of voice and conversions in a user behavior study run with Profound. It is not proof of causality, and we never present it as such. But it makes share of voice a reasonable proxy when most conversions arrive through the direct channel.

The layer everyone forgets: being cited correctly

Most GEO tracking stops at layer one: are we mentioned or not? That is not enough. Being cited is not the same as being cited correctly.

A brand named in the wrong category competes against the wrong players. A brand described with last year's offer sends buyers to a product that no longer exists. A brand reduced to one attribute when it has five loses every question about the other four. In each case, the presence metric looks healthy while the narrative is quietly working against the business.

That is why the Narrative layer sits at the heart of how we work, and of the exercise that opens our masterclasses. Ask the engines to describe your brand in one sentence. Then put that answer next to your own canonical sentence, the one that says who you are, what you do and for whom. The gap between the two is your first GEO roadmap. It is also, over time, one of the most telling measures of progress: when the engines start describing you in your own terms, narrative authority is taking hold.

How we build proof over time

Measurement is not a report you pull at the end. It is a design decision you make at the start. Three principles guide it:

Take the baseline before touching anything

We define the prompt panel with the client, typically the questions their buyers actually ask engines across the decision journey, and measure all four layers before a single page or press relationship changes. Without a baseline, every later number is an anecdote.

Measure over repeated runs, not snapshots

AI answers are volatile. The same prompt can return a different list of brands from one run to the next, and behavior varies strongly from one engine to another. A single screenshot proves nothing. We read presence across many runs and several engines before drawing any conclusion.

Keep a point of comparison

The strongest proof is a counterfactual: what happened where we worked, compared with where we did not yet. Phased rollouts make this possible without sacrificing anything. Work one product line, one market or one language first, and use the others as a temporary control. Belgium offers a natural setup: French and Dutch prompts, sources and press ecosystems can be worked in sequence, and the gap between them shows what the program changed. This kind of incrementality thinking is common in paid media, yet rarely combined with other methods: according to the IAB State of Data 2026 report, based on more than 400 senior US planning and analytics decision-makers, only 39% use incrementality, attribution and marketing mix modeling together.

Attribution still has a job, just not the judge's seat

None of this means abandoning attribution. AI referral traffic is incomplete, but it remains one of the few AI search signals you can observe directly, and it belongs in layer three. The mistake is letting it rule on a channel whose main effect happens before the click.

George Bonaci, who leads growth at Ramp, a US fintech valued at $44 billion in June 2026, says it well in Indig's piece: attribution has "become a crutch replacing critical thinking". His team still runs an attribution model, while constantly discussing its limits. He also admits that relying on attribution alone would have led them to cut their brand marketing, events and direct mail. That is exactly the decision many teams are about to make about GEO, for the same reason.

What to put in front of your CFO

If you need to defend a GEO budget this quarter, here is what we recommend

1. Fix the prompt panel and the baseline. Agree on the buyer questions that matter and measure all four layers now, before any work starts.

2. Write your canonical sentence. It becomes the reference against which the Narrative layer is scored.

3. Add AI to self-reported discovery. Put "ChatGPT / AI assistant" as an explicit option in your forms, and have sales ask the question on every first call. In the n8n case, that single question revealed about nine times more AI-driven conversions than analytics did.

4. Choose a control. A market, a language or a product line you will work later. Decide it before the program begins, not after.

5. Report a system, not a number. Show the four layers side by side, with their blind spots stated. Honest uncertainty is more defensible than a precise number that measures the wrong thing.

Stop asking the click to judge the brand

Nobody, us included, has the perfect GEO metric yet. The field is young and the instruments are still being built. But one thing is already clear: narrative authority compounds slowly, signal by signal, source by source. Judging it by the last click is a way of cutting it before it has time to work.

The question worth bringing to your next budget review isn't "how many conversions did ChatGPT send us?"

It's "are the engines naming us, describing us the way we describe ourselves, and is that showing up in how buyers look for us?"

FAQ

What is narrative authority measurement?

A way of assessing a brand's standing in AI answers across four layers: Presence (is the brand named?), Narrative (is it described correctly?), Recall (do buyers search for it afterwards?) and Revenue (does it reach the pipeline?). Each layer has a blind spot; read together, they give a defensible picture that no single metric provides.

Why can't click-based attribution measure GEO?

Because GEO works mostly before and without the click. An engine names the brand, the buyer remembers it and returns later through a direct visit or a branded search. Attribution was built to allocate ad budgets with cookies and clicks. In one case documented by Graphite, n8n's last-touch analytics credited AI answer engines with about 1% of conversions, while a post-conversion survey credited them with about 9%.

Why measure how AI engines describe a brand, not just whether they cite it?

Because being cited is not the same as being cited correctly. A brand placed in the wrong category, described with an outdated offer or reduced to one attribute has presence without authority. Comparing the engines' description with the brand's canonical sentence shows the gap and tracks progress over time.

How can you prove a GEO program works?

Take a baseline on a fixed prompt panel before any work starts, measure over repeated runs and several engines, and keep a point of comparison: a market, language or product line worked later serves as a temporary control. The difference between the two shows what the program changed.

Is AI referral traffic still worth tracking?

Yes. It is one of the few AI search signals you can observe directly. But it is incomplete, because most AI-influenced buyers arrive later through direct visits or branded searches. Use it as one input among others, never as the only yardstick.

Sources

• Kevin Indig, "The usefulness of attribution for AI Search", Growth Memo, 28 September 2026 (featuring George Bonaci, VP Growth at Ramp).

• Graphite, "Last-Touch Attribution Only Captures 10% of n8n's AEO Conversions", Five Percent, 2026.

• IAB and BWG Global, State of Data 2026: The AI-Powered Measurement Transformation, February 2026.

• CNBC, "Ramp hits $44 billion valuation as companies look to rein in AI spending", 4 June 2026.

• Profound, "The shortlist is the new shelf", user behavior study with Kevin Indig

-----

How this was written: Topic, angle, convictions: mine. AI assists the writing. I make the call.

Sabrina Bulteau is the co-founder of PingPrime.ai, GEO Expert and specialist in Narrative Authority in AI Search. She helps brands, institutions and media become the reference AI engines trust, cite, and repeat, not just one option among many, by working both sides of the signal: on-site (structure, narratives, architecture) and off-site (earned media, platforms, co-citations, sector press). She previously co-founded Be Connect (acquired by iO Group) and Sench, is active within CEC Belgium, and brings 25 years of experience in digital growth and strategic positioning.