Most teams measure answer engine optimization (AEO) the way they measure search: pull up GA4, sort by source, count the sessions that clicked through from ChatGPT or Perplexity — then conclude AI discovery is a rounding error and move on.

That conclusion is wrong, because it mismatches how the channel works against how you count it. On the Distribution Podcast, Ravish Agrawal of Gamma framed it in one line that reorganizes the whole problem: "As a channel, it behaves like Google. But as measurement, it behaves like Facebook." Execution looks like SEO; attribution looks like paid social. Measure it like SEO and you'll systematically under-count it.

TL;DR

AEO is a demand channel that hides from last-click analytics. GA4 only registers the visits where a user clicks a link out of the AI answer. Everything else is invisible.
The click-through is the small part. Ravish Agrawal of Gamma reports that even as Gamma's ChatGPT-referred traffic grew roughly 8x, click-through share stayed under 1% — the influence is real, the clicks aren't the whole story.
Treat it like Facebook, not Google. Paid social taught marketers to trust view-through and self-reported attribution because most impact never shows up as a last click. AEO needs the same instrumentation.
Three metrics matter: AI referral sessions (the floor), self-reported discovery (the true signal), and the correlation between AI mentions and branded/assisted demand.
The cheapest fix is a survey question. Add ChatGPT, Perplexity, Gemini, Claude, and Copilot as options in your "How did you hear about us?" field and you recover most of the signal GA4 drops.

The measurement gap

Search and AI answers feel similar, which is why teams reach for their search dashboards. But the two resolve into analytics very differently. With traditional search, the answer is a list of links and the next action is almost always a click. The click carries a referrer, GA4 logs a session, and "impression leads to click leads to session" holds up well enough to build reporting on.

AI answers break that chain. The model synthesizes an answer inline. Your brand can be named, described, compared, and recommended inside the response without the user ever leaving the chat. When they do act, they often act later — opening a new tab and typing your name directly, or searching your brand on Google. The discovery happened in the AI answer; the recorded session looks like direct or branded search.

This is exactly why the click-through numbers look so discouraging. Per Ravish Agrawal of Gamma, even while ChatGPT-referred traffic climbed roughly 8x, click-through share stayed under 1%. (He was describing Gamma's own experience over an unspecified window; treat the figures as directional, and note the denominator behind "under 1%" was left unclear on the podcast.) The takeaway is not "AI referrals are tiny." It's that the click is the visible tip of a much larger iceberg of influence you're not counting.

Execute like Google, measure like Facebook

Hold the two halves of Ravish Agrawal's line separately, because each points to a different discipline.

As a channel, it behaves like Google. Getting recommended by AI engines rhymes with SEO: you earn citations, show up where the model looks, and structure content so it can be lifted into an answer. If you already do organic discovery well, most of that muscle transfers. This is the answer engine optimization side of the house — the production and distribution work.

As measurement, it behaves like Facebook. Here the SEO instinct fails you. Paid social went through this exact reckoning a decade ago: marketers learned that last-click attribution wildly under-credited Facebook because most of its impact was view-through — people saw something, didn't click, and converted later through another door. The industry responded with view-through windows, incrementality tests, and self-reported attribution surveys instead of pretending the last click told the whole story.

AEO sits in the same spot. The "impression" is a mention inside an AI answer; the conversion often arrives days later as a direct visit or a branded search. If your dashboard only rewards the last click, you'll conclude the channel doesn't work — right up until a competitor who measured it properly has quietly owned the recommendation surface in your category.

The 3 metrics that actually matter

You need a stack of three, from easiest-and-least-complete to hardest-and-most-honest:

1. AI referral sessions (the floor)

This is what GA4 already gives you: sessions where the referrer is an AI surface. It's the under 1% — real, worth trending, but a floor, not a ceiling. Segment it out so ChatGPT, Perplexity, Gemini, Copilot, and other AI referrers aren't buried inside "direct" or "referral." Asva's AI Traffic Decoder pulls this signal cleanly out of your analytics so you can trend it on its own — but never mistake this number for the size of the channel.

2. Self-reported discovery (the true signal)

This is the Facebook lesson made concrete. Ask people how they found you and let them tell you "ChatGPT." When AI options appear in a "How did you hear about us?" (HDYHU) field, they capture the view-through discovery that never generated a click — the exact demand last-click analytics throws away. It's usually the single highest-value number you can add, because it measures influence at the point the buyer actually felt it, not at the accident of which tab carried the referrer.

3. Assisted and branded-search correlation

The third metric triangulates the first two. Track whether your visibility inside AI answers moves in step with branded search volume and assisted conversions. When your brand starts getting recommended by ChatGPT for a category question, branded queries and direct visits typically rise on a lag. That correlation is your incrementality proxy: it links the surface you can influence (mentions and citations) to the demand you can bank. Monitoring the mention side is what the Brand Visibility Tracker is built for.

None of the three alone is complete. Together they bracket the truth: the floor of counted clicks, the honest signal of self-report, and the correlation that ties influence to revenue.

How to set it up

You can stand up a credible AEO measurement layer in an afternoon:

Add AI engines as explicit HDYHU options. In your signup flow, demo request, and post-purchase survey, add ChatGPT, Perplexity, Gemini, Claude, and Copilot as named choices in the "How did you hear about us?" question. Don't hide them under a generic "AI" or "Other" bucket — the specificity is the data, and it tells you which surface to invest in.

Isolate AI referral traffic. Build a segment or channel group that captures known AI referrers so their sessions stop hiding inside direct and referral. Trend it weekly.

Instrument branded demand alongside it. Pull branded search volume and direct sessions into the same view as your AI-referral and self-report numbers, so the lagged correlation is visible when it appears.

Track your mention share on the answer side. Counting sessions tells you what landed; tracking whether AI engines name and recommend you tells you what's coming — the leading indicator the other metrics confirm after the fact.

Read the three together. A flat referral number with rising self-reported discovery and climbing branded search is a channel that's working — you're just seeing it through the right instruments.

Want a fast read on whether AI engines mention you today? Run a free AI visibility check — it's built to be repeated, so you can watch the mention side move as you invest.

Frequently asked questions

Why does GA4 under-count AI discovery?

GA4 is a last-click tool: it only logs a session when someone clicks a link out of the AI answer. Because AI engines answer inline, most brand discovery happens without any click, so it never reaches GA4 — or lands later as direct or branded traffic and gets miscredited. Ravish Agrawal of Gamma noted click-through share stayed under 1% even as ChatGPT-referred traffic grew roughly 8x, which is why clicks alone drastically understate the channel.

What does "measure it like Facebook" actually mean?

It means adopting the attribution tools paid social was forced to invent — view-through thinking and self-reported attribution — instead of relying only on last click. Ravish Agrawal of Gamma's framing was that AEO "behaves like Google" in execution but "behaves like Facebook" in measurement: most of its impact is influence that shows up as a later conversion, not an immediate click.

Is the "How did you hear about us?" survey reliable enough to base decisions on?

Any single response is imperfect, but in aggregate self-report recovers exactly the discovery that last-click analytics drops, measuring influence at the moment the buyer felt it. Used alongside AI-referral sessions and branded-search correlation, it's the most honest single number most teams can add.

Should I stop tracking AI referral sessions since they're under 1%?

No — keep them as your floor. They're real, they trend, and they're the cleanest hard number you have. The mistake is treating that floor as the ceiling. Pair the counted clicks with self-reported discovery and branded-demand correlation to read the whole channel, not just its visible tip.

Which AI engines should I list in the survey?

Start with the surfaces buyers actually name: ChatGPT, Perplexity, Gemini, Claude, and Copilot. List them individually rather than as a single "AI" option — knowing which engine drives discovery tells you where to focus your AEO work.

How do I know if my AEO work is paying off before conversions show up?

Watch the answer side. Whether AI engines mention and recommend your brand is a leading indicator; branded search and self-reported discovery confirm it on a lag. Tracking mention share across engines lets you see momentum building before it lands in revenue.

The bottom line

AEO looks small in most dashboards not because it is small, but because it's measured with the wrong instrument. Ravish Agrawal of Gamma's line is the fix in a sentence: build like it's Google, measure like it's Facebook. Add AI engines to your "How did you hear about us?" field, isolate AI referral traffic, watch it against branded demand, and track whether the engines actually recommend you. Do that and the channel stops being invisible.

If you'd rather see the mention side first, start with a free AI visibility report, explore your full picture in the Asva app, or book a 30-minute walkthrough and we'll map your measurement stack together.

How to Measure AEO: It Behaves Like Google, But You Track It Like Facebook