AEO has a measurement problem. Traditional attribution assumes a click happens before a conversion. But most AI-influenced revenue never produces a click. A buyer asks ChatGPT for a recommendation, hears your brand name, and types your URL directly into their browser. Or they Google your brand name a week later. In both cases, your analytics shows "Direct" or "Branded Organic." The AI influence is invisible.

This guide walks through a practical attribution framework that combines two imperfect signals (UTM tracking and HDYHAU surveys) into an estimate you can actually defend to a CFO. It won't be perfect. No attribution model is. But it will get you from "we have no idea" to "we can put a number on this."

Why Traditional Attribution Breaks for AEO

Traditional digital marketing attribution tracks clicks. A user clicks an ad, lands on your site, and converts. The click creates a clean line from spend to revenue.

AEO breaks this model in three ways:

• Zero-click influence. When ChatGPT recommends your product, the user gets the answer without clicking a link. They may never visit the AI platform's website. Your brand just entered their consideration set, and you have no record of it.
• Dark social referrals. Even when users do click links in AI responses, mobile apps (ChatGPT iOS, Claude app) often strip referrer headers. The traffic shows up as "Direct" in GA4. You can't tell it came from an AI engine.
• Delayed brand lift. A buyer asks Perplexity about your category today but doesn't visit your site for two weeks. When they do, the visit looks organic. The AI nudge happened outside your tracking window.

The result: your AEO spend is producing revenue, but your analytics can't see it. This is why most companies either don't invest in AEO or can't justify continuing.

The Two Signals You Actually Have

You can't track everything. But you have two reliable signals that, combined, give you a reasonable estimate.

Signal 1: UTM-Tagged AI Traffic

Some AI-driven traffic does carry referrer data. When a user clicks a link in the ChatGPT web interface on desktop, the referral header is preserved. GA4 can capture this as a session source if you've set up an "AI Search" custom channel group.

This gives you a hard floor. You can see exactly how many sessions came from AI engines with intact referrers, and you can track conversions and revenue from those sessions. This number is real and auditable.

The problem: it's a significant undercount. Estimates vary, but UTM-tracked AI traffic typically represents only 20-40% of actual AI-influenced visits, because mobile app traffic, in-app browsers, and delayed visits all get misattributed.

Signal 2: HDYHAU Survey Responses

"How Did You Hear About Us?" surveys are the oldest attribution tool in marketing, and they're surprisingly effective for AEO.

Add a post-signup or post-purchase question: "How did you first discover us?" Include options like "ChatGPT / AI assistant," "Perplexity," "Google AI Overview," and "AI search (other)." You can also include non-AI options like "Reddit," "Google search," "Referral from colleague," etc.

HDYHAU captures what analytics cannot: the user's actual memory of how they found you. It catches zero-click recommendations, mobile app referrals, and delayed visits.

The Multiplier Model

The core idea: UTM-tracked AI revenue is your known floor. The ratio between HDYHAU "AI" responses and UTM-tracked AI sessions tells you how much AI-influenced activity your tracking misses. Multiply to estimate total AI-attributed revenue.

The Formula

Let's break that down:

• chatgpt_utm_revenue: Revenue from sessions where GA4 captured an AI referral source. This is your hard number.
• hdyhau_ai_pct: The percentage of HDYHAU survey respondents who said they discovered you through an AI tool.
• total_revenue: Total revenue for the period.
• hdyhau_ai_responses: Raw count of people who said "AI" in the survey.
• chatgpt_utm_sessions: Sessions GA4 tracked as coming from AI engines.
• multiplier: The ratio telling you how many real AI-influenced users exist for each one your UTMs catch.

Worked Example

Say you're a B2B SaaS company. In Q1:

• Total revenue: $500,000
• GA4 tracked 120 sessions from AI sources, generating $8,400 in revenue
• 400 customers filled out the HDYHAU survey (5% sample rate)
• 32 of those said "ChatGPT / AI assistant" (8% of respondents)

Wait. The multiplier is below 1. This happens when your UTM tracking catches more sessions than your survey catches responses. When the multiplier falls below 1, you floor it at 1 (meaning "at minimum, the tracked revenue is the estimate"). In practice, this means your HDYHAU survey isn't collecting enough responses, or your UTM tracking is catching a higher share than expected.

Let's adjust the example to a more typical scenario where the survey catches more signal:

• GA4 tracked 45 AI sessions, generating $3,200 in revenue
• 400 HDYHAU respondents, 32 said "AI"

In this case, HDYHAU is doing the heavy lifting. The model says roughly $43,200 (8.6% of total revenue) was influenced by AI engines.

Adding Confidence Intervals

No estimate is exact. Apply a confidence range (typically +/- 25%) to communicate uncertainty:

For the example above: $32,400 to $54,000. Presenting a range instead of a single number signals intellectual honesty and gives stakeholders appropriate calibration.

The Correlation Model (12+ Weeks of Data)

Once you have three or more months of data, you can run a second model that validates (or challenges) the Multiplier Model.

The Correlation Model uses ordinary least squares (OLS) regression to find the statistical relationship between your Share of Voice in AI engines and your revenue, controlling for other factors like paid media spend and seasonality.

This tells you: "For every 1% increase in Share of Voice, revenue increased by $X, after accounting for changes in paid spend and seasonal patterns."

The Correlation Model requires an AI visibility tracking tool that monitors your Share of Voice over time. You'll need at least 12 weekly data points, ideally more.

Calculating Cost Per Citation

Once you have estimated AEO revenue, you can calculate the metric that makes AEO spend justifiable: cost per citation.

If you're spending $5,000/month on AEO and generating 80 citations, your cost per citation is $62.50. If those citations are driving an estimated $43,200 in quarterly revenue (roughly $14,400/month), your revenue per citation is $180. That's a 2.9x return.

Track cost per citation monthly alongside Share of Voice. As your AEO program matures and your citation count grows, cost per citation should decrease while revenue per citation holds steady or increases. If cost per citation rises without a corresponding lift in SOV or revenue, something in the program needs attention.

Setting Up the Tracking Stack

To run this framework, you need four things:

1. GA4 with an "AI Search" channel. Follow the GA4 AI tracking guide to set up a custom channel group that captures AI referral traffic. This gives you chatgpt_utm_sessions and chatgpt_utm_revenue.
2. A HDYHAU survey. Add to your signup flow, checkout, or post-purchase experience. Keep it short (one dropdown question). Include "ChatGPT / AI assistant," "Perplexity," "Google AI Overview," and "Other AI." Track your survey response rate.
3. An AI visibility tool. You need Share of Voice data for the Correlation Model. See the AI visibility tools comparison for options. At minimum, track SOV weekly for your top 20-50 queries.
4. A spreadsheet or dashboard. Pull these data points together into a single view: UTM revenue, HDYHAU percentages, SOV, paid spend, and total revenue per period. A Google Sheet works fine to start.

Frequently Asked Questions