How Should Retailers Measure AI Discovery?

Measurement is where most retailers begin their AI Discovery work. It is also where most retailers get stuck. This piece describes what to measure, what to be careful about, and how to separate diagnostic metrics from business outcomes.

Why is measuring AI Discovery hard?

AI Discovery is harder to measure than earlier commerce disciplines for four reasons.

Signals live outside the retailer's environment

SEO signals live in Google Search Console. Paid signals live in ad platforms. Many of the signals that matter in AI Discovery are fragmented across AI systems and are not fully exposed in a retailer's existing analytics. Some AI platforms are beginning to expose limited performance data, and third-party tools now surface additional visibility signals, but a unified view remains rare.

The customer journey is fragmented

A shopper may discover a brand in ChatGPT, compare alternatives in Perplexity, search the brand on Google, and eventually visit the retailer's website directly. Each step happens somewhere different, and most of them are not visible to the retailer.

Attribution models were built for a different world

Last-click and last-touch attribution assume the customer's final channel is the one that shaped the decision. AI Discovery increasingly breaks that assumption.

AI referral traffic can remain low even as AI influence grows

Similarweb's research on the AI consumer journey has reported AI referrals remaining below 1% of retail traffic while its consumer research shows AI playing a much larger role in discovery and evaluation.

What most retailers measure today

Most retailers measuring AI Discovery today measure visibility. Some common metrics:

• Citation rate. How often the retailer's brand or products appear in AI-generated answers.
• Prompt coverage. The set of consumer prompts where the brand shows up.
• Share of voice in AI answers. How the retailer's presence compares to competitors within AI systems.
• AI referral traffic. Visits from AI systems recorded in web analytics.
• Content readability for AI. How well the retailer's content can be parsed and cited by AI systems.

Each of these has value. They tell the retailer whether AI systems see them, understand them, and cite them.

They do not, on their own, tell the retailer whether customers are choosing them.

What's missing from visibility metrics

The gap between visibility and outcome is where the measurement problem lives.

A retailer can improve its citation rate significantly. That does not automatically mean customers are choosing the retailer more often.

A retailer can grow AI referral traffic. That growth may only reflect a fraction of AI's actual influence on the business.

A retailer can win the AI visibility dashboard against competitors. And still lose the customer.

Visibility is a diagnostic. Revenue is the outcome. Both matter. Do not confuse them.

What retailers should measure

There are two categories of measurement worth being explicit about.

Diagnostic measurement — how AI systems see the retailer

• Are we cited in the AI answers our customers are asking?
• Is our brand and product information represented accurately in those answers?
• Do we appear in the shortlist of options AI recommends?
• How does our presence compare to competitors?

These are useful. They point to where the retailer needs to improve.

Outcome measurement — whether AI Discovery is contributing to the business

• Is our new customer acquisition changing in ways consistent with AI influence?
• Are AI-referred visits behaving differently than other traffic (conversion, revenue per visit, engagement)?
• Are customers arriving with different intent signals than they did before?
• Are downstream channels (direct traffic, brand search) growing in ways that suggest upstream AI influence?

These are harder to measure and easier to argue about. They are also closer to the business.

The distinction between diagnostic and outcome

The most useful mental model is medical. A body temperature reading is a diagnostic. Whether the patient is healthy is the outcome. Both matter, but confusing them is dangerous.

An AI citation rate is a diagnostic. Whether AI Discovery is producing business results is the outcome. Both matter. But treating the diagnostic as the outcome causes retailers to invest in the wrong things.

The easiest way to fool yourself in AI Discovery is to optimize the dashboard instead of the customer.

Attribution: what to expect

Traditional last-click attribution can materially underweight AI's influence. This is not something last-click attribution alone can solve. Sophisticated incrementality tests, multi-touch modeling, and panel or survey-based measurement can recover more of the signal, but each has its own limits.

In a 2026 Similarweb behavioral study across Finance, Travel, and Beauty verticals, users who received a ChatGPT brand recommendation were 2.5 times more likely to visit that brand's website within seven days, compared with competitor brands. Most of that lift does not show up as AI referral traffic in the retailer's analytics.

The implication: retailers should expect a gap between what their analytics attribute to AI and what AI is actually contributing. Closing that gap will require new measurement approaches. It will not be closed by refining last-click models.

Last-click attribution is becoming last-century attribution.

What should retailers be asking?

• Are we measuring what is easiest to see, or what actually influences the business?
• If our AI citation rate doubled next quarter, would we expect the business to change?
• What business outcomes should we expect from AI Discovery over the next twelve months?
• Which of our existing analytics investments will need to be reworked to see AI influence?

These are commerce and analytics questions. Not marketing questions.