An ecommerce team should track five AI shopping KPIs: recommendation share, SKU coverage, merchant-routing share, AI-shopping sessions and AI-attributed revenue. Brand mentions, citations and sentiment still matter, but as diagnostics that explain why one of the five moved.

In July 2026, visits to US retail sites from AI sources grew 62% year over year, and those shoppers converted 60% better than shoppers from other channels, according to Adobe's AI Traffic Trends Report. That growth is only useful if you can see which products and sellers capture it, and each KPI below answers one question.

Key takeaways

• Five KPIs cover the path from AI answer to order. Report each on its own line.

• Brand presence can hide the losses that matter. In our sample, one brand had six of its nine featured products named, yet 29 of its 33 buy offers pointed to other sellers.

• SKU coverage and merchant routing catch products that never appear and buying journeys that end on another seller's site.

What are the five AI shopping KPIs?

The first three come from AI answers to a fixed question set. The last two come from your analytics.

Calculated: computed from AI answers you collect. Attributed: credited by your analytics to an AI source.

Why isn't brand visibility enough for an ecommerce team?

A brand can appear in most answers while its products lose the shortlist and its store loses the sale. OpenAI separates the two decisions. ChatGPT picks products by relevance to the shopper's intent, then ranks the sellers of each product on factors such as availability, price and quality (OpenAI Help Center); our page on what seller choice does to margin covers that second decision.

How we measured this. We scored five brands in beauty, jewelry and home goods between October 2 and 8, 2026, on the consumer ChatGPT and Google AI Mode interfaces. Each brand had 20 shopper questions, each run once per engine, for 200 answers. Featured products were each brand's best sellers, three to ten per brand, plus one entry-price model we chose for the home goods brand. The sample is small and came from sales conversations rather than at random. Answers vary between runs, so treat per-brand figures as directional. The sample shows why brand presence alone misleads; it does not measure the five KPIs themselves.

• When the question named the brand, 42 of 42 answers named it (100%).

• When the question did not, 75 of 158 answers named it (47%).

• All answers combined, a blended visibility score: 117 of 200 (59%).

• Featured products named in any answer, a looser cousin of SKU coverage: 19 of 42 (45%).

• Merchant offers pointing to the brand's own store, counted per offer shown across branded and unbranded answers: 41 of 135 (30%). Offers are not recommendations, so treat this as a proxy for routing share.

The blended 59% looks healthy, but it mixes questions a brand almost never loses with those that win new customers. On unbranded questions, results ran from 27% to 72% by brand. The home goods brand shows the gap between presence and where the sale goes: six of its nine featured products were named, yet 29 of its 33 merchant offers pointed to other sellers.

In our glossary's terms, a product loss means the AI recommended a rival's product. A merchant loss means it recommended yours but sent the shopper to another seller.

What does each AI shopping KPI look like in practice?

The examples use a fictional US lab-grown diamond jewelry brand with 120 priority products priced $300 to $1,200. Each month it runs 150 shopper questions three times on each of two engines: 900 answers.

Recommendation share

Recommendation share is how often your products make the AI's shortlist compared with competitors. Count only products offered to buy. Across 900 answers, the engines recommend 4,500 products and the brand holds 630, a 14% share. It is a consideration number; sales need their own KPI.

SKU coverage

SKU coverage is the share of priority products that appear for the shopper questions they should win. A $450 pair of studs fits "lab-grown diamond studs under $500 for a first anniversary." A $1,100 pair does not.

Here 31 of the 120 priority products appear at least once, so coverage is 26%. The useful output is the list of 89 products never considered. Google draws the same line between winning and breadth. Its Merchant Center AI performance report shows share of voice and, separately, "products showing", which "indicates how many of your products are appearing for top terms, popular attributes, and search intents" (Google Merchant Center Help).

Merchant-routing share

Merchant-routing share is the portion of your recommendations where the AI lists your own store as a place to buy. Merchant routing is which seller the AI sends the shopper to after recommending a product, and it decides margin and who owns the customer.

Sellers appear on 540 of the brand's 630 recommendations, and its own store on 189: a 35% routing share. Count only answers that show sellers. A wholesale-led brand may run a low share by design.

AI-shopping sessions

AI-shopping sessions are visits your analytics can trace to an AI assistant. Google Analytics 4 has a default AI Assistant channel for traffic from sources "like ChatGPT, Gemini, Deepseek, Copilot, or Grok" (Google Analytics Help).

Read it as a floor: GA4 counts a session as direct traffic "when no information about the referral source is available" (Google Analytics Help). Setup steps are in our guide to tracking AI-shopping traffic.

AI-attributed revenue

AI-attributed revenue is the revenue from orders placed in sessions with an identifiable AI source. A CFO can audit every order behind it, though attribution proves a path and some buyers would have bought anyway.

A shopper who asks ChatGPT on Monday and buys through branded search on Saturday shows up as organic. Estimate that influence separately and label it AI-assisted revenue (modeled).

How should you read the five KPIs together?

Read them in order. The earliest KPI to fall shows where the leak starts.

Illustrative figures for the fictional jewelry brand.

Recommendation share rose five points while coverage barely moved, so a few products are winning more often. Routing fell eleven points: marketplaces took much of the gain. Sessions and revenue grew, so a blended score would call it a good month.

The implication is different for each brand. A direct-to-consumer brand watches own-store routing closely, including how often its store is listed first. A retailer-led brand cares more about whether its product wins, whichever authorized seller gets the order. A multi-market brand can see one SKU behave differently by country as prices, stock and rivals change.

The trend only means something if the test stays the same, and answers vary between runs, so repeat each question several times per engine. Our guide to what shoppers ask AI covers the question set.

What should you check when a KPI moves?

Brand presence, competitor interception, citation share and sentiment live here, as diagnostics.

What can AI shopping KPIs not tell you?

These KPIs record what engines displayed and what your tracking caught. They cannot prove why an engine chose a product, or count every sale it shaped. Results hold only for the questions tested. A KPI that moves after a fix shows correlation; a controlled test, such as a holdout group of products, shows cause. Purchases that route to a marketplace never reach your analytics, and no vendor, Gravton included, can guarantee a recommendation.

Where does Gravton fit?

Gravton's free audit delivers a Demand Snapshot for your category: where your products appear, where they lose to a rival product or another seller, and ranked fixes for your site, content and product data. Measurement then tracks how often your brand and products are recommended and how your share moves after each change.

Get your free AI visibility audit and see which of your products AI recommends, and where the sale goes.

Frequently asked questions

Why aren't share of voice and citation share AI shopping KPIs?

Both describe the brand conversation. Share of voice shows how often you come up, and citation share shows which sources engines rely on. An ecommerce team needs to know whether its products make the shortlist and whether its store gets the buyer. Use both as diagnostics when recommendation share or SKU coverage drops.

What is a good SKU coverage rate?

We found no published benchmark. Coverage depends on catalog size and how strictly you define eligibility. In Gravton's five-brand sample, using the looser measure of featured products named in any answer, the one brand with three featured products had all three named, while brands with nine or ten ran from 20% to 67%. Set your own baseline, then work the list of absent products.

How often should leadership review these KPIs?

Monthly works for most leadership teams, since a single week of answers can mislead. Operating teams should look more often when prices, stock, assortment or campaigns change, or when an engine changes how it shows products. Keep each review on the same question set.

Can you measure AI-assisted revenue exactly?

No. If AI shaped the choice but the order arrived through search or a direct visit, you can only estimate its role. Our page on measuring AI-shopping performance covers the methods. Name the method, report the estimate in its own line and never add it to AI-attributed revenue.

Should the five KPIs be combined into one score?

A composite can show a trend but should never replace the five lines. The example month would score as a gain while the brand lost nearly a third of its own-store routing. A score nobody can trace to a product or seller cannot drive a decision.

What should you read next?

• What Do Shoppers Ask AI Before They Buy? covers the question set every KPI is measured on.

• Recommendation Share: How Much of the AI Shortlist Is Your Brand Winning? goes deeper on the first KPI.

• How Do I Track AI-Shopping Traffic in GA4 and Shopify? sets up the session and revenue KPIs.

• Why AI Discovery Is a P&L Problem turns the KPIs into margin.

Sources

1. Adobe Digital Insights. "AI Traffic Trends Report." August 19, 2026. July 2026 data.

2. OpenAI Help Center. "Shopping with ChatGPT search." Accessed October 23, 2026.

3. Google Merchant Center Help. "About AI performance insights." Accessed October 23, 2026.

4. Google Analytics Help. "Default channel group." Accessed October 22, 2026.

5. Google Analytics Help. "Campaigns and traffic sources." Accessed October 23, 2026.

6. Gravton Labs. Scorecard data for five ecommerce brands, 200 answers from ChatGPT and Google AI Mode, October 2 to 8, 2026. Internal dataset, anonymized.