Agentic commerce is shopping in which an AI agent, such as Google’s AI Mode, ChatGPT or Perplexity, finds products, compares them and sometimes completes the purchase for the user. Google’s Universal Commerce Protocol (UCP) and OpenAI’s Agentic Commerce Protocol are attempts to standardize how agents talk to merchants. Measuring it means tracking agent-referred sessions and orders separately, checking feed and catalogue quality, and comparing margin and returns with other channels.
What is agentic commerce?
Traditional e-commerce assumes a person clicks through pages. In agentic commerce, part of that journey happens inside an AI system. A user asks for “a waterproof hiking jacket under 150 euros that ships to Istanbul by Friday,” and the agent searches catalogues, filters options and presents a short list. In some cases it can check out with stored payment details after the user confirms.
Several efforts are underway at the time of writing. OpenAI introduced Instant Checkout in ChatGPT with the Agentic Commerce Protocol, built with Stripe. Google announced UCP as an open standard for agents to discover products, manage carts and complete purchases with merchants, alongside agentic checkout features in AI Mode and Gemini. Perplexity and others offer their own shopping flows. Details and availability differ by country and platform.
Why is it hard to measure?
Measurement is difficult because the agent sits between you and the customer. Some agent journeys end in a click to your site, which analytics can see. Others end in a checkout handled through a protocol, so the order reaches your platform without a normal session. And many agent interactions influence a purchase that happens days later through another channel.
There is also no shared reporting standard yet. Each platform reports differently, and some report very little. A measurement plan has to combine what you can see directly with careful estimation.
Which KPIs actually matter?
Start with a short set of KPIs grouped by stage. Fewer, well-defined metrics beat a dashboard full of numbers nobody trusts.
Visibility
- Agent referral sessions: visits from known AI referrers such as chatgpt.com, perplexity.ai and gemini.google.com, tracked as a separate channel.
- Product inclusion rate: for a fixed set of test prompts, the share of answers in which your products appear.
- Feed coverage: share of active products with complete titles, attributes, prices, stock and shipping data in Merchant Center and other feeds.
Conversion
- Agent-originated orders: orders placed through agentic checkout or tagged as coming from an agent platform.
- Conversion rate of agent referral sessions compared with organic and paid search.
- Average order value for agent orders versus site orders.
Economics
- Contribution margin per agent order after fees.
- Return rate for agent orders, since products chosen by an agent may match expectations less well.
- Customer acquisition share: how many agent orders come from new customers.
How do you set up tracking?
Good measurement starts with clean data in the systems you already own.
- Create a custom channel group in Google Analytics 4 for AI assistants, using a regular expression on session source for known AI referrer domains, and update it as new platforms appear.
- Make sure order data from any agentic checkout includes a source field your e-commerce platform stores, such as an order tag or sales channel in Shopify.
- Keep Merchant Center feeds complete and current; agents rely heavily on structured product data.
- Use UTM parameters where platforms allow you to control the link.
- Join order data with CRM or customer data to see new versus returning buyers.
For the test prompt metric, run the same 20 to 50 shopping prompts each month across the main agents and record whether your products appear. Answers vary, so run each prompt more than once.
What should the dashboard look like?
A useful agentic commerce dashboard fits on one screen. Put visibility metrics at the top, conversion in the middle and economics at the bottom. Show monthly trends rather than daily figures, since volumes are often small.
Looker Studio works well for combining GA4, Merchant Center and order data. Add a small table showing the test prompt results per platform. Include a notes column for changes such as feed fixes or new platform launches, so trends can be explained.
How do you prove value to leadership?
Leadership will ask whether agent orders are incremental or simply moved from other channels. Two approaches help. Compare new customer share for agent orders against your site average. And run a holdout where possible, for example pausing an agent integration in one market or product category for a few weeks, to see whether overall sales fall.
Be honest about uncertainty. Report what is measured, what is estimated and what is unknown. A dashboard that admits gaps earns more trust than one that claims every sale is tracked.
Pitfalls to avoid
Three mistakes distort agentic commerce reports. Counting every visit from an AI domain as agent-driven inflates the channel, since some users simply copy a link. Ignoring dark traffic, where users paste a product name into search after an AI conversation, undercounts it. And comparing tiny monthly volumes as percentages produces dramatic swings that mean nothing. Report absolute numbers next to rates until volumes grow.
How Alien Road applies agentic commerce measurement
In a typical engagement our e-commerce team starts with a data audit: feed completeness in Merchant Center, product attribute quality, analytics channel setup and how the store platform records orders from external sources. We list which agent platforms are relevant in the client’s markets.
Setup covers fixing feed gaps, building the AI assistant channel group in GA4, configuring order tagging for any agentic checkout and creating the monthly test prompt set. We then build the one-screen dashboard in Looker Studio.
Measurement runs monthly, with the prompt test and dashboard refreshed together. Reporting explains movements in each KPI, notes platform changes and recommends where to improve product data next.
Where to start this week
Open Google Analytics 4 and check whether traffic from chatgpt.com, perplexity.ai and gemini.google.com is visible in your referral reports. If it is mixed into “referral,” create a custom channel group for AI assistants. Then pick ten shopping prompts your customers might use, run them in two AI assistants and write down whether your products appear. That gives you a baseline before any larger investment.