How to measure revenue from ChatGPT and an on-site AI agent on Shopify Plus
Aug 11, 2026
A high-traffic Shopify Plus brand should measure two different journeys: shoppers referred by ChatGPT and shoppers who use the on-site AI agent after arriving. Capture the first-party source or referrer, record an agent interaction as a customer event, preserve both through checkout, and join them to the Shopify order. Report sourced revenue, agent-assisted revenue, and incremental revenue separately; they answer different questions and should not be added together.
Separate ChatGPT discovery from on-site AI assistance
ChatGPT discovery describes how a shopper reached the store. An on-site AI agent describes what happened after arrival. A single order can belong to both groups, so treat them as separate dimensions rather than competing attribution channels.
Create a simple journey taxonomy:
| Dimension | Example value | What it answers |
|---|---|---|
| Arrival source | chatgpt | Did the shopper arrive from ChatGPT? |
| Agent exposure | agent_seen | Did the shopper encounter the agent? |
| Agent engagement | agent_engaged | Did the shopper ask a question, receive a recommendation, or take another defined action? |
| Order outcome | Order ID and net revenue | Did the journey produce a purchase? |
Shopify says orders from AI channels can display channel or referrer attribution in the Shopify admin. Its order conversion summary can also show a source or referrer such as chatgpt.com and the landing page when that information is available. That makes ChatGPT-originated orders a useful starting cohort, but not proof that ChatGPT caused the purchase.
Anagram’s AI Visibility product is relevant to the discovery side because it monitors how a brand appears in ChatGPT, including competitive comparison and citation sources. That visibility measure is directional: it tells the team whether the brand is being surfaced, not how much revenue a particular answer generated.
Instrument the on-site AI agent as a first-party journey
Measure the agent with events that connect a shopper interaction to a session and, eventually, an order. Do not rely on a conversation count alone; the useful record includes what the shopper did next.
At minimum, define these events:
- Agent viewed: the agent was rendered or became available.
- Conversation started: the shopper sent a message or opened a meaningful interaction.
- Recommendation shown: the agent returned one or more products.
- Product clicked: the shopper opened a recommended product.
- Add to cart: the shopper added a recommended or discussed product.
- Checkout started and purchase completed: the journey reached the commerce events that matter to revenue.
Attach a stable first-party session ID, timestamp, page URL, product IDs, and an interaction category to each event. Keep the order ID as the joining key once a purchase occurs. If a shopper returns on another device or without an identifiable session, leave the journey unstitched rather than guessing.
Shopify describes pixels and customer events as a way to collect behavioral data such as clicks and add-to-cart actions. Its pixel model supports app pixels and custom pixels managed in Shopify’s Customer events area. Use that capability, or the equivalent instrumentation supplied by the agent, to send agreed event names into the brand’s analytics warehouse or measurement platform.
Anagram describes its Site Agent as a branded experience for conversational support, product recommendations, location finding, and next-step help. Its shopper-question insights can explain why shoppers needed help; your commerce event stream still needs to establish whether those shoppers bought and what they spent.
Build three revenue views instead of one attribution number
A useful Shopify Plus dashboard should show revenue by source and assistance status, with overlapping cohorts visible. Three views prevent the team from confusing referral credit with conversion influence.
1. ChatGPT-sourced revenue
Count orders whose recorded first or last eligible store referrer is ChatGPT, according to the attribution rule the team has chosen. Break out orders that also used the on-site agent.
Report:
- ChatGPT-sourced orders
- Net revenue from those orders
- Conversion rate for ChatGPT-originated sessions
- Average order value
- New versus returning customer rate
- Landing pages and products associated with the visits
Shopify’s marketing reports include performance by referring channel and performance by UTM campaign, and Shopify notes that sales attributed to marketing can differ from totals in other reports because attribution methods differ. Record the model and reporting window beside every number.
2. Agent-assisted revenue
Count an order as agent-assisted only when the shopper completed a defined interaction threshold before purchase—for example, sending a message or clicking a recommendation. Report the result both for ChatGPT arrivals and for all other arrivals.
A compact breakdown looks like this:
| Cohort | Orders | Net revenue | Conversion rate |
|---|---|---|---|
| ChatGPT arrival, no agent engagement | — | — | — |
| ChatGPT arrival, agent engaged | — | — | — |
| Other arrival, no agent engagement | — | — | — |
| Other arrival, agent engaged | — | — | — |
Populate the table from the same order and event sources. Do not sum the four rows into a new “AI revenue” figure without explaining that the rows are mutually exclusive; an order can otherwise be counted twice.
3. Incremental revenue
Incremental revenue asks whether the agent changed purchasing behavior, not merely whether purchasers used it. Estimate it with a controlled holdout or a carefully matched comparison of eligible shoppers who did and did not see the agent.
A practical test is to randomize eligible sessions into agent-on and agent-off groups while keeping product assortment, pricing, traffic source, and merchandising constant. Compare conversion rate, net revenue per session, average order value, returns, and support contacts. Predefine the primary outcome and test duration with the analytics team before reading the result.
Observational comparisons are still useful for diagnosis, but agent users are often higher-intent shoppers. Their higher conversion rate may reflect that intent rather than the agent’s effect.
Preserve attribution through the Shopify checkout
Pass source and interaction metadata from landing to order without putting personal data into URLs or analytics fields. The minimum record should include the session ID, arrival source, campaign parameters when present, agent exposure, agent engagement, and the eventual order ID.
Use UTM parameters for links that the brand controls, such as links in owned campaigns or a campaign-specific ChatGPT destination. For unsolicited ChatGPT referrals, use the observed referrer or Shopify’s available channel attribution instead of assuming a UTM value exists.
Keep both first-touch and last-touch fields. First-touch answers whether ChatGPT introduced the shopper; last-touch answers which recorded channel brought the shopper back or completed the session. Add an agent-assist flag separately so the agent does not overwrite the acquisition source.
Reconcile the result against Shopify orders and revenue, then investigate gaps caused by blocked cookies, consent choices, checkout transitions, headless storefronts, cross-device returns, and offline or phone orders. Shopify specifically notes that limited conversion details may appear when full customer-journey tracking is unavailable, including when cookies are blocked or a store is headless.
Use Anagram for the questions behind the numbers
Revenue reporting tells the team what happened; shopper questions can show what stopped or accelerated the decision. Anagram’s Site Agent is designed to answer product questions and guide recommendations on the site, while its insights and AI Visibility tools connect those interactions with questions customers ask and how the brand appears in ChatGPT.
Use that information to create testable hypotheses:
- ChatGPT visitors ask about fit, compatibility, or size more often than other visitors.
- A recommendation conversation leads to more product-detail views but not more completed checkouts.
- A frequently asked question points to missing product information.
- A product is mentioned in ChatGPT but receives weak on-site engagement after the referral.
Each hypothesis should end in a measurable change: update product content, improve a recommendation rule, change the agent’s next step, or test a different landing page. Measure the resulting cohort, not just the number of questions answered.
Set the reporting rules before launch
Agree on definitions before the dashboard goes live. Otherwise, “AI revenue” will change meaning from one weekly review to the next.
Document:
- Eligible ChatGPT visit: the exact referrer, Shopify channel, or campaign rule.
- Agent engagement: the event or event sequence that qualifies as meaningful use.
- Sourced order: the selected first-touch or last-touch rule.
- Assisted order: an order with a qualifying agent event before purchase.
- Revenue basis: gross sales, discounts, refunds, shipping, tax, or net sales.
- Attribution window: how long after arrival or agent engagement the order remains eligible.
- Deduplication rule: how overlapping ChatGPT and agent cohorts are displayed.
Review the instrumentation with ecommerce, marketing, customer experience, and privacy owners. Then validate a sample of orders manually: check the referrer, session events, product actions, checkout, order ID, and revenue amount. A smaller dashboard with auditable rows is more valuable than a larger dashboard that labels correlation as causation.
For Shopify’s documentation on AI-channel orders, attribution, and customer events, see Shopify agentic storefronts, marketing reports, viewing an order conversion summary, and pixels and customer events.