A customer can discover a product in an AI answer and later reach your store through a different route. They might type your domain, search your brand name, or switch devices before purchasing. The customer's account of discovery and the visit recorded by your analytics can therefore describe different parts of the same journey.
The practical response is to measure three things separately: observed visits and orders, customer-reported discovery, and your brand's presence in a defined set of AI answers. These signals can inform each other, but they should not be added together as though they measured the same event.
Understand the question your report answers
Shopify's marketing documentation describes attribution through recorded referrers and interactions. It also explains that attribution models allocate credit differently: last click includes direct visits, while last non-direct click excludes them when an eligible non-direct source exists. Check the selected model before interpreting an order or comparing reports. [Shopify marketing reports](https://help.shopify.com/en/manual/reports-and-analytics/shopify-reports/report-types/default-reports/marketing-reports)
Now consider a hypothetical shopper who reads a recommendation but never clicks its link. The next day, they enter your store address and buy. That earlier reading experience has not created a store visit for your analytics to record. Changing an attribution model cannot manufacture a missing interaction.
This does not mean all AI-related visits become direct traffic. Inspect the data available in your own store instead of assuming a universal rule about how every assistant, browser, or shopping experience passes referrals.
Build a report around observable evidence
Start with the referral and acquisition data you can actually inspect. Identify sources associated with AI services when they appear, then review landing pages, sessions, orders, and sales under a consistent attribution model and date range.
Document how the segment is defined. If it uses a list of referring domains, keep that list with the report and revisit it periodically. If your account offers a relevant channel report, record its definition and coverage before comparing it with another analytics system. Different report definitions can produce different totals without either one being broken.
When the sample is small, show counts beside rates. Two orders from a handful of sessions can create an impressive conversion rate without establishing a stable pattern. Compare similar periods, note major promotions, and avoid describing an emerging channel as predictably superior based on a few purchases.
Ask customers a question analytics cannot answer
A short post-purchase question can capture remembered discovery: “Where did you first hear about us?” Include an AI assistant option alongside other realistic channels, an “other” field, and an option for people who do not remember.
If customers choose an AI assistant, an optional follow-up can ask what they were trying to find. A response such as “a bag that fits under an airline seat” is more actionable for content planning than a channel name alone.
Treat survey answers as self-reported information. People may remember the last influential experience rather than the first, and respondents may differ from customers who skip the question. Report the number of responses and response rate so the context remains visible.
Avoid leading questions such as “Did ChatGPT help you discover our excellent products?” Neutral wording makes the results more useful.
Keep visibility monitoring in its own column
A visibility check answers whether your brand appeared in the prompts you tested. It cannot, by itself, connect that appearance to a particular customer's order.
Track these three questions separately:
1. Which visits and sales can your store identify?
Check your analytics for recorded referral sources, the pages visitors landed on, and the orders attributed to those visits. This tells you which identifiable sources brought people to your store and whether those visits led to purchases.
For example, a recorded referral from an AI service is evidence of a visit from that source. It does not tell you about shoppers who read an AI recommendation and later typed your store address themselves.
2. Where do customers say they first found you?
Use a post-purchase survey to ask how customers heard about your brand. Record their answers and how many customers responded. This can reveal discovery experiences that your store analytics did not capture.
For example, a customer might select “AI assistant” even though their order shows direct traffic. Keep both observations: one describes what the customer remembers; the other describes the recorded visit.
3. Does your brand appear in the AI answers you test?
Run a consistent set of relevant shopping questions. Record whether your brand appears, whether the product information is accurate, and which sources the answer links to. Use those observations to identify questions your website needs to answer more clearly.
This measures your selected tests. It does not show how many real shoppers saw your brand or how many orders those answers generated.
If the same order appears in an acquisition report and a survey, it is still one order. Keep overlapping evidence attached to the order where your process permits, rather than counting it as two channel wins.
Use disagreement to guide the next investigation
Suppose several customers report AI discovery but identifiable referrals remain low. Review their landing pages and the buying questions they remember. Then check whether your site provides a useful answer to those questions.
If visibility improves but sales do not, examine the fit between the tested prompts and your customers. The test may be covering broad informational questions, the product may be unavailable in the relevant market, or the measurement period may be too short. These are possible explanations to investigate, not conclusions the dashboard can establish alone.
Metis AI by MICROFACTOR's [GEO platform] describes daily AI probes, visibility trends, and competitor monitoring. Those capabilities can support the visibility portion of this workflow. They should be evaluated alongside your store analytics and customer research, with each source retaining its own meaning.
The goal is a better decision: which product page to improve, which buying question to answer, and which observation to test again. A clear account of what you know is more useful than assigning an exact revenue figure to discovery you did not directly observe.