You can track whether ChatGPT and other AI assistants recommend your Shopify store three ways: manually prompting the assistants on a fixed schedule, subscribing to a monitoring tool (Otterly.AI, Peec AI, Profound, Arbling), or checking your own analytics for AI-referred sessions. Do all three. Each one catches signals the other two miss, and the manual protocol below costs nothing but an hour a month.
The manual protocol
The manual protocol is a fixed list of questions, asked the same way every month, with answers logged in a spreadsheet. Write 10 prompts: two brand checks ("what is [store], is it legit"), five buyer questions phrased the way your customers talk ("best minimalist gold hoops under $200 that won't tarnish"), and three category questions ("who makes reliable third-party-tested magnesium supplements"). Ask each prompt in ChatGPT with web search on, in ChatGPT from memory (search off), in Perplexity, and in Gemini or Google AI Mode. For every answer record two things: were you named, and was your domain linked as a source. Repeat each prompt at least three times per platform; assistants sample, and single runs mislead. When we ran our own August 2026 measurement, the repeat sampling is what separated a real zero from bad luck — 290 calls gave us rates, not anecdotes. One hour a month buys you a real baseline.
Two details make or break the data. First, keep the prompt list frozen: the moment you reword a question, this month's numbers stop being comparable to last month's. Add new prompts as a separate list if you need to. Second, test the memory surface and the search surface as separate rows. ChatGPT answering from its training data and ChatGPT reading live search results are effectively two different recommenders — in our own audit, the search-enabled surface pulled citations on 47 of 50 calls while the memory surface pulled zero on all 50. Merchants routinely score well on one and not the other.
What to look for in your own analytics
Your analytics show the arrival side of AI recommendations: sessions referred from assistant domains and sessions carrying AI query parameters. In Shopify or GA4, segment by referrer for chatgpt.com, chat.openai.com, perplexity.ai, gemini.google.com and copilot.microsoft.com. Separately, watch landing URLs for utm_source=openai — when ChatGPT cites pages in shopping answers, appended tracking parameters show up in your logs, and we saw exactly that pattern in our own August 2026 citation data. Two caveats keep the numbers honest. AI referral share is still small in absolute terms: Similarweb put ChatGPT referrals at roughly 7.1% of referral traffic to retail sites in 2026, just under paid search at 7.8%. And most AI-influenced buying never shows a referrer at all — a shopper asks ChatGPT, reads the answer, then types your store name into a browser an hour later. That lands as direct traffic. Treat measurable AI referrals as the floor.
The cleanest fix for the invisible-referrer problem sits at signup, not in analytics: ask "how did you hear about us?" with an explicit AI-assistant option. Self-reported attribution is imprecise, but it is the only instrument that catches the ask-ChatGPT-then-type-the-URL path, which referrer data structurally cannot.
When a monitoring tool earns its fee
A monitoring tool automates the manual protocol: it asks your prompt list on a schedule, across platforms, and charts mention and citation rates over time. Otterly.AI tracks brand mentions and website citations across ChatGPT, Perplexity and Google AI Overviews. Peec AI adds position and sentiment analytics aimed at marketing teams. Profound covers enterprise scopes with content workflows attached. Arbling — us — runs measurement as part of a merchant loop: audit, fix the product data underneath, republish, remeasure on a frozen panel. The build-vs-buy line in 2026 is roughly this: one brand, ten prompts, no competitor tracking — the spreadsheet works fine. The tools earn their subscription when you need competitor benchmarks (whose mention rate is rising when yours is flat), defensible month-over-month deltas, or more surfaces than you will manually poll. What no tool changes: if the answer is "you are not recommended," the fix lives in your product data and your content, not in the dashboard.
Reading your first results
Expect a specific and slightly painful pattern on the first run: brand queries fine, buyer queries empty. When we measured ourselves in August 2026, brand questions scored a perfect mention rate on all eight surfaces we test, and every buyer-intent question — the "who should I buy from" phrasing that precedes actual purchases — returned zero mentions and zero citations. Across the 800+ stores we have analyzed, that split is the rule, not the exception. It happens because assistants answer brand questions from general knowledge about you, but answer buyer questions from retrieved sources — comparison pages, buying guides, review roundups — and most merchants have never published anything that functions as a source. So interpret results in pairs: mentioned and cited means you are working as a source; mentioned but never cited means assistants know you exist but read about you elsewhere; absent from buyer queries means the conversation about your category happens without you, and the competitors filling that space are your real rivals in AI search.
What to do about an empty buyer column is its own topic — our GEO playbook for e-commerce covers the fix side, and what an AI visibility audit checks covers getting the measurement done professionally. Or start with the version we run for clients at arbling.com.