7 Shopify prompt classes ChatGPT actually cites
Every sentence the assistant writes back is either a citation Shopify stores can win or a citation they've already lost.
Surfient
Between November 2025 and March 2026 we logged 3,912 commerce prompts to ChatGPT, Perplexity, Claude, and Google AI Overviews — sampled from merchant analytics, observed shopper traces, and a panel of 240 US and UK consumers who agreed to let us inspect their assistant history for two weeks. Seven intent classes cover 96% of them. The remaining 4% are typos, abandoned half-prompts, and one-off edge cases (someone asked Claude for a poem about their cat's dietary restrictions). For practical purposes, there are seven.
If you run a Shopify store, this is the taxonomy your product copy has to answer. Not one of the seven. All seven. Because assistants don't pick a favourite prompt type — shoppers do — and the ones you haven't written for are the ones where a competitor steals the citation. For the underlying retrieval mechanics we covered in an earlier post, see how ChatGPT actually cites commerce content; this piece is about the intent side of that same pipeline.
What we sampled and why it matters
The panel skewed toward considered purchases ($60+ AOV, non-impulse): furniture, small appliances, apparel with a fit concern, skincare, pet nutrition, replacement parts, outdoor gear. We excluded impulse categories (trending beauty, fast fashion, branded snacks) because those queries mostly still go to TikTok and Instagram search — the assistant funnel hasn't absorbed them yet, and OpenAI's own shopping rollout notes describe the same mid-to-high-consideration skew.
We also excluded research-only prompts (“how does a DC motor work”) and pure-navigation prompts (“Wayfair returns policy”). The 3,912 remaining prompts are what the industry calls buyable — the shopper is, in that moment, willing to be told which specific product to purchase. Every sentence the assistant writes back is either a citation Shopify stores can win or a citation they've already lost. If you want a refresher on how we measure citations at all, start with citation rate and share of AI voice in the glossary.
Why a taxonomy, not a keyword list
The seven intent classes, at a glance
Before we unpack each one, here is the distribution across the sample. The chart is the same data the rest of the post references — a screen reader will read the labels and values verbatim, and so will a retriever that scans the HTML.
Each class is defined by the ranking cue the assistant uses to pick a product — the specific sentence shape it prefers to quote. Miss the ranking cue and your PDP is indistinguishable from every other PDP in your category, and the model defaults to whichever brand it has the longest training-data history with (usually not yours).
1. Product discovery — “best X” (28.4% of the sample)
The starter prompt. “What's the best standing desk?”, “best indoor rower 2026”, “best cast iron skillet”. The shopper has a category in mind and wants a shortlist of three to five brands with a sentence each. This is the only class where assistants will freely name brands without a constraint — every other class narrows the answer set. Our engine-specific playbooks go deep here: rank Shopify in ChatGPT and rank Shopify in Perplexity.
VISIT US:- https://www.surfient.com/blog/ai-shopping-prompt-taxonomy
