ChatGPT Shopping's Discover-Compare-Decide Flow: What Your Product Data Needs at Each Stage
By Steve Merrill | April 12, 2026
ChatGPT's April 2026 shopping update was described in OpenAI's release notes as improvements to "discover, compare, and decide on products." Simple enough language. But what it actually means is that ChatGPT now runs three distinct product evaluation phases, and each one pulls different data from your Shopify store.
If you haven't thought about which phase is failing you, you're probably failing at all three.
What Is ChatGPT's Discover-Compare-Decide Flow?
It's the structured sequence ChatGPT now uses when a user asks a product-oriented question. Discovery is the filtering stage: which products are even candidates for this query? Compare is the evaluation stage: how do the candidates stack up on the attributes that matter? Decide is the conversion stage: which one should the user actually buy?
Different data requirements at each stage. A product that passes Discovery can fail at Compare if the attribute data isn't there. A product that wins Compare can lose at Decide if the trust signals are weak.
Most Shopify stores haven't thought about their product data in these terms at all. They've got a product title, a description, some photos. That was fine for Google Shopping. It's not enough for ChatGPT's staged evaluation.
What Does Your Product Data Need for Discovery?
Discovery is about matching your product to the right intent. ChatGPT isn't just matching keywords, it's matching intent, use case, and buyer context.
Generic titles fail here. "Ceramic Dog Bowl - 3 Colors Available" doesn't tell ChatGPT enough to surface your product when someone asks "what's a good slow feeder bowl for a lab puppy with digestion issues?" The intent is specific. The product data needs to speak to it.
What actually works at this stage:
- Use-case language in product descriptions: what problem the product solves and for whom, not a feature list
- Product category schema that's specific, Google's product taxonomy is more granular than most Shopify stores use. ChatGPT benefits from it.
- Buyer intent keywords in title and description, the language customers use when they're in research mode, not the language a manufacturer uses in a spec sheet
I've seen stores where a simple rewrite of the first two sentences of each product description, adding use-case context, doubled their appearance rate in ChatGPT discovery queries within three weeks. The product data was there. It just wasn't framed for how AI reads intent.
What Does Your Product Data Need for Comparison?
Comparison is where structured data becomes the deciding factor. ChatGPT is building a table. It needs attributes it can actually compare across products.
If those attributes aren't in your schema or clearly structured on your page, ChatGPT either leaves your product out of comparisons or lists "unavailable" for attributes it can't find. Either way, you lose.
The attributes that matter most for comparison depend on your category. But the ones Search Engine Journal's research on AI-ready product feeds consistently flags as gaps:
- Material composition: the actual materials, spelled out, not "premium" as a stand-in
- Dimensions and weight (in standardized formats)
- Certifications and compliance (safety ratings, eco certifications, third-party testing)
- Warranty terms (explicit duration and coverage)
- Compatibility specs (what this works with, what it doesn't)
These need to be in your Product schema as structured properties. ChatGPT doesn't parse unstructured prose well enough to reliably extract comparison attributes from product description text.
What Does Your Product Data Need for the Decision Stage?
Decision is where buyers convert. ChatGPT's decision-stage signals are trust-weighted, it's essentially asking: "Is it safe to recommend buying this right now?"
The factors that shift ChatGPT toward recommending your product at the decision stage:
- Aggregate review score and review volume in structured data, AggregateRating in Product schema, not purely visual stars on the page
- Clear return policy, in schema or plainly stated near the purchase option. Ambiguous return policies reduce recommendation confidence.
- Inventory availability, Offer schema with
availability: InStock. Out-of-stock or ambiguous availability gets filtered out. - Price accuracy, current price in schema that matches the actual page price. Stale schema prices cause ChatGPT to flag the product as unreliable.
- Brand authority signals, Trustpilot, press mentions, industry certifications that ChatGPT can verify from external sources
Decision-stage failures are often invisible. Your product gets discovered, it gets compared, and then it quietly loses the recommendation to a competitor who has stronger trust data. You never see it happen.
How Do You Audit Which Stage Is Failing You?
Manual testing is the most reliable method right now. Run 10-20 queries in ChatGPT that a typical buyer for your products would ask, broad category searches, comparison questions, "which is best" queries. Watch where your products appear and where they drop out.
Showing up in discovery, missing from comparisons: attribute data problem. Add structured comparison fields to your Product schema.
Showing up in comparisons, losing the final recommendation: trust signal problem. Audit your AggregateRating schema, review volume, and return policy clarity.
If you're not showing up at discovery at all: indexing or content problem. Check Bing Webmaster Tools coverage and rewrite product descriptions with use-case intent language.
Each stage has a fix. The audit tells you which fix to focus on.
Frequently Asked Questions
What changed in ChatGPT's April 2026 shopping update?
OpenAI restructured ChatGPT's shopping flow into three explicit stages: Discover (surface relevant products), Compare (present side-by-side attributes), and Decide (reduce purchase friction). Each stage pulls different product data signals.
What product data does ChatGPT use for the discovery stage?
Discovery relies heavily on product category signals, use-case descriptions, and buyer intent keywords in your product content. Generic titles like "Blue Widget Model X" perform worse than titles that answer "what problem does this solve."
What data does ChatGPT need to include a product in comparisons?
Comparison requires structured attribute data, materials, dimensions, certifications, warranty, weight, compatibility specs. If these aren't in your Product schema or clearly structured on the page, ChatGPT can't populate comparison tables consistently.
How does ChatGPT decide which product to recommend at the decision stage?
Decision-stage recommendations weight trust signals heavily: aggregate review score, review count, return policy, shipping time, and brand authority signals. Products with strong structured review data and clear return policies outperform those with better specs but weak trust data.
