ChatGPT Shopping, Perplexity, and Google AI Mode: How Each AI Agent Picks From Your Shopify Catalog

By CatalogScan · Published 2026-05-31 · 8 min read

Three AI shopping agents have crossed the threshold from "interesting demo" to "measurable referral traffic." Each one reads the same Shopify catalog differently. Here is what each one actually weights—and the cross-platform optimization sequence that covers all three without tripling your workload.

The three agents and how they differ

ChatGPT Shopping, Perplexity Commerce, and Google AI Mode all send traffic to Shopify stores. They surface stores in different query surfaces (a chat reply, a shopping panel, an AI Overview) and they weight catalog signals differently—but they all read the same underlying data sources: your store’s crawlable pages, your structured data, and in some cases your Merchant Center feed.

The important distinction is not where they appear but how they decide which stores to recommend. ChatGPT Shopping optimizes heavily for exact-match product queries. Perplexity Commerce optimizes for natural-language recommendation queries. Google AI Mode extends the existing Shopping Graph you’ve already built for Google Shopping. All three punish the same category of failure: a catalog that cannot be read at machine speed.

The practical consequence: fixing your catalog for one agent typically moves you forward on the other two. But the priority order depends on which agent is sending your category’s traffic—and each agent has a specific signal it weights that the others treat as secondary.

ChatGPT Shopping: how it picks your store

ChatGPT Shopping runs on a hybrid architecture. GPT-4o uses Bing’s Shopping index as its real-time product catalog, supplemented by OpenAI’s own crawl of Shopify storefronts via GPTBot and OAI-SearchBot. When a user asks “where can I buy a YETI Tundra 45 in white,” the model cross-references the GTIN for that specific variant (888830018989) against every retailer it has ingested. A store without that GTIN set on that variant is not a candidate for that query, even if inventory is ready and price is competitive.

This is the exact-match problem. ChatGPT Shopping does not approximate or infer product identity the way a human would when reading a page. It requires a machine-readable product identifier that traces to a GS1-registered barcode. Stores that rely on their product names and descriptions for Google Shopping’s text-match retrieval do not carry that advantage into ChatGPT Shopping.

What ChatGPT Shopping reads from your Shopify store

1. /products.json — the bulk product feed that GPTBot uses to ingest your entire catalog at once. Headless Shopify stores (Hydrogen, Next.js) that don’t expose this endpoint are invisible to initial crawl-based indexing. See the shopify-feed signal guide for the exact failure pattern.
2. ProductGroup JSON-LD with GTIN on each hasVariant — the structured data shape that maps each variant to its barcode. A parent-only Product schema or a per-variant fragmentation pattern (one Product per color/size) scores partial credit but breaks ChatGPT Shopping’s ability to match specific variants by GTIN. See the product-jsonld signal guide.
3. GPTBot access in robots.txt — the access gate. A Cloudflare Bot Fight Mode rule, an AI Scrapers managed rule, or a blanket User-agent: *\nDisallow: / will block GPTBot at the network layer before any content is read. See Cloudflare for Shopify: the three settings that silently block AI shopping agents.
4. Sitemap.xml — the discovery surface GPTBot reads before crawling individual PDPs. A missing or stale sitemap means GPTBot may miss product pages that /products.json did not surface.

Verifying GPTBot access

curl -si -A "GPTBot/1.0 (+https://openai.com/gptbot)" https://yourstore.com/robots.txt | head -5
curl -si -A "OAI-SearchBot/1.0 (+https://openai.com/searchbot)" https://yourstore.com/products.json?limit=5 | head -5

Both should return HTTP/2 200 with the body. A 403, a Cloudflare challenge page, or a 200 that returns HTML instead of JSON on the second command all indicate a blocking issue.

Perplexity Commerce: how it picks your store

Perplexity Commerce operates differently from ChatGPT Shopping at the query level. Where ChatGPT Shopping excels at exact-match lookup (“find this specific product”), Perplexity’s shopping surfaces handle recommendation queries: “what’s the best cast iron skillet under $80” or “running shoes for wide feet.” For these queries, Perplexity needs enough structured content to generate a confident recommendation with citations—product name, price, availability, review score, and why this specific store is credible.

The consequence for Shopify stores: Perplexity does not tolerate thin product pages. A page with a product name, a single image, and a price will not be cited. A page with a rich description (materials, dimensions, use case, differentiator), an embedded AggregateRating from a reviews app, and a structured FAQ has multiple citation surfaces Perplexity can pull from to answer a complex shopping question with confidence.

What Perplexity Commerce reads from your Shopify store

1. AggregateRating schema on PDPs — Perplexity uses review scores as a trust signal when generating recommendations. A store with Judge.me or Yotpo reviews enabled on their JSON-LD gives Perplexity a numeric confidence anchor. 78% of the top 100 DTC stores in our scan fail this signal despite having review apps installed. See the per-app AggregateRating guide.
2. Description richness — Perplexity uses product descriptions as citation material. Short, marketing-copy-only descriptions ("crafted for the adventurer in you") give the model nothing to quote. Descriptions that include materials, dimensions, use-case specifics, and care instructions give Perplexity the factual content it needs to generate a recommendation paragraph that cites your store.
3. Structured FAQ — If your product pages or cornerstone pages include FAQ schema, Perplexity can surface those answers directly in its response. See the product-jsonld guide for how to embed FAQPage schema alongside your ProductGroup.
4. PerplexityBot access in robots.txt — same access gate as GPTBot. Both must be allowed. Perplexity crawls from PerplexityBot/1.0 (+https://docs.perplexity.ai/guides/bots).

Google AI Mode: how it picks your store

Google AI Mode (previously AI Overviews, now a dedicated search tab in markets where it has rolled out) is the easiest win for most established Shopify merchants—and also the most squandered. Unlike ChatGPT Shopping and Perplexity, Google AI Mode draws from the Shopping Graph you have already been building through Google Shopping and Merchant Center. If your Merchant Center has an active, approved feed, you have a baseline presence in Google AI Mode at launch.

The ceiling for that presence is set by two factors: canonical URL consistency and GTIN coverage. Google’s Shopping Graph uses GTINs as its cross-retailer product identity layer. A store with GTIN populated on all variants and a consistent canonical URL (no ?variant= parameter, no www vs non-www split, no accessible .myshopify.com origin that creates a competing entity) allows Google to cluster all price and availability signals for a given product into a single Shopping Graph node. See the canonical-pdp signal guide.

What Google AI Mode reads from your Shopify store

1. Google Merchant Center feed — the primary data source. Google AI Mode draws real-time price and availability from your Merchant Center data, not from a fresh crawl. A stale, disapproved, or gap-ridden feed is your biggest single Google AI Mode problem, and it will not show up in a CatalogScan scan (we test the live storefront, not your Merchant Center).
2. Canonical URL on PDPs — Google consolidates all signals for a product under its canonical URL. <link rel="canonical"> must point to the clean root URL (https://yourstore.com/products/product-handle) without variant parameters or UTM decoration. Shopify’s default theme handles this correctly; headless builds frequently do not. See headless Shopify: the four signals you silently lose.
3. GTIN coverage — used for cross-retailer price comparison. A Google AI Mode response for a branded product query ("where is the cheapest [brand] [model]") requires GTIN to match store inventory against the same product at other retailers. See Shopify GTIN requirements for AI shopping agents.
4. ProductGroup JSON-LD — used to associate variant availability with a single product entity in the Shopping Graph. Google’s Rich Results crawler (AdsBot-Google, Google-Extended) reads this schema to supplement Merchant Center data for PDPs not individually included in the feed.

Signal priority table: all three side by side

Reading across any row tells you how different the platform priorities are for a given signal. The biggest differences are in AggregateRating (Perplexity weights it critical, ChatGPT Shopping treats it as low priority) and Merchant Center (essential for Google AI Mode, not applicable to the other two).

What our 100-store scan revealed about platform differences

We scanned 100 top-DTC Shopify stores and graded each on the catalog signals AI shopping agents read. Layering that against the signal priorities above reveals a clear pattern: most stores are optimized for the signal that is easiest to set and weakest in the signals that matter most for the highest-value queries.

• 60% fail Product JSON-LD entirely. This is the highest single-signal failure rate of all 15 signals we measure, and it is the signal that directly gates ChatGPT Shopping’s ability to identify and recommend specific products at the variant level. These stores are invisible to ChatGPT Shopping’s most commercially valuable query type.
• 22% have zero GTIN coverage. Another 47% have partial coverage (GTINs on parent SKUs but missing on color or size variants). Combined, that is 69% of top-100 DTC stores that are disqualified from variant-level exact-match queries on both ChatGPT Shopping and Google AI Mode.
• 78% fail AggregateRating despite having reviews apps installed. Most stores have Judge.me or Yotpo installed and visible stars on every PDP. The schema is not in the JSON-LD. Perplexity’s recommendation confidence signal fires on schema, not on visible HTML. These stores get zero credit for their reviews corpus from a machine’s perspective.
• ~40% of stores have no accessible /products.json feed. Many of these are headless Shopify builds (Hydrogen, Next.js + Storefront API) where the standard public endpoints were not replicated. These stores are invisible to GPTBot’s bulk-ingest crawl.

The pattern that emerges: fixing Product JSON-LD, GTIN coverage, and AggregateRating schema addresses the top failure modes for all three platforms simultaneously. These three signals represent different fix timelines (JSON-LD: theme edit, one afternoon; GTIN: CSV reimport, one week; AggregateRating: app toggle, 15 minutes) but they are the highest cross-platform leverage points available without touching Merchant Center.

The 3-pass optimization sequence

The goal is maximum cross-platform coverage with the least duplicated work. The passes are ordered by speed and breadth of impact: the first pass fixes the single most common disqualifying failure across all three agents at once, the second pass opens the two crawl-based agents’ highest-value query type, and the third pass covers the remaining platform-specific gaps.

User-agent: GPTBot
Allow: /

User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Google-Extended
Allow: /

5 mistakes that sink you on all three platforms simultaneously

FAQ