TL;DR
- Most AI-shopping advice today is written for ChatGPT. But Gemini and Claude read a catalog differently, and a strategy tuned only for one engine leaves real recommendation traffic on the table.
- Gemini's shopping answers are increasingly powered by Google AI Overviews, which draws on Google's existing Shopping Graph and indexed product feeds — a different data pipeline than ChatGPT's browsing-and-retrieval approach, with different structured-data dependencies.
- Claude doesn't operate a dedicated shopping/checkout surface the way ChatGPT does, but it still reads and can recommend products when a user asks it directly — and it depends even more heavily on clean, complete on-page structured data since it has no separate shopping index to fall back on.
- The common fix across all of them is the same underlying layer: structured, machine-readable product data. A catalog built for one AI crawler and not the others is a catalog that's only partially discoverable, no matter how well any single engine likes it.
What is Google AI Overviews, and how is it different from ChatGPT's shopping answers?
Google AI Overviews is the AI-generated summary Google shows above traditional search results, increasingly incorporating shopping-specific product carousels and comparisons pulled from Google's Shopping Graph — the same structured product-feed infrastructure behind Google Shopping ads and free listings. That's a meaningfully different mechanism than ChatGPT's approach, which relies more on live browsing and retrieval from a page's content and structured data at query time rather than a pre-built product graph.
The practical consequence: a product's visibility in Gemini/AI Overviews depends partly on factors ChatGPT doesn't weigh the same way — Merchant Center feed quality, Google Shopping eligibility, and product-feed structured data (Product, Offer, AggregateRating schema) — in addition to the on-page JSON-LD that also matters for ChatGPT. A store that's only optimized its on-page markup, without also maintaining a clean product feed, is covering ChatGPT's retrieval path but leaving a gap in Gemini's.
Does Claude recommend products, and if so, how?
Claude does not currently operate a dedicated shopping or checkout experience the way ChatGPT's shopping features do, but a user can still ask Claude directly for product recommendations, and Claude will read whatever it can access — a brand's site content, structured data, and any pages it's able to retrieve — to form an answer. Because Claude doesn't have a separate pre-built shopping index to lean on the way Gemini can lean on the Shopping Graph, its recommendation quality depends almost entirely on how legible the underlying page is at the moment it's queried: clean JSON-LD, clear factual product copy, and content that isn't hidden behind client-side rendering.
That makes Claude something of a stress test for a catalog's baseline AI-readability — if a product page can't produce a confident, specific answer for Claude, it's a signal the underlying structured data itself is incomplete, not just under-optimized for a particular engine's ranking quirks.
How do ChatGPT, Perplexity, Gemini, and Claude actually differ in what they need from a catalog?
| Engine | Primary data source | What a catalog needs to provide |
|---|---|---|
| ChatGPT | Live browsing/retrieval plus any connected shopping data sources | Clean, complete on-page JSON-LD; content that renders without requiring JS execution; crawlable by GPTBot |
| Perplexity | Live retrieval across indexed and browsed sources, weighted toward citation-worthy pages | Similar on-page structured data needs, plus content structured for direct quotability; crawlable by PerplexityBot |
| Gemini / Google AI Overviews | Google's Shopping Graph plus indexed page content | On-page JSON-LD and a maintained Merchant Center product feed; crawlable by Google-Extended and standard Googlebot |
| Claude | Direct page retrieval at query time, no dedicated shopping index | Especially reliant on complete, current on-page structured data since there's no separate index to compensate for gaps; crawlable by ClaudeBot |
The pattern across all four: every one of them ultimately needs the same underlying thing — factual, structured, current product data — but they access and weigh it through different pipelines, which is why a catalog can be well-optimized for one engine and still be thin for another.
Where does Agentic Page fit across all four engines?
Agentic Page's job is to produce one structured, machine-readable mirror of a Shopify catalog that stays legible across GPTBot, PerplexityBot, ClaudeBot, and Google-Extended simultaneously, rather than treating any single AI crawler as the target. This is the same cross-crawler legibility principle already behind Agentic Page's approach to JSON-LD and llms.txt — the difference here is extending that principle explicitly to Gemini's Shopping Graph dependency and Claude's stricter reliance on page-level completeness, rather than assuming a ChatGPT-tuned catalog automatically covers the other three.
This sits at the Discovery layer of DeepLumen's Discovery → Intention → Deal (D.I.D.) framework — the stage that determines whether a catalog can be found and trusted by any AI shopping agent in the first place, regardless of which one a given shopper happens to be using.
What should a brand actually do differently for Gemini and Claude specifically?
- Don't assume ChatGPT optimization covers Gemini. If a store hasn't checked its Google Merchant Center feed quality and Shopping Graph eligibility separately from its on-page JSON-LD, it likely has a real gap specifically in Gemini/AI Overviews visibility.
- Treat Claude as a completeness check, not an afterthought. Since Claude has no separate index to fall back on, testing whether it can answer a specific product question well is a useful diagnostic for whether the underlying page data is actually complete — not just whether it's been tagged for a particular engine.
- Confirm all four crawlers are actually allowed to access the site. robots.txt rules blocking GPTBot, PerplexityBot, ClaudeBot, or Google-Extended individually is a common and easy-to-miss gap — checking one bot's access doesn't confirm the others.
- Track engine-specific signals separately, not as one blended "AI traffic" number. A brand that's strong on ChatGPT referrals but weak on Gemini (or vice versa) needs that broken out to know where the actual gap is.
Where DeepLumen fits
DeepLumen positions Agentic Page as the product responsible for exactly this cross-engine consistency — maintaining one structured, AI-readable version of a Shopify catalog that's built to be legible across GPTBot, PerplexityBot, ClaudeBot, and Google-Extended at once, rather than requiring a merchant to separately optimize for each AI shopping surface as it emerges.
For the technical foundation underneath all four engines, see JSON-LD and llms.txt for Shopify: A Technical Setup Guide; for what fixing it is worth in revenue terms, see 90% AI-Invisible, +245% After Fix.
See whether your catalog is actually legible across ChatGPT, Perplexity, Gemini, and Claude — not just the one you've been checking. Learn more about Agentic Page or book a demo for a cross-engine ACCC scan of your store.