AI shopping channels — ChatGPT Shopping, Perplexity AI shopping results, and Google AI Mode product surfacing — are fundamentally different from traditional Google Shopping in how they evaluate and select products. Understanding these differences changes what "feed optimisation" means for AI channels.
The short version: traditional Shopping rewards keyword density and GTIN completeness. AI channels reward semantic accuracy, factual specificity, and description quality. The same product data that performs well on Google Shopping may be average on ChatGPT Shopping — not because it's wrong, but because it's optimised for a different evaluation model.
How ChatGPT Shopping selects products
ChatGPT Shopping processes natural language queries ("I need a waterproof jacket for hiking in Scotland in November") rather than keyword searches. The underlying LLM evaluates product descriptions against query intent using semantic understanding, not keyword matching. A description that says "waterproof" once and doesn't elaborate is less likely to be surfaced than one that says "fully waterproof with a Hydrodry 10,000mm waterproof rating, taped seams, and adjustable hood — suitable for sustained rain in outdoor conditions."
The ChatGPT JSONL specification explicitly states that descriptions should be "informative and accurate". This is OpenAI's way of saying: the model evaluates description quality, not just the presence of keywords. Descriptions that are clearly AI-generated boilerplate, overly promotional, or thin on specifics perform worse than descriptions that genuinely explain what the product is and who it's for.
The three description types and their AI performance
Type 1 — Promotional copy: "Elevate your outdoor adventures with our premium waterproof jacket. Featuring cutting-edge technology and superior craftsmanship, this jacket is the perfect companion for all your outdoor pursuits." This is the worst type for AI channels. Vague, promotional, no specific attributes. ChatGPT's model gives it low relevance to specific queries.
Type 2 — Feature list: "Waterproof. Breathable. Adjustable hood. Multiple pockets. Lightweight. Available in Navy, Black, Green." Better than promotional, but lacks context and specificity. The model can extract some attributes but can't assess whether this product genuinely fits a specific use case query.
Type 3 — Factual specification description: "Lightweight waterproof hiking jacket with a Hydrodry 10,000mm waterproof and 5,000g breathability rating. Fully taped seams prevent water ingress in sustained rain. Features an adjustable storm hood, two hip pockets, and one chest pocket with YKK zips. 100% recycled nylon shell. Pack weight 380g. Suitable for hiking, trail running, and outdoor activities in wet UK conditions." This is the ideal type for AI channels — and it will also perform well on Google Shopping because it naturally includes relevant keywords and is factually specific.
Perplexity AI: product schema and crawling
Perplexity sources product data differently from ChatGPT Shopping. Rather than a submitted JSONL feed, Perplexity primarily crawls your product pages and reads structured data from your product schema markup. Optimising for Perplexity means ensuring your Shopify or WooCommerce product pages have complete, accurate Product schema including: name, description, offers (price, availability, currency), brand, gtin, and image. Google-rich-results-compliant Product schema is directly readable by Perplexity's crawlers.
The practical implication: your ChatGPT Shopping feed optimisation (better descriptions, factual specificity) also improves your website's product schema quality if you push those improved descriptions back to your store via AI enrichment. Better feed descriptions that are also on your product pages serve both ChatGPT and Perplexity.
Google AI Mode: Shopping feed meets LLM
Google AI Mode (formerly SGE) surfaces products in AI-generated search responses by drawing from your Google Shopping feed via Merchant Center. The same feed that powers your Shopping ads feeds Google AI Mode. This means traditional Shopping feed optimisation (better titles, complete GTINs, correct categories) continues to be relevant — but the AI layer also rewards description quality for queries where AI Mode generates a narrative response around products.
The intersection point: products with complete, high-quality descriptions in both the Shopping feed and on the product page (read by Google's Googlebot) appear more frequently in AI Mode responses for conversational queries. Products that have feed title optimisation (good keyword matching) AND description quality (good AI channel relevance) perform across both traditional Shopping and AI Mode placements.
Trajekt's AI enrichment for AI-channel optimised descriptions
Trajekt's Claude-powered description enrichment generates factual, specification-led descriptions from your source product data. The enrichment prompt is configured specifically for AI channel requirements: plain text output (no HTML), 150–300 words, factual and specific, uses attributes present in the source data, no promotional language, describes what the product is and who it suits.
Descriptions are cached by product and only re-generated when the source product data changes — so the ongoing API cost scales with how frequently your product range changes, not with catalogue size. For a stable catalogue of 5,000 products, the daily enrichment cost after the initial batch run is typically under £1.
The unified feed strategy
The good news is that AI-channel-optimised feed content isn't at odds with traditional Shopping optimisation. The qualities that make a product description effective for ChatGPT (factual, specific, complete) are the same qualities that make it effective for Perplexity (specific attributes, readable prose) and compatible with Google AI Mode (information-dense, query-relevant). The worst-performing description type — vague, promotional — is bad on all four channels.
A single feed investment — improving description quality across your catalogue — compounds across traditional Shopping, ChatGPT Shopping, Perplexity, and Google AI Mode simultaneously.
Trajekt is a UK-built product feed management platform. Our editorial team covers feed optimisation, Google Shopping strategy, ChatGPT commerce, and ecommerce channel management for UK merchants and agencies.