AI Shopping Assistants Are Here: What They Read From Your Catalog

AI shopping assistants and your catalog: what matters now

AI shopping assistants are no longer a future concept — they are actively recommending products, comparing options, and influencing purchase decisions at scale. Google AI Mode handles shopping queries by pulling from Merchant Center feeds and crawled product pages. Amazon Rufus answers questions and compares products using catalog data. ChatGPT and Perplexity are building shopping features that retrieve product information from across the web.

For ecommerce teams, this creates a new requirement: your catalog must be readable by machines, not just attractive to humans. The product page that converts a browsing shopper may be invisible to an AI assistant if the data is incomplete, unstructured, or inconsistent.

The practical impact is already measurable. AI-assisted shopping sessions tend to convert at 2-3x higher rates because the assistant has already addressed the shopper's objections. But your products only benefit if the assistant can find and parse them.

This guide covers what each major AI assistant reads, which fields matter most, and what you can fix today. For a look at how Amazon's assistant is specifically evolving, see our article on Amazon Rufus and agentic shopping.

How each AI assistant reads product data

Each assistant has a different data pipeline, but they converge on similar requirements.

Google AI Mode

Google AI Mode is the most feed-dependent assistant. It draws from:

• Merchant Center product feeds — title, description, price, availability, brand, GTIN, image, product type, and custom attributes
• Crawled product pages — HTML content, schema markup (Product, Offer, AggregateRating), and structured specs tables
• Google Shopping Graph — an internal knowledge graph that aggregates product data from feeds, crawled pages, and third-party sources

If your Merchant Center feed is strong and your PDP has complete schema markup, you have the best chance of appearing in AI Mode shopping results. See our Google Merchant Center feed optimization guide for feed-specific best practices.

Amazon Rufus

Rufus operates within Amazon's ecosystem and reads:

• Listing content — title, bullet points, A+ content, backend keywords
• Structured attributes — product type, category attributes, item specifics
• Customer reviews — review text, ratings, common themes
• Q&A data — questions and answers on the listing
• Pricing and availability signals — price history, stock patterns, Prime eligibility

The key takeaway: Rufus can answer questions about your product that go far beyond what your listing headline says. If your backend attributes are incomplete, Rufus has less to work with.

ChatGPT shopping

ChatGPT's shopping experience retrieves product data from:

• Crawled web content — product pages, blog reviews, comparison articles
• Affiliate and commerce APIs — structured product feeds from partner networks
• Schema markup — Product and Offer schema on indexed pages
• User reviews and community content — aggregated sentiment from crawled review sites

ChatGPT tends to recommend products it can confidently cite with specific attributes. Vague product pages with thin content rarely appear.

Perplexity shopping

Perplexity follows a similar pattern to ChatGPT but with a stronger emphasis on source citation. It pulls from:

• Crawled product and review pages
• Schema markup
• Trusted comparison sources
• Recent content (Perplexity tends to favor freshness)

The fields AI assistants prioritize

Across all assistants, certain product data fields are consistently more influential:

Tier 1: Required for recommendation

These fields are essential. Without them, your product is unlikely to be recommended:

Tier 2: Critical for comparison queries

These fields determine whether your product wins in head-to-head comparisons:

Tier 3: Differentiators and trust signals

These fields help the assistant choose between similar products:

For a complete field-by-field reference, see our product data quality checklist.

Common catalog gaps that block AI recommendations

Most ecommerce catalogs have systematic gaps that prevent AI assistants from recommending products. Here are the most common:

• Missing structured attributes — the description says "waterproof" but there is no waterproof rating attribute, so the AI cannot filter on it
• Vague descriptions — "premium quality materials" tells the AI nothing. "650-fill-power goose down, 20D ripstop nylon shell" tells it everything.
• Inconsistent pricing — the PDP shows one price, the feed another, and the marketplace a third. The assistant loses confidence.
• No schema markup — the product page has all the data visually, but no JSON-LD for machines to extract it
• JavaScript-rendered content — key specs are loaded via JavaScript. AI crawlers (GPTBot, ClaudeBot) do not execute JavaScript, so they see an empty page.
• Missing variant data — the parent product exists but individual size/color variants are not in the feed or schema

Tools like Lasso are designed to close these gaps at scale — enriching missing attributes from manufacturer data and web sources, generating structured descriptions, and standardizing variant data across channels. See how teams across different industries apply this in our use cases.

How to audit your catalog for AI readiness

Run this practical audit to assess whether AI assistants can effectively read your catalog:

Step 1: Test with the assistants directly

Ask each major assistant about your products:

• "What's the best [your product category] under [price]?"
• "Compare [your product] with [competitor product]"
• "[Your brand name] [product name] specs"

If your product does not appear, the assistant either cannot access your data or considers it incomplete.

Step 2: Check technical accessibility

• Robots.txt — verify that GPTBot, ClaudeBot, and other AI crawlers are not blocked
• JavaScript rendering — test if your product pages serve complete HTML without JavaScript execution
• Schema markup — validate Product and Offer schema with Google's Rich Results Test
• Feed completeness — audit your Merchant Center feed for optional-but-valuable attributes you are not sending

Step 3: Score attribute completeness

Pick your top 100 SKUs and score them:

• Do they have all Tier 1 fields filled? (target: 100%)
• Do they have Tier 2 fields filled? (target: 80%+)
• Are descriptions specific and attribute-rich? (not vague marketing)

The gap between your current score and the targets is your optimization backlog.

Making your catalog AI-assistant ready

The shift to AI-assisted shopping rewards the same thing good product data management has always required: completeness, accuracy, and structure. The difference is that now, gaps are not just a missed SEO opportunity — they are a direct exclusion from a growing sales channel.

Start with your highest-value products. Enrich their attributes, add schema markup, verify feed accuracy, and test with the assistants directly. Then scale the process across your catalog.

If you want to accelerate this work, explore how Lasso enriches and structures product data or get in touch for a walkthrough tailored to your catalog.