I created custom metafields for products and I don’t think they are getting picked up by AI LLMS
What is the importance of JSON structured product feeds and structured schema.
Should I create a /pages/ai-feed and point to this information
Also, what is the value of intent-graph pages
Hello Feel_Good_Haus, my name is Emmanuel, founder of FeedArmy, and your host for today’s Shopify AMA for Google’s AI Overviews.
Good questions, let me take them one at a time, because the first one contains the actual answer.
1. Why your metafields aren’t “picked up” by LLMs: metafields are backend data. AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended) see exactly what Googlebot sees, the rendered HTML of your pages. If your theme doesn’t output the metafield values into visible page content or into your JSON-LD, then no crawler, AI or otherwise, has ever seen them. The Fix: surface the metafield data on the product page itself (spec tables, description blocks, FAQ sections) via your theme/Liquid, and map the relevant fields into your Product schema. Also confirm the content is server-rendered, anything injected client-side by JS is unreliable for many AI crawlers.
2. Value of JSON structured feeds and schema: real, but for specific systems. Product JSON-LD gives Google machine-confident facts (name, price, availability, ratings) and feeds its shopping/AI surfaces; your Merchant Center feed does the heavy lifting for Shopping. But for LLM citation specifically, what matters most is plainly stated facts in the visible content, schema labels your facts, it doesn’t replace having them on the page. Do both: clear visible product info first, accurate schema describing that same info second. (Keep them consistent, mismatches between schema and page content hurt trust.)
3. Should you create /pages/ai-feed? No. This is the llms.txt idea in a different outfit, and the evidence is in: Google has explicitly stated it does not use llms.txt (or similar files) for Search or its AI features, its 2026 guidance literally lists it as a tactic that doesn’t help, and no major AI company has publicly committed to reading such files in production. A custom /pages/ai-feed is even weaker than llms.txt, because bots don’t even know to look for a URL you invented. And if that page duplicates your product content in indexable form, you risk duplicate-content dilution that hurts your real product pages. If you want a cheap hedge, put a standard llms.txt at your domain root, it costs minutes and helps some AI agent tooling, but don’t expect it to move citations or rankings.
4. Intent-graph pages: the idea underneath is legit, the buzzword is marketing. Strip the jargon and it’s topic clusters, separate pages answering genuinely distinct user intents. That works because AI answers are assembled via “query fan-out”: the systems run multiple related sub-queries, and pages that answer those sub-questions get cited even when they don’t rank for the head term. So yes, build pages around real sub-intents your customers have (comparisons, use cases, how-tos, sizing/compatibility questions). Just don’t pay a premium for it because someone rebranded topic clusters as an “intent graph.”
To increase visibility in Google’s AI surfaces (AI Overviews, AI Mode, and conversational shopping), the highest-impact lever for products is your Google Merchant Center product data. Google’s AI shopping answers are built from the Shopping Graph, which is fed by your product feed, so the more complete and structured your data, the more accurately AI systems can understand, represent, and recommend your products when shoppers ask questions in natural language.
I’ve made a video guide covering this here: https://www.youtube.com/watch?v=p3Ay8iy8v0A
Start with these two established attributes:
Product Highlight [product_highlight] : short bullet-style selling points AI can quote directly: Product highlight [product_highlight] - Google Merchant Center Help
Product Detail [product_detail] : structured technical specs (section/name/value), ideal for spec-driven questions: Product detail [product_detail] - Google Merchant Center Help
Then add the new conversational attributes, which Google rolled out specifically to help AI systems and conversational agents understand product nuances. They’re optional, submitted via a supplemental feed, primary feed, or the Merchant API, and adding them won’t affect the approval status of your existing products:
Question and Answer [question_and_answer] : arguably the highest-leverage one. FAQ pairs in plain language directly match how people ask AI (“does it have a headphone jack?”). If your product pages already have FAQs, most of this data exists, it just needs mapping: Question and answer [question_and_answer] - Google Merchant Center Help
Variant Option [variant_option] : lets you declare variant dimensions beyond the standard color/size/material set (e.g. “graphics card”, “scent”, “wattage”), so AI doesn’t have to guess variant differences from titles. Used together with item_group_id: Variant option [variant_option] - Google Merchant Center Help
Item Group Title [item_group_title] : a clean shared title for the whole variant family, so AI can present the product group first, then drill into a specific variant: Item group title [item_group_title] - Google Merchant Center Help
Related Products [related_product] : declares accessories, spare parts, substitutes, “often bought with” relationships, which powers cross-sell answers like “what accessories work with this”: Related product [related_product] - Google Merchant Center Help
Document Link [document_link] : URLs to hosted PDFs (manuals, spec sheets, assembly guides, ingredient lists) that AI agents can reference for detailed questions your description doesn’t cover. Most valuable for electronics, appliances, furniture, tools: Document link [document_link] - Google Merchant Center Help
Popularity Rank [popularity_rank] : a 0–100 value ranking each product against the rest of YOUR catalog, helping AI answer “best” and “most popular” queries. Note this ranks your own products against each other, not against competitors, setting everything to 100 won’t help. Keep it updated as sales data changes: Popularity rank [popularity_rank] - Google Merchant Center Help
A few practical tips:
1. Don’t duplicate data, Google explicitly says not to repeat information already in description, product_highlight, or product_detail inside the conversational attributes.
2. Keep feed and landing page consistent, the details in your feed must match what’s shown on the product page.
3. If your catalog is large, start with your top-selling SKUs: question_and_answer and popularity_rank first, then variant_option + item_group_title if you sell variants.
And beyond these, aim for overall feed completeness, the full product data specification is here, and every accurately filled attribute is another fact AI systems can use: Product data specification - Google Merchant Center Help
You can use apps such as multifeed: Multiple Google Shopping Feeds – Ecommerce Plugins for Online Stores – Shopify App Store
To add custom metafields, that way Google Merchant Center can pick it up.