When a shopper asks ChatGPT or Google to find the best waterproof trail-running shoes under $150, an AI agent does the choosing. It does not browse your storefront or admire your hero image. It reads structured product data: titles, descriptions, specifications, prices, and attributes, then matches them against the question. Google's own product structured data documentation is explicit that merchant listings are built from machine-readable fields like name, image, and a priced offer, and OpenAI's product feed for agentic shopping works the same way. The persuasive marketing language your team labored over is mostly invisible to the model.
This is a quiet but sweeping change for e-commerce. For twenty years, product copy was written to convince a human who had already landed on the page. Now the first reader is an agent deciding whether your product even enters the shortlist. Getting recommended means feeding that agent clean, accurate data in the fields it parses. This post breaks down what AI shopping agents read, and how to write product content, from titles and specs to descriptions and answers, that earns the recommendation.
Key Takeaways
- AI shopping agents rank products from structured fields (title, description, price, availability, attributes), not marketing prose.
- OpenAI's Agentic Commerce Protocol requires nine core fields per item, and Google's Product markup requires name, image, and a priced offer at minimum.
- Complete, consistent specs and identifiers (GTIN, MPN, size, color, material) are what let an agent match your product to a query.
- Thin, duplicated manufacturer copy and mismatched data between page and feed are the fastest way to get filtered out.
What AI Shopping Agents Actually Read
An AI agent ingests a structured record and treats each field as a fact it can trust or discard. OpenAI's Agentic Commerce Protocol product feed spec makes the requirement concrete: every item needs at least nine fields, including item_id, title, description, url, brand, seller_name, image_url, availability, and price, organized across roughly fifteen field groups. Google's merchant listing structured data is just as blunt: the Product needs a name, an image, and an Offer with a price and a three-letter currency code. Both systems reward products described in structured, non-promotional terms, and the recommended fields, such as ratings, reviews, GTIN, brand, and category, are the signals that decide which of ten similar products gets surfaced. If your data is thin or contradictory, the agent has no reason to pick you and every reason to pick a cleaner competitor.
Write Titles That Describe, Not Sell
The product title is the single most important field an agent reads. A title like "The Only Rain Jacket You Will Ever Need" tells a model nothing it can match. Compare "Patagonia Torrentshell 3L Waterproof Rain Jacket, Men's, Black, Large," which names the brand, product type, key attribute, gender, color, and size, precisely the language a shopper's query contains.
The ACP spec recommends keeping titles to at most 150 characters and including variant details like size and color directly in the title. Front-load the information that matters: brand, product type, then defining attributes. Drop the superlatives and the SEO-era keyword stuffing. An accurate, descriptive title lets the agent confirm your product answers the exact question asked, and it does double duty for classic search. Our guide to product page SEO that drives sales covers how the same title discipline lifts human conversion too.
Complete Specs and Structured Attributes
Attributes are the tiebreakers. When a shopper asks for a "size 10 waterproof hiking boot with a Vibram sole," the agent is filtering on structured attributes. Google recommends adding brand, gtin or sku, category, and availability alongside the required fields, and the ACP feed carries a full item-information group for material, color, size, dimensions, and weight. Product identifiers matter more than most merchants realize: a valid GTIN or MPN lets an agent match your listing to a known product across the web and reconcile it with reviews and pricing elsewhere.
Fill every attribute that applies, and fill it consistently. Leave "material" blank and your boot silently loses every query that specifies one. The more attributes you populate accurately, the more specific queries your products can satisfy. For stores managing thousands of SKUs, our category page strategy for large catalogs shows how to keep attribute data structured at scale.
Descriptions: Benefit Plus Fact, Never Boilerplate
The description field is where writing craft still matters, but not the way it once did. The ACP spec asks for plain text, factual content up to roughly 5,000 characters, and Google lists description as a recommended property meant to convey real product details. That rules out two habits. Marketing-only prose that states no facts gives the agent nothing to index. Worse is thin duplicated manufacturer copy: when hundreds of retailers paste the same supplier description, none stands out, and Google's 2026 core updates lean hard against thin, mass-produced pages.
Write descriptions that pair a fact with the benefit it delivers. "A 3-layer waterproof membrane keeps you dry in sustained rain" beats both "stay dry out there" and a copied spec sheet. Cover what it is made of, what it fits, and how it performs. Unique, factual, benefit-anchored descriptions are readable by humans and parseable by agents, the combination that wins in AI-era commerce.
Answer the Questions Shoppers Actually Ask
Agentic queries are conversational, so the products that win are the ones whose pages already answer follow-up questions like "Is this true to size?" or "What is the return window?" Add a genuine Q&A or FAQ block to your product pages and answer these plainly. Agents lift concise, self-contained answers the same way AI Overviews do, which is why the mechanics overlap with winning featured snippets for product queries.
Reviews matter here too. Google lists aggregateRating and review as recommended properties, and the ACP feed reserves a group for ratings because they improve ranking and trust. Honest, marked-up review data tells an agent that real customers have validated your claims, often the deciding factor when two products match a query equally well on paper.
Keep Your Data Consistent Across Page and Feed
The fastest way to get filtered out is contradiction. If your on-page price says $129, your structured data says $149, and your feed says out of stock, an agent cannot trust any of the three and moves on. Google explicitly recommends providing both structured data on your pages and a Merchant Center feed, because the two cross-verify each other. Every channel an agent reads, your Shopify page, your feed, and any marketplace listing, should state the same price, availability, and attributes.
Consistency is an operational problem, not a copywriting one. It requires a single source of truth for product data and automated syncing so a price change or a stockout propagates everywhere at once. This is the foundation we build for merchants through our Shopify SEO agency services, aligning on-page content, structured data, and feeds so agents read one clean, current story about every product.
AI shopping agents have not made product content less important, they have made accuracy and structure the whole game. Write titles that describe, fill every attribute, replace boilerplate with factual benefit copy, and keep your data identical across page and feed. Do that and your products become the ones ChatGPT and Google reach for. To see how AI agents currently read your catalog and where your data contradicts itself, talk to our team and we will map every field that matters.
Read More
- Product Page SEO That Drives Sales
- Winning the Featured Snippet for Product Queries
- Category Page Strategy for Large Catalogs
- Agentic Search: How to Be the Business AI Picks, Calls, and Books
Frequently Asked Questions
What product content do AI shopping agents actually read?
Agents read structured fields, not marketing prose: the title, description, price, availability, images, brand, and attributes like size, color, material, and identifiers. OpenAI's Agentic Commerce Protocol feed requires nine core fields per item, and Google's Product markup requires a name, an image, and a priced offer at minimum.
How should I write a product title for AI agents?
Describe the product instead of selling it. Lead with brand and product type, then add defining attributes and variant details like color and size. OpenAI's spec recommends keeping titles under about 150 characters. "Patagonia Torrentshell 3L Waterproof Rain Jacket, Men's, Black, Large" matches shopper queries far better than a slogan.
Why is duplicated manufacturer copy a problem?
When many retailers use the same supplier description, no listing stands out and the page reads as thin, mass-produced content that Google's recent core updates demote. Agents also gain nothing unique to index. Write your own factual descriptions that pair each fact with a benefit.
Do product identifiers like GTIN and MPN matter for AI search?
Yes. A valid GTIN or MPN lets an agent match your listing to a known product across the web and reconcile it with reviews and pricing elsewhere. Google recommends including gtin or sku, and the Agentic Commerce Protocol treats identifiers as a core way to reduce catalog matching errors.
Why does data need to match between my page and my feed?
If your page, structured data, and product feed disagree on price or availability, an agent cannot trust any of them and skips your product. Google recommends providing both on-page structured data and a Merchant Center feed so they cross-verify. Keep price, stock status, and attributes identical across every channel so ChatGPT, Google AI Mode, and other shopping agents read one clean story about every product.
Want this handled for you? Our eCommerce SEO services can help you scale rankings and revenue across your catalog.
Aashirvad Kumar
Writes about SEO, digital marketing and growth at Optimize For SEO.