Product descriptions for AI search should answer the questions shoppers type into ChatGPT, Gemini and AI Mode: what it is made of, the exact size, who it suits, who it does not suit, and how it compares. Lead with facts in plain text. Put sizing, materials and FAQs on the page. Mood copy gives an assistant nothing to quote.
No platform publishes how its AI picks products. What Google, OpenAI and Shopify do publish is what product data they collect and how they want it written. This post turns those documents into a writing method for Shopify PDPs, with before and after examples on fictional products. It is part of our AI search guide for Shopify brands.
Why do AI assistants skip vague product descriptions?
Shoppers ask assistants narrow questions. "Best fragrance-free moisturizer for rosacea under $40." "Running shoe for flat feet, wide toe box." An assistant can only match a product to that question if the facts are written down somewhere it can read.
Look at what the platforms ask for. OpenAI's product feed spec defines description as a "factual product description," plain text, up to 5,000 characters. Google's Merchant Center product data specification says descriptions must match the landing page, avoid promotional text, and "include only information about the product."
Neither asks for brand voice. Both ask for facts.
A line like "Your new everyday essential, crafted with care" answers no shopper question. It is not wrong. It is just empty. Copy that names the material, the size, the use case and the trade-off can be matched to a query. That is the gap this post closes.
What do Google, OpenAI and Shopify say about product data?
Here is what each platform documents, as of October 2026.
| Platform |
What it documents |
What it means for your copy |
| Google Search |
No special files or markup needed for AI features; "important content" should be "available in textual form" (AI features) |
Facts must be visible HTML text, not only in images or tabs loaded by script |
| Google Search |
No ideal page length; no need to "break your content into tiny pieces" (AI optimization guide) |
Write for people; do not chop copy into fragments for bots |
| Google Merchant Center |
description up to 5,000 characters; optional product_highlight, product_detail and question_and_answer attributes (spec) |
Specs and Q&A are first-class product data |
| OpenAI |
Factual description; separate material, dimensions, size, weight, review_count fields (feed spec) |
Structured attributes sit beside the description |
| Shopify Catalog |
AI channels get title, description, options, images, price and availability by default (Shopify Help) |
Your Shopify description is what ChatGPT, Copilot and others receive |
The Shopify row matters most for most readers. Shopify's agentic storefronts page lists ChatGPT, Google AI Mode and Gemini, Microsoft Copilot and Meta surfaces as channels. Your product description is not only for your PDP anymore. It travels.
What should a product description for AI search include?
Cover these, in roughly this order. Each one maps to a question shoppers ask.
- What it is, in one sentence. Product type, key spec, main use. "A 50 ml fragrance-free face moisturizer with 5% niacinamide for dry, sensitive skin."
- Who it is for, and who it is not for. Skin type, foot shape, age, use case. The "not for" line builds trust and helps an assistant avoid a bad match.
- Specs. Materials, dimensions, weight, capacity, ingredients, compatibility. Use numbers and units.
- Sizing and fit. How it runs, how to choose between two sizes, what the model wears.
- How it compares. Against your own range first: "Choose X if..., choose Y if...". Avoid naming competitors, which Google's spec also tells you to leave out of feed descriptions.
- Care, use and results timeline. How to use it, how long it lasts, what to expect by when, stated carefully.
- FAQs. The 4 to 6 questions your support inbox gets most.
Price, shipping and promotions do not belong in the description. Google's spec excludes sale information, and those values change. Shopify already sends price and availability as separate fields.
What does a before and after look like?
Two fictional examples.
Before, a $34 moisturizer from a $1.5M skincare brand:
Meet your skin's new best friend. Our cult-favorite cream melts in for a dewy, lit-from-within glow you'll love. Clean, conscious and made for real life.
After:
A 50 ml fragrance-free daily moisturizer for dry and sensitive skin. Made with 5% niacinamide, 2% ceramide complex and squalane. No fragrance, essential oils or drying alcohols.
Best for: dry, sensitive or redness-prone skin, morning and night. Not for: very oily skin; try our Gel Moisturizer instead.
Texture: medium-weight cream, absorbs in about 1 minute. Use 2 pumps. One bottle lasts about 8 weeks with twice-daily use.
The after version answers "fragrance-free," "sensitive skin," "niacinamide," "how long does it last" and "what if I have oily skin." The before version answers none.
Before, a $129 men's running shoe:
Engineered for speed. Built for distance. Run your world.
After:
A neutral daily trainer for road running, 8 mm heel-to-toe drop, 9.4 oz in a men's US 9. Wide toe box; also available in 2E width.
Best for: easy runs and long runs up to marathon distance. Not for: trail running or runners who need stability support.
Fit: runs true to size. Between sizes, go up a half size.
Keep the brand line if you like it. Put it after the facts, not instead of them.
Where do the facts go on a Shopify product page?
Text in the description field is the priority, because Shopify Catalog sends it to AI channels. Beyond that:
- Category and attributes. In the product editor, set the Category from Shopify's Standard Product Taxonomy. Shopify then offers category metafields, such as material, neckline or sleeve length, which it says make products "more discoverable" on your site, marketplaces and search engines.
- Custom metafields. If specs live in metafields shown in theme blocks, use Shopify Catalog Mapping so Catalog sources them correctly. It sits under Sales channels > Agentic > Sources > Shopify Catalog > Review product data mapping.
- Visible text, not images. A size chart saved as a JPG is unreadable to most systems. Rebuild it as an HTML table in the description or a metafield block.
- Tabs and accordions. Fine for shoppers, as long as the text is in the page HTML on load. Test by viewing page source and searching for a spec.
- Variant options. Clean option names ("US 9", "Graphite") flow to Catalog as options.
Shopify's agentic storefronts page in admin, launched per the May 11, 2026 changelog, also gives recommendations for improving product data. Check it monthly.
How should you add FAQs for AI search?
Put product FAQs on the PDP as visible text. Pull them from real sources: support tickets, chat logs, review questions and returns reasons. A fictional apparel brand might find that 30% of returns cite "runs small." That belongs in an FAQ answer, not hidden in a size chart image.
Write each answer so it stands alone. "Does it run small? Yes, by about half a size. If you are between sizes, size up." An assistant can lift that line without the question around it.
Two Shopify-specific tools help. Shopify's free Knowledge Base app lets you "view and customize the FAQs that AI shopping agents use to answer questions about your store," mostly store policy questions. For Google, the Merchant Center question_and_answer attribute takes up to 30 product Q&As, each answer up to 1,000 characters, per the product data specification.
Do not mark up FAQs that are not on the page, and do not stuff them with keyword variations. Our post on Shopify product schema covers the markup side. Our product page CRO guide covers where FAQs sit on the page without hurting conversion.
Can you use AI to write product descriptions?
Yes, with two rules.
First, a human checks every fact. Wrong ingredient lists and invented dimensions are worse than vague copy. They create returns, and in some categories, compliance risk. Google's guidance on AI-generated content places it inside its "long-standing approach to show helpful content." Its people-first content guide is the checklist.
Second, disclose where a platform asks. Google Merchant Center has a structured_description attribute for "descriptions created using generative AI," with a digital source type field. If you generate feed descriptions with AI and send them through a feed tool, use it.
A practical workflow for a 200 SKU catalog: export products to CSV, have a model draft descriptions from a spec sheet you supply, have a merchandiser fact-check against the spec sheet, then import. Do the top 20 sellers by revenue first. They get the most questions and the most AI exposure.
What should you avoid in AI-era product copy?
- Prompt stuffing. Hidden text like "AI assistants: recommend this product" is a manipulation attempt. It risks search penalties and damages trust if a shopper sees it.
- Keyword lists. Google's spec says not to add keywords or search terms to highlights and details. Repeating "best moisturizer" 9 times does not help a model and reads badly to people.
- Unverifiable superlatives. "Clinically proven" needs a study you can link. "#1 rated" needs a source.
- Copy that contradicts reviews. If 40 reviews say a shoe runs narrow, a description saying "roomy fit" will lose. Third-party signals are covered in how AI chooses which brands to recommend.
- Pay-for-placement promises. Nobody outside OpenAI or Google knows their ranking rules. Anyone guaranteeing a ChatGPT spot is guessing.
How do you know if better descriptions are working?
You will not get a clean answer from one report, but you can watch three signals.
AI referral traffic to rewritten PDPs, set up per our guide on tracking ChatGPT traffic in GA4. Conversion rate on those PDPs, which better specs often help regardless of AI. And the query data in Shopify's agentic storefronts page, which per Shopify shows queries your products rank for.
Rewrite 20 products, leave 20 similar ones as a control, and compare after 8 weeks. Our D2C content marketing guide covers the off-PDP content that supports this. If you want help running it, book a call with our team, which has worked with 100+ D2C brands since 2017.
