How do you write product descriptions that AI shopping assistants can use?
Write facts, not slogans. Open with what the product is and who it's for, then list the details a shopper would ask about: size, material, weight, compatibility, what's in the box. Keep the same facts on your page, in your schema markup and in any product feed. OpenAI asks for a factual description, and Google bans promotional text in it.
- ChatGPT reads your description: OpenAI's help centre lists "structured metadata ... (e.g., price, product description)" among the things ChatGPT considers when it picks products to show.
- It may rewrite it: OpenAI says ChatGPT may generate simplified product titles and descriptions, because merchants describe the same product in different ways.
- OpenAI's feed rules: the description field is required and should be a "factual product description" in plain text. OpenAI suggests aiming for at most 150 characters for titles and 5,000 for descriptions.
- Google's rules for Merchant Center: up to 5,000 characters, the most important details in the first 160–500 characters, and no price, shipping, sale dates, company name or comparisons with other products.
- AI-written copy is labelled at Google: descriptions created with generative AI must go in Merchant Center's
structured_descriptionattribute, flagged astrained_algorithmic_media. - No special AI format: Google says there are no extra requirements to appear in AI Overviews or AI Mode, and that structured data should match the visible text on the page.
Why descriptions matter more in a chat
On a search results page, a shopper types "trail running shoes" and scans the list. In ChatGPT or Google AI Mode, they write the whole brief at once: "waterproof trail shoes for wide feet, under €120, that ship to Portugal". Then they follow up: "which of these is lightest?"
An assistant can only match your product to that brief if the facts are written down somewhere it can read. If your page says "the ultimate shoe for every adventure" but never says it's waterproof, comes in wide fit or weighs 290 g, the assistant has nothing to match. It will pick a store whose page does say those things.
OpenAI describes the same idea in its guide to shopping with ChatGPT: a product appears when ChatGPT judges it relevant to the shopper's intent, and the factors shift with the question. Its example: give a $30 budget and ChatGPT focuses more on price; leave price out and it looks at other things.
What OpenAI and Google ask for
Both companies publish rules for product data. They mostly agree.
| OpenAI product feed | Google Merchant Center | |
|---|---|---|
| Title | Required. Product name, including the selected variant when relevant. Aim for at most 150 characters | Required. 1–150 characters. Put the most important details first; Google says users usually notice only the first 70 or fewer characters |
| Description | Required. "Factual product description", plain text, aim for at most 5,000 characters | Required. 1–5,000 characters. Most important details in the first 160–500 characters |
| Promotional text | Not covered in the field rules; price and sale price have their own fields | Not allowed: no price, sale dates, shipping, delivery dates or company name |
| Comparisons | Not covered | Not allowed ("even better than X") |
| Variants | One row per variant, each with its own price, stock, URL and images | Put each variant's distinguishing details (colour, size) in the title |
| Attributes in their own fields | Material, colour, size, dimensions, weight, condition, GTIN, reviews, returns | Colour, size, age group, material, pattern and more |
| AI-written text | Not covered | Must use structured_description with trained_algorithmic_media |
Two takeaways. First, the description should describe the product and nothing else: Google says not to include your business history or policies, and to put price, shipping and returns in their own attributes. Second, both systems want the key attributes as separate data too, not only buried in prose.
If you don't send a feed at all, the same advice holds for the product page itself, because that's what ChatGPT's search crawler and Google read. We explain how products reach ChatGPT, with or without a feed, in How do shoppers use ChatGPT to find products?
A 6-part template for a product description
Use this order. Each part answers a question shoppers ask assistants.
- What it is, for whom, in one sentence. "Waterproof trail running shoe for runners with wide feet." This is the line an assistant is most likely to lift.
- The defining specs. Material, size range, weight, dimensions, capacity. Use numbers and units: "290 g (EU 42)", not "lightweight".
- What it's good and not good for. "Built for muddy and rocky trails; not designed for road running." Honest limits help an assistant match the right shopper.
- Compatibility and fit. Sizing advice ("runs half a size small"), what it works with, what it doesn't.
- What's in the box. Especially for bundles, kits and electronics.
- Care and materials detail. Washing instructions, origin of materials, certifications you can back up.
Keep shipping costs, return terms and prices out of the description itself. Show them on the page as plain text and in their own fields. We covered how in How do you show your shipping costs and return policy to Google and AI shopping assistants?
Before and after: an example
This is an illustrative product, not a real listing.
Before
NEW!! The ULTIMATE trail shoe 🔥 Our best-selling shoe is back and better than ever. Perfect for every adventure. Free shipping this week only!
After
Waterproof trail running shoe for runners with wide feet. Weight: 290 g (EU 42). Upper: recycled polyester mesh with a waterproof membrane. Outsole: rubber with 5 mm lugs for mud and loose rock. Drop: 6 mm. Sizes EU 38–47, wide fit only; runs half a size small. Not designed for road running. In the box: one pair of shoes and a spare pair of laces. Care: hand wash, air dry.
The "before" version breaks four of Google's rules at once: capitals for emphasis, symbols, a time-limited offer and shipping information. It also answers none of the questions a shopper would ask. The "after" version can be matched to "waterproof", "wide feet", "lightweight", "muddy trails" and "true to size?".
Matching titles
The title does the first round of matching. Google's examples lead with the product name, then the details that define it. For variants, add the detail that tells them apart:
| Weak title | Better title |
|---|---|
| Trail Shoe PRO SALE | Northline waterproof trail running shoe, wide fit, black, EU 42 |
| Mug | Blue ceramic mug, 350 ml, dishwasher safe |
| Dress - New Collection | Sleeveless polka dot skater dress, black, size M |
OpenAI's own feed examples follow the same shape: "Trail running shoes — black, size 10" and "Blue ceramic mug, 350 mL".
Mistakes that get products skipped
| Mistake | Why it hurts | Fix |
|---|---|---|
| Key facts only in images | Text inside an image may be missed; Google's AI guidance asks for important content in textual form | Repeat sizes, specs and ingredients as text |
| Same description on every variant | The assistant can't tell the black size 42 from the green size 44 | Give each variant its own title and attributes |
| Marketing claims with no facts | "Best quality" matches no question | Replace each claim with the fact behind it |
| Copy from the manufacturer, unchanged | Every store selling the product says the same thing | Add your own fit notes, use cases and photos |
| Page, schema and feed disagree | Google says structured data should match the visible text; mismatches lead to wrong answers | Update all three from one source |
| Policies inside the description | Google says not to include them | Use a policy page plus shipping and returns fields |
Should you use AI to write descriptions?
You can, but check every fact against the product itself. A model can invent a material or a weight that sounds right. If you use AI-written copy in Google Merchant Center, Google requires it in the structured_description attribute with the digital source type set to trained_algorithmic_media. The Merchant Center description guidelines show the format.
Where to start this week
- Pick your 10 best sellers. Rewrite their titles and descriptions with the template above.
- Write down the questions customers ask you in emails, chats and reviews. Make sure each answer appears on the product page.
- Check your schema and feed match the page. Price, availability, brand, size and colour should be identical everywhere.
- Ask an assistant your customers' questions. Type a detailed brief into ChatGPT or Google AI Mode and see which stores come up, and what they say about your products.
- Repeat for the next 10.
What this won't do
Better descriptions don't guarantee a recommendation. OpenAI says not every relevant product will be shown, and it ranks stores on factors such as availability, price, quality and whether you're the maker or main seller. Reviews on other sites also feed ChatGPT's review summaries. What good descriptions do is remove one common reason to skip you: the assistant couldn't tell your product fits.
Check your site
The free Web2Agents AI check scores your homepage on the basics AI tools look for, with results on screen. The €19 AI-Ready Kit runs all 36 checks on up to 25 pages, including product pages, and ranks the fixes.
FAQ
How long should a product description be for AI shopping assistants?
Long enough to answer the questions shoppers ask. Google allows up to 5,000 characters and advises putting the most important details in the first 160–500. OpenAI also suggests aiming for at most 5,000 characters in its product feed.
Should I put the price and shipping in the description?
No. Google's Merchant Center rules say not to include price, sale dates, shipping or delivery dates in the description. Show them on the page as plain text and send them in their own feed or schema fields.
Does ChatGPT use my product description word for word?
Not always. OpenAI says ChatGPT may generate simplified product titles and descriptions from the information it receives, because merchants describe the same product in different ways. Clear facts give it the right material to work with.
Can I use AI to write my product descriptions?
Yes, if you check every fact. In Google Merchant Center, descriptions created with generative AI must be submitted in the structured_description attribute, marked as trained_algorithmic_media.
Do I need special markup for AI Overviews or AI Mode?
No. Google says there are no extra requirements or special schema for its AI features. Product structured data still helps Google understand your pages, and it should match the visible text.
Sources
- Shopping with ChatGPT Search, OpenAI Help Center
- Products feed reference, OpenAI Developers
- Description and structured description, Google Merchant Center Help
- Title and structured title, Google Merchant Center Help
- AI features and your website, Google Search Central
- Introduction to Product structured data, Google Search Central
