AI shopping assistants answer product questions using structured data and text they can interpret. An AI-readable feed is complete, consistent and verifiable: clear attributes, current prices, visible policies and markup faithful to the page.
The essentials in a definition
An AI-readable feed is a complete, consistent and verifiable set of product data that automated systems can interpret and cite. It builds on the Merchant Center specification, product structured data and the Schema.org vocabulary.
What is machine-readable
A machine reads identifiers, attributes and values better than adjectives. “Waterproof jacket, size M, recycled polyester, 129 euros, in stock” informs more than “the perfect jacket for every adventure”. The more factual and consistent the description, the easier it is to cite correctly.
The Merchant Center specification defines attributes and formats. Meeting it is the starting point.
Attributes that make a difference
Beyond the basics (title, price, link, image), fill in brand, identifiers such as GTIN, colour, size, material, condition and dimensions. Include shipping and returns information, and keep availability updated. Category-specific attributes help distinguish similar products.
Avoid promotional text in fields that should be descriptive, and do not fill attributes with data you cannot confirm.
From feed to citable answer
- Meet the feed specification
- Complete category-specific attributes
- Align markup, feed and page
- Test answers and fix the source
Markup and consistency
Use product structured data and the Schema.org vocabulary to mark up name, image, price, availability and genuine reviews, faithfully to the visible content. Check that price and stock match across feed, markup and page.
Inconsistencies are among the most common causes of disapprovals and wrong descriptions.
Factual, consistent data gets cited; adjectives do not.
Checking the result
Ask assistants about your products and compare answers with the catalogue. When there are errors, look for the source: a missing attribute, ambiguous text, an outdated page. Fix at the source and repeat the test after some time.
Record what changes to understand the effect, knowing the result is neither guaranteed nor controllable.
Decision table
| Field | Readable example | Avoid |
|---|---|---|
| Title | Brand, type, key attribute | Promotional phrases |
| Description | Facts and verifiable differences | Vague adjectives |
| Identifiers | GTIN, brand, reference | Empty fields |
| Price and stock | Current values equal to the page | Mismatches |
Common mistakes to avoid
- Titles with excess promotion. They harm interpretation and may cause disapproval.
- Markup that does not match the page. It breaches guidelines and breeds distrust.
- Not updating frequently. Outdated prices and stock are cited as current.
How to measure the result
Track the rate of approved products, filled attributes, price mismatches and the accuracy of descriptions in answer samples.
Limits and exceptions
Assistants use several sources and methods you do not control, and correct citation is never guaranteed. Specifications and formats evolve; confirm official documentation. Incorrect markup can have consequences, so always validate before publishing.
Illustrative example
Illustrative example: an appliance store rewrites titles as “brand, type, capacity, energy class”, fills in GTIN and dimensions and aligns price and stock in near real time. Products stop being disapproved for inconsistencies and, when asking assistants, the characteristics cited start to match the catalogue.
In practice: checkpoints
Run a monthly audit of ten key products: compare feed, page, markup and what two or three assistants say. Log differences, fix the source and retest. Keep changes small and documented, since the aim is accuracy and not tricks.
Frequently asked questions
Does a good feed serve ads and assistants?
Largely yes: completeness and consistency benefit both.
What errors are most common?
Missing attributes, misaligned prices and promotional descriptions in descriptive fields.
Should I mark up reviews?
Only genuine reviews, in line with applicable guidelines.
How often should I update the feed?
As often as price and stock change; daily at minimum for dynamic catalogues.
Do I need a developer?
Many commerce platforms export well; for complex cases, technical support can help.
An AI-readable feed is factual, complete and consistent with the page, and is verified by asking assistants and fixing what is wrong at the source.
Sources and recommended reading
- Merchant Center: product data specification
- Google Search Central: product structured data
- Schema.org: Product
Last editorial review: October 2026. Informational content; always confirm current conditions in the official sources before deciding.