Make your site readable to AI shopping agents

November 9, 2026 · Clever Bean

DRAFTED OVERNIGHT. APPROVED AND PUBLISHED BY DARIA BYKOVA. November 9, 2026

Shopify's AI shopping visits rose 2.6 times year over year and that product catalog searches convert at twice the rate of general AI queries. (Nov 3, 2026 earnings call, reported by pymnts).

An AI shopping agent reads text, links and structured data, not layout. To be usable by one, give every product its own page with a plain name, price, stock status, variants and policies, mark them up with schema.org Product data, keep menus as real links and make checkout work without hidden steps.

What is an AI shopping agent and how does it read a website?

An AI shopping agent is software that takes a buyer's request, such as "a silk scarf under a set budget that ships to Switzerland", then searches, compares products and sometimes completes the purchase. It acts on the buyer's behalf.

It reads your site as text and structure. It sees page titles, headings, link labels, the words in your product description, any structured data in the page code and the fields in your forms. It does not see your photography the way a person does, and a banner that says "new" in an image says nothing to it.

That gives you a simple working rule. If a fact matters to a buyer, it must exist as words or data on a page with its own address.

How do AI agents navigate a site's menus and category structure?

Agents follow links. A menu built from real links with plain labels is easy to follow. A menu that opens only on hover, loads after a script runs or hides behind an icon with no text label is easy to miss.

Three habits help:

  • Name categories the way buyers search. "Silk scarves" works. "Collections" tells an agent nothing.
  • Keep the path short. Home, category, product should be reachable through ordinary links, with each level on its own address.
  • Link products from their category pages as text, not only through an image carousel or a filter that needs a click to load results.

If you sell both products and services, give each a clear category so an agent can tell a thing you ship from a session you book.

What product data do AI shopping agents need?

An agent comparing five products needs the same fields for each one. Missing fields push your product down the list or out of it, because the agent cannot compare what it cannot read.

The core set is name, a short factual description, price and currency, stock status, variants such as size or colour with their own availability, shipping regions and costs, return terms and a clear product identifier where one exists. Show every one of them in visible text and in structured data, and make sure the two match. A visible price that differs from the marked-up price is a reason for an agent to distrust both.

Product page data checklist

0 of 7 done

  • Details

    Use the name a buyer would type, with material, type or size where it matters.

  • Details

    Both must show the same number. Update both together when the price changes.

  • Details

    A page that says "in stock" for every size when one is gone teaches agents to stop trusting you.

  • Details

    Sizes, colours and options should appear as readable choices, not only as swatches.

  • Details

    Put them on the product page or link to one plain policy page.

  • Details

    State the window and who pays for return shipping.

  • Details

    Barcode or SKU helps software match your product to the same item elsewhere.

Every product page should state name, price, currency, stock per variant, variant options, shipping and return terms as visible text, and repeat the same facts in structured data. Add a barcode or SKU where you have one. The main failure is mismatch: marked-up facts that differ from what the page shows.

How do I write product descriptions that AI agents can parse and compare?

Write attributes as facts, then add the story. An agent comparing two scarves wants size, material, weight, origin, care and use. It cannot compare "luxuriously soft with an unforgettable feel".

A structure that works:

  1. One sentence saying what the product is and who it suits.
  2. A short list of attributes with units: dimensions, material, weight, country of origin, care.
  3. One paragraph on what makes it different, with something checkable, such as how it is made.
  4. The practical answers buyers ask: sizing, delivery, returns, what is in the box.

Use the same attribute names across products so comparisons line up. If one page says "material" and another says "fabric", an agent may treat them as different things.

Skip adjectives that carry no fact. If a sentence survives without the word, delete the word.

Should I use schema.org Product markup for AI shopping?

Yes. Schema.org Product markup with an Offer is a shared, documented way to tell software your product name, price, currency, availability and identifiers. Search engines have used it for years, and it gives any agent a clean version of the facts on your page.

Two cautions. First, markup helps only when it is accurate and matches the visible page. Second, I cannot show you that any specific assistant ranks products higher because of it, and nobody outside those companies can. It removes guesswork and costs little, which is reason enough.

Most store platforms generate basic Product markup. Check what yours outputs rather than assuming. Look for price, currency, availability per variant, brand, image and an identifier. Gaps usually sit in variants and stock.

What changed on 2026-11-05, and what do we think?

The martech trade press keeps a running list of AI releases, and its 2026-11-05 edition is a reminder of how fast the tooling around agents moves. Our opinion, written on 2026-11-08, stays the same whichever tool gets announced next.

Do not rebuild your site around one vendor's agent. Each announcement changes who gets to act for the buyer. None of them changes what the agent needs from you, which is facts it can read, a path it can follow and a checkout it can complete. Brands that have those three in order can use whatever arrives. Brands that chase each release end up with a site tuned to last month.

Spend the effort on your product pages and data first. They pay off for Google, for AI answers and for human buyers at once.

How do I make checkout and add-to-cart usable without a visual interface?

An agent completing a purchase needs the steps to be predictable. Every step that depends on a pop-up, a drag gesture or an image puzzle is a place where it stops.

What to check:

  • The add-to-cart control is a real button with a text label, and variants are chosen through standard form fields.
  • Guest checkout exists. A forced account creation is the most common wall.
  • Shipping cost and delivery time show before the final step, not after the card number.
  • Error messages are written in text and say which field needs fixing.
  • Verification steps such as bot challenges are used only where you have a real fraud problem, since they stop legitimate agents too.
  • Payment options follow common, documented methods rather than a custom flow.

If a human using only a keyboard and a screen reader can buy from you, an agent probably can. That is a useful rough test, and it is also good for your customers.

Product feed, schema markup and llms.txt: what each does

Where it lives
Inside each product page
What it carries
Name, price, availability, identifiers for that page
Who reads it
Search engines and any software that parses page code
Best used for
Every product page, always

Schema markup sits on every product page and is the baseline. A product feed is for platforms that ask for your catalogue in their format. An llms.txt file is an optional site guide that only some software looks for. Do the first, add the second where a platform requires it and treat the third as extra.

Do I need an llms.txt file or a product feed?

A product feed, yes, if you sell on a platform that takes one. It is how that platform gets accurate price and stock without crawling every page. Keep it in sync with the site, because a feed that disagrees with the page is worse than none.

An llms.txt file is a proposed convention: a plain text file listing your most useful pages with short descriptions. It is cheap to make, and it does no harm. I cannot show you that the major assistants read it, so it comes after accurate product pages and markup, never instead of them.

If you add one, list your categories, your best product pages, your shipping and return policy and your contact page. Keep descriptions factual.

How do I test whether AI shopping agents can find and buy my products?

Test in three layers and write down what you see each time.

  1. Page layer. View your product page with styling off, or copy the text. Does it still show price, stock, variants and policies? Check your markup with a structured data testing tool.
  2. Question layer. Ask several assistants the questions your buyers ask, such as "best [product type] for [use] under [budget]". Note whether you appear, what they say and whether the facts are right.
  3. Purchase layer. Buy something from your own store using only a keyboard, then try an assistant that offers to shop for you if you have access. Note every point where it stalls.

Nobody can read live ChatGPT queries, so you cannot see what real buyers asked. Repeat the same test each month and track changes. Results take time, and a page you fix today may take weeks or months to show up in answers.

Could an AI agent buy from your site today?

Question 1 of 5

Does each product have its own page with a price in text?

Score yourself on five points: product pages with text prices, matching markup, guest checkout, shipping and returns stated early, and plain text menu links. Zero to two means you are mostly readable. Three to six means gaps an agent will hit. Seven or more means an agent will likely stall. Start with one product page either way.

What is an AI shopping agent?

Software that takes a buyer's request, searches and compares products and sometimes completes the purchase on the buyer's behalf. It reads your site as text, links and structured data.

Do I need to redesign my site for AI agents?

Usually not. Most gains come from clearer product pages, accurate markup and a checkout without hidden steps, which you can change on your current site.

Does schema.org markup guarantee AI recommendations?

No. It gives software clean facts about your product. Nobody can guarantee recommendations, and results take time.

Can I see what shoppers ask ChatGPT about my products?

No. Nobody outside the AI companies can read live queries. You can ask assistants the questions your buyers likely ask and record the answers.

Is an llms.txt file required?

No. It is an optional convention. Accurate product pages and markup matter more.

  1. The latest AI-powered martech news and releases 2026-11
  2. Agentic Commerce Drives Shopify Revenue Up 32% 2026-11

Agents reward the same things careful customers do: clear names, honest facts, a visible price and a checkout that works. Fix the product page first, then the menu, then checkout and test after each change. Results take time and nobody can read live ChatGPT queries, so measure what you can see. To find out how your business appears in Google and five AI assistants today, run the free check at /check.