AI search, agents and your catalog
Shoppers now ask ChatGPT, Gemini and a wave of personal agents what to buy, and increasingly let them buy it. What those assistants recommend depends on what they can read about you.
This page explains how AI search and shopping actually work, why your product catalog has become your storefront, and what your options are for getting it in front of every assistant.
The shopper does the comparing, across tabs. Your website does the persuading. You win by ranking.
The AI does the comparing before anyone reaches your site. Your data and your reputation do the persuading. You win by being in the answer.
Nobody visits your homepage. The agent reads your catalog and acts through it. You win by being easy to buy from.
You'll hear this called AEO (answer engine optimization) or GEO (generative engine optimization). Same idea: make sure AI can find you, trust you and describe you correctly.
"linen duvet"
"A breathable linen duvet for a hot sleeper, under $300, that arrives by Friday and is easy to return."
The AI breaks that sentence into checks. Each one needs a fact it can find and trust:
linen duvet
AI has to confirmProduct type and material
Where it looksProduct attributes in your feed and page
for a hot sleeper
AI has to confirmThe product actually solves this
Where it looksYour description, then reviews and forums that agree
under $300
AI has to confirmCurrent price, for this size
Where it looksPrice per variant, kept current
arrives by Friday
AI has to confirmIn stock and ships in time
Where it looksLive inventory and shipping data
easy to return
AI has to confirmThe returns policy
Where it looksPolicy as readable text or data, not hidden in a pop-up
| The shopper said | AI has to confirm | Where it looks |
|---|---|---|
| linen duvet | Product type and material | Product attributes in your feed and page |
| for a hot sleeper | The product actually solves this | Your description, then reviews and forums that agree |
| under $300 | Current price, for this size | Price per variant, kept current |
| arrives by Friday | In stock and ships in time | Live inventory and shipping data |
| easy to return | The returns policy | Policy as readable text or data, not hidden in a pop-up |
If any one fact is missing, AI can't confirm you qualify, so it recommends someone it can.
01
Crawlers read the page source and stop. Most don't run JavaScript, so anything that loads later, like prices, reviews or policy text, may be invisible.
02
A structured file of every product, variant, price, stock level and policy, sent directly or through your platform. The cleanest signal you control.
03
Reviews, Reddit, press, creators and marketplaces. This is where AI decides whether to believe your claims, and where it learns things you didn't say.
04
At the moment of decision, better assistants confirm price and stock. Stale data means a wrong answer, or a sale that falls apart at checkout.
When these four agree, AI trusts you. When they disagree, say your site says a 100-night trial and an old forum post says 30 days, AI will just pick one, and it may be inaccurate.
The biggest door. Shopping now focuses on discovery and comparison, with most purchases completing on your own site.
Gets products from: OpenAI's product feed, platforms that syndicate to it, and crawled pages
Answers above the links, built on Google's product database. Checkout inside Google is live for eligible merchants through an open commerce standard.
Gets products from: Google Merchant Center feeds and crawled pages
Built into Windows, Edge and Bing, so it reaches shoppers who never open a chatbot. Some merchants can sell inside Copilot directly.
Gets products from: Microsoft Merchant Center (can reuse your Google feed) and platform syndication
Smaller, with deep researchers. Every answer shows its sources, so being a cited source with accurate details is the whole game.
Gets products from: its free merchant program, partner feeds and crawled pages
Shopping answers draw on creator posts and brand content across Instagram, Facebook and Threads. The Muse agent can browse, fill in forms and check out for the user.
Gets products from: your Meta catalog, social content, and your website
Instinct, Muse and others that people simply text: "find it, book it, buy it". They use your site the way a person would, only faster and with less patience.
Gets products from: your website directly. Guest checkout and clear policies matter.
Requirements for these channels: be readable without a browser, be described accurately, and be easy for a machine to act on.
In AI search, homepage is not doing the selling anymore. Your product data does. Most brands under-invested in the product catalog.
One identity per product. The same item described the same way on your site, in every feed and on every marketplace.
Fully describe every variant. Size, colour, material, fit and use case as fields, not buried in a paragraph.
Live price and stock. Updated in minutes or hours, not overnight, especially around promotions.
Structured data for shipping and returns policies. Delivery windows and policies that a machine can read and quote.
Social proof. Ratings, review counts and real images that travel with the product wherever it goes.
There are four ways to get your products in front of AI assistants, and most brands end up using two. The right mix depends on your platform and who owns your product data.
Effort: low · Best when: you want the fastest start
Several commerce platforms now send your catalog to AI assistants for you, with a switch per channel. Coverage, checkout and controls vary a lot between platforms, so check exactly which assistants yours reaches.
Watch: it sends what you have. Gaps in your product data go out to every channel at once.
Effort: medium · Best when: you sell across many channels, AI plus marketplaces and ads
One source, many destinations, with rules per channel. Managed services suit lean teams; self-serve tools cost less but need an owner.
Watch: cost grows with catalog size, and you've added another layer between you and your product data.
Effort: low to medium · Best when: your platform has no AI channels and you don't want to migrate
Some platforms and startups now run shared product catalogs that AI assistants draw from, open to brands that don't run their store with them.
Watch: another company holds a copy of your product data. Check how often it syncs, who sees your customer data, and how orders get back to you.
Effort: high · Best when: a few channels drive most of your sales and you want full control
Google and Microsoft Merchant Center, OpenAI's product feed, Perplexity's merchant program and your Meta catalog each have their own setup.
Whichever route you choose
Same price, same stock status, same return policy. AI often checks one against the other. If your feed says $289 and in stock but the page shows $319, or no price at all, AI can't tell which is right and may leave you out. For your developers: add product markup (schema.org) to every product page.
It still feeds Google's AI answers, and disapproved items there quietly cost you visibility.
Start with the connection you already have and put the effort into the product data itself: complete descriptions, every variant, correct prices and policies. Good data improves every channel at once; messy data shows up wrong everywhere. And give one person ownership of it. In most companies it's split across merchandising, marketing and engineering.
These are the free checks I recommend. Each takes a few minutes and gives you a useful first look. Not affiliated with any of them.
Scores how well AI agents can find, read and use your website, with evidence for each issue.
Run Is AgenticBuilt for commerce brands. Shows how visible you are across AI search and shopping assistants, with a few places to start.
Run Yotpo's reportA visibility score across ChatGPT, Gemini and Google's AI results, compared with other brands Semrush places beside you.
Run the checkerA note on reading them: each tool asks different questions and scores in its own way, so the same brand can look strong in one and weak in another. Use them to spot patterns rather than settle on a number. The harder questions, like why it's happening and which fixes are worth the effort, usually take someone looking at your site, catalog and competitors together.
Run and reviewed by Rohit
Coming soonEverything on this page, applied to your brand. Written for the leadership team, readable in two minutes, with the technical detail underneath for your team, then a 30-minute walkthrough with me.
Leave your details and you'll be first to know.
They overlap, and good SEO still helps. But AI answers often cite sources outside the top search results, read your page without running JavaScript, lean on reviews and forums, and increasingly pull from catalog feeds rather than pages. Strong rankings don't guarantee you're in the answer.
Less than you'd think. Every major platform has a route into AI channels; what matters more is the quality of your product data. The options section above lays out the trade-offs.
It's a business choice. Crawlers that power search and shopping answers send you customers; crawlers that collect training data don't directly. Many brands allow the first and decide separately on the second. The worst outcome is blocking both by accident, which is common.
No. Everything uses your public website and public AI assistants, the same things your customers see. No logins, no code, no data sharing.
The technical checks are repeatable. AI answers vary from one ask to the next, which is why each question is asked several times and reported as "named in X of Y answers". It's a dated snapshot, not a tracking service.
Based on public sources as of October 2026: OpenAI commerce documentation and product updates; Google announcements on AI Mode and the Universal Commerce Protocol; Shopify announcements on Agentic Storefronts and the Agentic plan; Meta announcements on Muse Spark and Muse; vendor documentation from Microsoft, Perplexity and feed platforms; reporting by TechCrunch and Reuters. Not affiliated with any platform named.