| The three assistants do not share a product source, and that is the part most guides skip. ChatGPT reads a product feed you submit to OpenAI, or your Shopify catalog. Gemini and Google AI Mode read Google Merchant Center, which feeds the Shopping Graph. Alexa for Shopping reads your Amazon listing and quotes it back to the shopper. Doing the work for one of them does almost nothing for the other two. |
The mistake almost every guide makes
Search for how to get recommended by AI and you will find a dozen posts telling you to write clearer product descriptions, add schema markup and answer buyer questions. All of that is good advice. None of it tells you where each assistant actually gets its products.
That matters because the three biggest ones work in three different ways, and the work is not interchangeable.
The discipline has a name now. Generative engine optimization, usually shortened to GEO, and answer engine optimization, or AEO. Both describe the same job: making a product easy for a machine to describe accurately to somebody who asked a question. But how to optimize for generative engine optimization is not one task. It is three, and they run on three separate rails.
| Assistant | Where the products come from | What you control |
|---|---|---|
| ChatGPT | A product feed submitted to OpenAI, or your Shopify catalog if you sell on Shopify, plus other structured third party data | The feed. Accuracy of price and availability, and whether you are listed as the maker or the seller |
| Gemini and Google AI Mode | Google Merchant Center, which populates the Shopping Graph, plus structured data on your own product pages | The Merchant Center feed, your GTINs and product identifiers, and your schema markup |
| Alexa for Shopping | Your Amazon listing itself. It quotes your copy back rather than summarising it | Your title, bullets, attribute fields, A plus content and Q and A |
So a brand that spends a month rewriting Amazon bullets has done nothing for ChatGPT. A brand that submits a perfect feed to OpenAI has done nothing for Alexa. Both think they have covered AI.
Engine one. ChatGPT reads a feed, not your website
OpenAI states in its own help centre that ChatGPT considers structured metadata from first party and third party providers, alongside price, reviews and availability. There are two ways to get your catalog into that structured layer.
• If you sell on Shopify, your catalog is already integrated through Shopify Catalog. Nothing to submit, but everything depends on your product data being complete
• If you do not, you apply for direct product feed access and submit a feed by SFTP, file upload or a hosted URL
OpenAI’s commerce documentation is explicit that feeds exist to keep pricing and availability fresh. A feed that goes stale is worse than no feed, because it puts a wrong price in front of a buyer who is ready to act.
How to get products on ChatGPT Shopping, in order
Four steps, and the order matters because each one depends on the last.
• Get your catalog complete first. A feed only carries what your product data already holds, so a feed built on blank fields ships blanks faster
• Confirm which route applies to you. Shopify merchants are already integrated. Everyone else applies for direct feed access
• Submit the feed by SFTP, file upload or a hosted URL, and set it to refresh on the same schedule your prices change
• Check your work by asking ChatGPT the buying question a customer would ask, then look for your own sentences in the answer
Two things OpenAI says that change how you should plan
| 1. You cannot buy your way in. OpenAI states plainly that product results are selected independently by ChatGPT, are not ads, and are not influenced by any OpenAI partnerships. There is no paid placement to buy. Anyone selling you one is selling you something else. 2. Being the maker or the primary seller counts. OpenAI says merchant ranking considers availability, price, quality, and whether they are the maker or primary seller. If you own the brand, say so clearly and consistently in your data. If you resell someone else’s brand, expect to sit behind the people who make it. |
Engine two. Gemini and AI Mode read Google Merchant Center
Google’s product answers are built on the Shopping Graph, reported to hold more than 50 billion product listings and interpreted by Gemini models. The way in is Google Merchant Center, and no special account type is needed. A standard Merchant Center account makes you eligible.
What actually decides whether your product is usable to it:
• Product identifiers. GTINs or MPNs. Without them your product is hard to match against everything else in the graph, and matching is the entire mechanism
• Correct Google product category, not the closest guess
• Schema markup on your own product pages, because Google reads those directly as well as the feed
• Titles that say what the product is and what makes it different, rather than a keyword pile
If you sell only on marketplaces and have no site of your own, this engine is largely closed to you. That is worth knowing rather than discovering.
Engine three. Alexa reads your Amazon listing and quotes it
Alexa for Shopping is the odd one out, and the easiest to influence, because there is no feed to submit. It reads the listing you already have. The important mechanic is that it quotes your copy rather than summarising it. The sentence you wrote is the sentence the buyer hears.
That makes empty attribute fields a missing answer rather than an untidy record. We wrote up the seven changes we made across our accounts in our Alexa for Shopping post, so this section stays short.
One thing worth adding here. Being the Featured Offer still decides whose copy gets read out, which is why Featured Offer eligibility quietly became an AI visibility problem and not only a Buy Box one.
The one thing all three actually reward
Three different sources, one shared preference: structured completeness. Every one of these systems is trying to answer a question, and every one of them prefers a product it can describe confidently over a product it has to guess about.
Which means the highest value work is almost never the clever sentence. It is the field you left blank.
There is a useful test hidden in that. To get cited by ChatGPT, or by any of them, your product has to give the model a sentence worth repeating. A model will happily describe around a vague listing without ever naming it. It will only quote a line that answers something. That is why listing optimization stopped being a copywriting exercise and became a data one.
Before and after: three rewrites
These are illustrative examples rather than client listings, but the pattern is the one we apply. Each one takes a line that describes and turns it into a line that answers.
1. A title
| Before | After |
|---|---|
| PREMIUM STAINLESS STEEL WATER BOTTLE 32oz Insulated Double Wall Vacuum BPA Free Leakproof Sports Gym Travel | Insulated 32oz Water Bottle, Keeps Drinks Cold 24 Hours, Fits Standard Car Cup Holders |
The first is a keyword pile. The second answers the two questions that actually decide the purchase, how long it stays cold and whether it fits the car. Every engine above can use the second. None of them can do much with the first.
2. A bullet
| Before | After |
|---|---|
| DURABLE CONSTRUCTION: Made with premium high grade materials and reinforced stitching for maximum durability and long lasting performance | Rated for 50lb loads and tested to 10,000 open and close cycles, which is about three years of daily commuting |
The first could describe any product ever made. The second contains two numbers an assistant can quote and a buyer can check.
3. The part nobody rewrites
| Field | Before | After |
|---|---|---|
| Material | blank | 18/8 stainless steel, BPA free liner |
| Capacity | blank | 32 fl oz, 946 ml |
| Compatibility | blank | Fits cup holders 2.9 inches and wider |
| Care | blank | Dishwasher safe, top rack |
| Warranty | blank | 2 years, replacement not repair |
This is the least glamorous work on this page and it is the highest return. Those five fields answer five separate buyer questions, they feed the Shopping Graph match, they populate a ChatGPT feed, and Alexa can read them out loud. One afternoon of catalog work does more than a month of copywriting.
If you sell on marketplaces and on your own site
Most of the brands we work with do both, and the two jobs do not overlap as much as people assume.
| Where you sell | What to do first | What will not help |
|---|---|---|
| Amazon only | Attribute fields, then bullets rewritten as answers. Alexa is the engine you can reach | A product feed. You cannot submit someone else’s marketplace listing to OpenAI |
| Own site only | Google Merchant Center with GTINs, then schema on product pages, then a feed to OpenAI | Amazon listing work, since you do not have any |
| Both | Do the catalog data once, properly, then push it to both. The attribute values are the same values | Treating them as two separate projects with two separate owners, which is how they drift apart |
That last row is the one that costs money. We run 60+ marketplaces across our client base and the most common failure we see is not bad data, it is two versions of the same data. If your Amazon listing says one capacity and your website feed says another, you have not given three assistants a product. You have given them a disagreement to resolve, and they resolve it by recommending somebody else. Keeping those in step is a core part of managing a catalog across marketplaces.
What to do this week
Pick one product. Your best seller. Then do this, in this order.
• Ask ChatGPT, Gemini and the Amazon app the same buying question a customer would ask before paying. Write down which of your sentences each one quotes back, and which one does not mention you at all
• Open that product in your catalog and count the blank attribute fields. Fill every one that is true
• Rewrite the first bullet as the answer to the question you just asked, with one number in it
• If you have your own site, check whether that product has a GTIN in Merchant Center. If it does not, nothing else on this list matters for Google
That is an afternoon. Do it on one product, see which engine changes, then decide where the rest of the catalog effort goes.
See Where Your Products Are Actually Visible
If you want us to run that test across your top products and tell you which of the three engines you are actually visible in, that is what our free Amazon analysis covers. We look at catalog completeness, listing structure and feed readiness, and tell you which one is costing you the most. No commitment attached.
Frequently Asked Questions
No. A product feed is submitted by the merchant who controls the catalog and the checkout. For products you sell through Amazon, the route to ChatGPT and to Alexa is the quality of the Amazon listing itself. If you also sell the same products on your own site, that storefront can carry a feed.
For Gemini and Google AI Mode, yes, and if you have no site of your own then that engine is largely closed to you. It is one of the strongest arguments for running a direct storefront alongside the marketplaces rather than instead of them.
No. OpenAI states that product results are selected independently, are not ads, and are not influenced by OpenAI partnerships. There is no placement to buy, which is unusual and worth appreciating while it lasts.
No, it sits on top of them. All three engines lean on the same underlying assets: accurate structured product data, clear answers to buyer questions, and being findable in the first place. Work that improves AI visibility almost always improves conventional rankings too.
Size is not the gate, data quality is. OpenAI says ranking considers availability, price, quality and whether you are the maker or the primary seller. A small brand that makes its own product, keeps stock accurate and describes it completely is in a better position than a large reseller with a thin catalog record.
Two checks. Ask the assistants the same buying questions monthly and record whether your sentences come back. And watch referral traffic from ChatGPT, Perplexity and Gemini in your analytics, which is small for most brands today and is the fastest growing line on our own report.
