If you sell on Amazon, Booking or Expedia, the AI built into those platforms doesn't go and read your website to answer customers: it uses the internal product listings, and only those. A poorly written or incomplete listing means the AI recommends the competitor instead of you — inside the very platform where the customer is already buying. Optimizing those listings specifically for AI answers is one of the few actions with a direct, measurable impact on sales within a matter of weeks.
Activate Microsoft 365 Copilot, open Word, type “suggestions for a weekend in Capri”. See who it recommends? Microsoft Copilot takes into account your geographic location and the document context — different from ChatGPT, which draws from its training index and the canonical web sources.
This is the point: every AI platform has a privileged source from which it extracts its answers. And when you enter a marketplace — Amazon, Booking, Expedia, Trivago, eBay — the privileged source becomes one alone: the product listings internal to the marketplace itself. Not your website, not your blog, not your backlinks. The listings.
If you run a 5-star boutique hotel in Capri, your Booking listing is already today the chunk that feeds Booking AI when a user asks “best luxury hotel in Capri with a view of the Faraglioni”. The same goes for those selling extra virgin olive oil on Amazon, wine on Tannico, handcrafted footwear on Farfetch.
In this series I’m taking apart the AI platforms one by one. Here we talk about a category of its own: the AI inside marketplaces.
Why the marketplace is a closed universe
When ChatGPT answers “best hotel in Capri”, it draws from open web sources: TripAdvisor, reviews, travel magazine articles, the hotel’s official website. It’s the citation model I described in the articles on E-E-A-T for AI and on the backlink as a citation proxy.
When instead Booking AI answers the same question inside the Booking app, the universe is closed: the AI can only cite properties present on Booking. Your listing is the only source. Same mechanism for Rufus on Amazon: if you sell Apulian extra virgin olive oil, Rufus answers “best oil for dressing raw dishes” by drawing from the Amazon listings, not from your company website.
There is no academic paper certifying this behavior — it’s a product mechanism, not published research. From this follows a consequence that Italian entrepreneurs underestimate: your SEO/GEO asset on the marketplace is the listing, not generic presence. And the listing must be treated as structured content for AI, not as a commercial showcase.
What changes for your business
If you have a boutique hotel in Capri, until now you’ve optimized the Booking listing with internal ranking in mind (photos, price, reviews, conversion rate). The next level is to write it thinking that every field is a citable chunk for the marketplace’s conversational AI.
Three operational changes:
- The long description is not marketing copy, it’s a mini-FAQ that answers natural questions like “hotel in Capri with infinity pool and sea view”, “Anacapri boutique hotel with Michelin-starred cuisine”, “romantic stay in Capri off-season”.
- The structured amenities (Booking fields: spa, sea view, restaurant, parking) are not search filters: they are entities the AI uses to match intent-based queries.
- Reviews with specific language (“breakfast served on the terrace with a view of the Faraglioni”) become material that Booking AI can cite verbatim.
The same reasoning applies beyond tourism: a Neapolitan pastry shop selling sfogliatelle on Amazon, a winery on Etna on Tannico, a dental practice in Salerno on MioDottore. The marketplace changes, the principle doesn’t.
Brochure-style short description: 200 characters of marketing, zero concrete specifics.
The test you can run yourself in 10 minutes
This is a basic, entry-level audit. A serious analysis of your listing requires professional tools and a longitudinal comparison of rankings.
- Open the marketplace where you sell (Booking, Amazon, Trivago, Tannico, Farfetch).
- If it has a built-in AI assistant (Booking AI, Amazon Rufus, Trivago AI), run 5 natural queries like “X in Y with feature Z” that relate to your industry.
- Note: is your listing cited? Which competitors are? What do they have in their descriptions that you don’t?
- Open 3 cited competitor listings and compare them with yours: description length, number of completed amenities, presence of FAQs, language in the reviews.
Binary threshold: if the marketplace AI never cites you on queries that describe exactly your offering, the listing needs rewriting. It’s not a problem of price or reputation, it’s a problem of machine readability.
Rewrite the long description of your main listing (Booking, Amazon, industry marketplace) as a 600-800 character mini-article that answers 4-5 natural questions typical of your segment.
The test I ran: Microsoft Copilot vs ChatGPT on 15 Capri queries
Indicative test, not a study. Small sample but a clear pattern.
I ran 15 “destination” queries on Capri — the typical questions of someone planning a stay: “5-star boutique hotel Capri with Faraglioni view”, “Michelin-starred restaurant Anacapri with sea view”, “romantic weekend Capri October”, “hotel Capri near Marina Piccola”, and so on. I passed them both to ChatGPT in search mode and to Microsoft Copilot (the copilot.microsoft.com chat with Bing as the retrieval engine).
Patterns that emerged:
- Across 15 queries, ChatGPT cited on average 4-5 properties per answer, mixing historic hotels, Condé Nast Traveler articles and travel-writer blogs. The citations converged on the well-known names (the 3-4 most-reviewed grand hotels).
- Microsoft Copilot cited on average 6-8 properties, including less prestigious ones but with a strong presence on Booking and Tripadvisor — the two indexes from which Bing draws heavily.
- On 4 queries out of 15, Copilot directly proposed Booking links in the answer. ChatGPT never did.
The limit of the test: 15 queries don’t make statistics, and AI results vary over time. But the product signal is clear: Microsoft Copilot, when the user is logged into the Microsoft ecosystem, favors sources on which Bing has deep indexing and commercial agreements. Booking is one of these.
For your boutique hotel in Capri this means two things: the Booking listing works twice (for the internal Booking ranking + for the Copilot citations), while the official website works mainly for ChatGPT, Perplexity and Google AI Overviews.
A non-trivial detail that emerged from the test: Microsoft Copilot took into account the geographic location declared in the profile and the context of the conversation in which I had mentioned an itinerary in Campania. It proposed Capri in a way consistent with the history of the exchange. ChatGPT in search mode answered more “neutrally”, like a general-purpose search engine. It’s a subtle difference but it changes the game for anyone selling local experiences: the user context becomes part of the query.
The mistakes I see most frequently on marketplace listings
- Brochure-style short description: 200 characters of marketing, zero concrete specifics. The AI has no citable material and cuts you out.
- Half-completed amenities: the optional fields left empty (pet policy, flexible check-in hours, detailed spa services) are missing entities. If a user searches “pet-friendly hotel Capri with spa” and you have both but haven’t ticked them, you’re invisible.
- Missing FAQs: many marketplaces have a questions & answers section (Amazon Q&A, Booking property surroundings). Filling it in is the equivalent of the inverted pyramid applied to content: the blunt answer first, the context after.
- Internal jargon language: “Junior Suite Deluxe Premium room”. The AI doesn’t know what it is. Better “35 sqm room with balcony and open sea view”.
These mistakes are the marketplace version of the same pattern I described for Named Entity Recognition: if you don’t make your features readable as entities, the system doesn’t recognize them.
What to do now in concrete terms?
- Rewrite the long description of your main listing (Booking, Amazon, industry marketplace) as a 600-800 character mini-article that answers 4-5 natural questions typical of your segment.
- Fill in 100% of the structured fields. Even the ones that seem marginal: each is a potential entity.
- Add 5-10 real FAQs (if the marketplace allows it) with natural, non-commercial language.
- Compare your listing with the 3-5 competitors that the marketplace AI cites most often on the queries in your industry.
- Review the copy every 3 months: marketplaces update their AI algorithms silently.
The marketplace as an AI platform in its own right
Marketplaces have become full-fledged AI platforms, with a closed universe of sources and their own citation algorithm. Your visibility in AI answers is no longer played out only on ChatGPT, Perplexity and Google: it’s also played out inside Booking, Amazon, Trivago, Tannico, MioDottore. Every marketplace is a vertical AI engine.
In the upcoming articles of this series I analyze Bing Copilot in depth (how the Microsoft 365 integration works and how to exploit it for B2B), Perplexity vs Google AI Overviews for commerce, and the vertical industry platforms. The thread is always the same: knowing which source feeds which AI is the difference between being cited and not being there.
