In 2025 I wrote two posts on AI for marketplaces. Rereading them now, the honest verdict is that they were lists of tools. Pricing tables for APIs that have since changed twice, a tour of vendors, a future vision. Useful for a month, then stale. What they did not do was answer the question a founder actually has, which is narrower and more practical: where in my marketplace does this technology move a number I care about, and which side of the market does it help?
This post is the rewrite. It skips the tool catalogue, because tools turn over faster than anyone can review them, and it organises AI by the job it does inside a marketplace: helping buyers find a match, helping suppliers create listings worth finding, keeping both sides safe, and getting your marketplace mentioned when a buyer asks an AI assistant for a recommendation instead of searching for one.
Start With The Side That Is Stuck
A marketplace has two or more sides, and at any given moment one of them is the constraint. Sometimes buyers arrive and cannot find what they want. Sometimes suppliers sign up and never finish a listing. Sometimes both sides match but nobody trusts the other enough to pay. The right AI project is the one aimed at whichever of those is binding right now.
This sounds obvious, and it is genuinely hard to hold onto, because the AI features that demo best are rarely the ones aimed at your constraint. A conversational shopping assistant looks impressive in a pitch. If your real problem is that sixty percent of supplier signups never publish a listing, the assistant is answering questions about inventory that is not there yet.
| If the stuck side is | The AI job | The number it should move |
|---|---|---|
| Buyers cannot find a match | Search and matching | Search-to-contact or search-to-book conversion |
| Suppliers do not finish listings | Listing creation | Signup-to-published-listing rate |
| Matches happen but nobody pays | Trust and safety | Contact-to-transaction rate, dispute rate |
| Transactions happen but go wrong | Support and disputes | Time to resolution, repeat rate after a dispute |
If you cannot yet say which row you are in, that is the first piece of work, and it is not an AI project. Our marketplace KPIs post covers the handful of numbers that tell you.
Search And Matching: The Demand Side
Classic marketplace search matches words. A buyer types “quiet place to work near downtown” and the search engine looks for listings containing those words. The listing that would suit them perfectly, described by its owner as “calm studio, five minutes from the city centre,” shares almost none of them and never appears.
Semantic search, which matches on meaning rather than exact words, is the single most practical AI upgrade for most marketplaces with more than a few hundred listings. It is also among the cheapest, because it is now a standard feature of the hosted search services many marketplaces already use. The work is less in the technology than in deciding what a good match means in your category, which only you know.
The way to judge it is your zero-result and low-result searches. Pull the last ninety days of searches that returned nothing, or returned results nobody clicked. If a meaningful share of them had suitable listings that the search simply missed, better matching will move conversion directly. If they genuinely had no suitable supply, the problem is supply, and no search upgrade fixes it. Our liquidity playbook is the better read in that case.
Listing Creation: The Supply Side
Listing creation is where AI earns its keep most reliably, and it is the use I would look at first on almost any marketplace where supply is the constraint. Writing a good listing is tedious. Suppliers are busy, many are not natural writers, and a long form with twelve fields is exactly where signups go to die.
Three patterns work well:
- Draft from photos. The supplier uploads pictures and the system proposes a title, category and description for them to edit. Editing a draft is far easier than facing a blank box.
- Fill the structured fields. Size, condition, materials, service area. Extracted from what the supplier already wrote, then confirmed with a tap. Structured fields are what make listings filterable, which feeds straight back into buyer search.
- Suggest, do not decide, a price. A range based on comparable listings helps a first-time supplier who has no idea where to start. The supplier keeps the final say.
One caution. Generated text tends toward the same polished, generic voice, and a marketplace full of listings that all sound alike loses the texture that makes buyers trust real people. Keep the supplier in the loop as editor, and keep their own words where they are specific. A draft is a starting point, not the listing.
Trust And Safety: Both Sides
Trust is the quiet constraint on most marketplaces. Buyer and supplier find each other, then hesitate, because they are strangers about to exchange money. Anything that makes a stranger safer to deal with lifts the contact-to-transaction rate on both sides at once.
AI helps here in three established ways. Identity verification services compare a photo of a government ID against a selfie, which is now a commodity you integrate rather than build. Moderation models screen new listings and messages for prohibited items, scams and abuse before a human sees them, so your small team reviews the flagged few instead of everything. And pattern detection spots behaviour that looks like fraud: a new account messaging fifty suppliers in an hour, or a buyer trying to move payment off the platform.
That last one matters commercially as well as for safety. Conversations that drift toward “just pay me directly” are the moment your marketplace loses its take rate. Catching them is a trust feature and a revenue feature at the same time.
Support And Disputes
Support on a marketplace is harder than support in a normal shop, because most tickets involve two parties who disagree. An assistant that answers the routine questions, such as how payouts work, how to cancel or where an order is, frees your team for the cases that need judgment.
Disputes are the cases that need judgment. AI can help assemble the facts: summarising the message history, the listing as it was at the time of booking, the photos both sides uploaded. The decision itself should stay with a person, at least until you have a long record of how disputes in your category resolve. A dispute handled badly costs you both sides of the transaction, and often their friends as well.
“The best AI feature on a marketplace is usually invisible. Buyers notice they found the right thing faster. Suppliers notice listing took five minutes instead of thirty. Nobody notices the scam that never reached them.”Darren Cody, Marketplace Studio
Showing Up When Buyers Ask An AI
The second half of this topic is outward-facing. More buyers now ask an AI assistant where to find something instead of typing a query into a search box and scanning ten links. If your marketplace is the answer to that question, you want to be the one named.
The reassuring part is that this is mostly ordinary search work. Google states in its Search Central documentation that to appear as a supporting link in its AI Overviews and AI Mode, a page needs to be indexed and eligible to show in Search with a snippet, and that there are no additional technical requirements. The fundamentals that win classic search are the same ones that get you cited.
For a marketplace, those fundamentals have a particular shape:
- Indexable category and location pages.Airbnb is the classic example: it has a crawlable page for stays in almost every city people search for. Your equivalent is a real page for each combination buyers actually look for, such as “wedding photographers in Calgary,” with listings and a short, genuinely useful introduction. Be selective. Indexing every filter combination creates thousands of thin pages that help nobody.
- Direct answers to real questions. How much does it cost to list? How are suppliers vetted? What happens if a booking is cancelled? Answer each plainly, in a sentence or two, on a page that is clearly about that question. Assistants quote direct answers. They struggle to quote marketing copy.
- Structured data that matches the page.Markup from schema.org labels prices, reviews, locations and FAQs so machines read them without guessing. Google's guidance is that structured data should match the visible text on the page, so mark up what is there, not what you wish were there.
- Let the right crawlers in. Check your robots.txt. OpenAI documents separate crawlers: OAI-SearchBot surfaces sites in ChatGPT search, while GPTBot collects content that may be used to train its models. They can be allowed or blocked independently, so you can stay visible in AI search answers without opting into model training.
The products here change quickly, and I would treat any specific tactic, including the ones above, as something to recheck every few months. The underlying principle has held for twenty years: a fast, well-structured page that clearly answers the question a buyer asked.
Build, Buy, Or Wait
For nearly every use in this post, buying beats building. Semantic search, verification, moderation and support assistants are all mature services you connect to your marketplace rather than invent. Building your own makes sense only when the capability is the product itself, and that is rare in the first two years.
Waiting is also a legitimate answer. If you have forty listings, AI matching has nothing to match. Hand-sourcing your first suppliers will do more for you than any model, which is the argument of our post on why the first twenty listings are hand-sourced. If the question is whether AI tools can build the marketplace itself, that is a separate topic, covered in our no-code marketplace builder guide.
The way to sequence it: name the stuck side, pick the one AI job aimed at it, define the number it should move before you start, and give it ninety days. If the number moves, keep going. If it does not, the constraint was somewhere else, and that is useful to know too.
Working out which side is stuck, and which fix is worth paying for, is the core of our marketplace consulting work. If any of this vocabulary is new, the marketplace glossary covers it.
