Search behaviour

How AI is changing the way users search, and how marketplaces are adapting

Mira team4 October 20265 min read

People now search by describing what they need, in full sentences, with context and follow-up questions. AI assistants taught them this habit, and queries typed into AI tools are several times longer than classic keyword searches. Marketplaces are responding in three ways: conversational search on their own site, a presence inside AI assistants, and listings that machines can read.

What has changed in the way people search?

For twenty years, search meant guessing the right keywords. A user who wanted a family car typed a make, a body type and maybe a price. The search box rewarded short, precise inputs, and people adapted to it.

AI assistants reversed that. At Google I/O in May 2026, Google reported that queries in AI Mode are three times longer than traditional searches, and that follow-up questions were growing 40% month over month. An analysis by SOCi found that prompts in large language models average 23 words, almost six times a typical search query. Semrush estimated that 65% to 85% of ChatGPT prompts have no matching keyword in its database.

Users have moved from keywords to descriptions. A keyword query names a product. A conversational query describes a situation: who the item is for, how it will be used, and what matters most. The second kind carries far more intent, and much of it is lost when a search box only accepts keywords.

How many people already search this way?

The habit is no longer limited to early adopters. According to Eurostat, 32.7% of people aged 16 to 74 in the EU used generative AI tools in 2025. OpenAI reported in February 2026 that ChatGPT has 900 million weekly users.

Shopping is a meaningful share of that use. A working paper by OpenAI's economic research team, reported by Modern Retail, estimated that around 2% of ChatGPT queries are shopping-related, which comes to roughly 50 million shopping queries a day.

Why does this matter for a marketplace's own search?

Most marketplace search is built on structured fields. A car listing has a make, a model, a year, a mileage and a number of previous owners. Filters work well when the user already knows which values they want.

Conversational users start somewhere else. They say what they need, and the words rarely map to a field:

“Something reliable for my daughter’s first car” has no field. It implies low mileage, a small engine, good safety ratings and low running costs.
“Room for our big dog in the back” means an estate body type and a large boot, but neither word appears in the request.
“Cheap to run for city driving” touches fuel type, consumption and insurance group at once.

When the search cannot translate, the user gets zero results or a page of irrelevant ones. Many leave without contacting a seller, and the marketplace never sees the demand that walked away.

The gap on most marketplaces is a translation gap. Users describe needs, and listings are written in attributes. Conversational search sits between the two: it turns a description into several structured searches and explains the results in the user's own terms.

Do users who arrive from AI behave differently?

Data from retail suggests they arrive with clearer intent. Adobe Analytics reported that in July 2026 AI-referred traffic to U.S. retail sites grew 62% year over year, and that those visits converted at a rate 60% higher than non-AI traffic, with 53% more revenue per visit.

This is retail data, not classifieds, so the numbers do not transfer directly. The direction is still useful: people who have already described their need to an assistant arrive further along. A marketplace that can keep that conversation going on its own site is better placed to turn it into a lead.

How are marketplaces adapting?

Three responses are emerging, and many marketplaces are pursuing more than one.

1. Conversational search on their own site

Some portals are adding search that accepts plain language. Finn, in Norway, is testing an AI-first home search with a limited audience, and Homes.com in the U.S. has launched a conversational assistant for property search. The goal is to keep users on the marketplace with the same experience they get from an assistant.

2. A presence inside AI assistants

A growing share of discovery now starts inside an assistant. A report by OC&C Strategy Consultants describes how open protocols such as MCP give marketplaces a standard way to expose product search to AI agents. In practice this means publishing an app or connector, so that a user who asks an assistant about a car, a sofa or a flat can see the marketplace's live listings.

3. Listings that machines can read

Both moves depend on listing data that AI systems can interpret. Adobe found that many U.S. retail sites are not entirely readable by machines, which limits how often they appear in AI answers. On marketplaces where users write the listings, incomplete listings and free text make this harder. That is why the translation layer matters as much as the interface.

What should a marketplace team do first?

A useful first step costs little and needs no new technology:

Read your zero-result and refined queries. They show where users already write in sentences and the search fails them.
Collect ten real questions your users ask, in their own words, and run them on your current search. Note what comes back.
Decide where the conversation should happen first: on your site, inside an AI assistant, or both.
Keep your own search and inventory as the source of truth, so every answer points to a real, live listing.
Measure the result in your own numbers: tag every click from conversational search to a listing, and compare those leads with the ones from your usual search.

Where Mira fits

Mira on your site (widget) is a conversational search layer for marketplaces in any category. It connects to the search a marketplace already has with one line of code, understands what users describe in their own words, and shows the listings that fit. The same engine also powers your MCP, which we package for the AI assistants you choose, from ChatGPT to Claude, and you publish under your own name. See how Mira works for marketplaces.

Frequently asked questions

Will conversational search replace filters?

No. Filters remain the fastest tool once a user knows exactly what they want. Conversational search helps earlier, when the user can describe a need but not yet the values that match it, and it can apply the right filters on their behalf.

Is this only relevant for large marketplaces?

No. Smaller and vertical marketplaces often have the most specific vocabulary, and the gap between how users speak and how listings are written can be wider there.

What is an MCP for a marketplace?

An MCP (Model Context Protocol) is the standard way AI assistants connect to outside services. For a marketplace, it lets assistants like ChatGPT or Claude search its live listings, under its name, and link back to the listing on its site.

How do you know if conversational search works?

Look at your own numbers. Tag every click from conversational search to a listing, for example with a dedicated UTM, and compare those leads with the ones from your usual search.

Curious how your users' questions perform on your search?

Let's talk