Search is one of the most important experiences on AutoScout24. For many users, it is the moment where a broad intention (“I need a new car”) becomes a concrete shortlist of vehicles worth considering. But automotive search is rarely simple. Some buyers know exactly what they want down to the brand, mileage and fuel type. Others start with lifestyle needs, such as wanting a spacious electric SUV for the family. Some are still exploring and need guidance before they can define the right criteria at all.
That is why search on AutoScout24 is evolving beyond a single search pattern. Today, different search experiences serve different user needs: classical structured search, AI semantic search and AI conversational search. Together, they help buyers move from intent to relevant vehicle results in ways that fit how they actually think, compare and decide.
Three search experiences for three different user needs
Classical Search
The first and most established search experience is classical search, also known as parametric or structured search. This is the familiar filter-based approach where users manually configure their criteria through checkboxes, radio buttons, range sliders, and dropdowns. They can define attributes such as brand, mileage, fuel type, transmission, price and many other vehicle-specific details.
Classical search works especially well for high-intent buyers who already know their exact requirements. A user searching for a Volkswagen Golf under a certain price, with limited mileage, automatic transmission and petrol engine, expects deterministic control. They want every filter adjustment to behave predictably, with clear boundaries and a visible impact on the result count. This experience serves the analyst persona: pragmatic, methodical and confident in the technical specifications that matter to them.
AI Semantic Search
The second experience is AI semantic search. Here, users can enter a natural language description into a search bar instead of manually selecting every parameter. The AI interprets the underlying intent, identifies concepts that match AutoScout24’s available filters and translates the user’s phrase into a structured search.
This is particularly valuable for buyers who think in terms of real-world needs rather than catalogue specifications. Someone might search for a “spacious electric family SUV with more than 500 km range”. Rather than requiring that person to know which filters to combine, semantic search reduces friction by converting a multi-faceted need into actionable parameters. It is designed for the expressive searcher: someone who is goal-oriented, values convenience and describes what they need in practical language.
AI Conversational Search
The third experience is AI conversational search. Instead of relying on a single query, users can engage in a back-and-forth dialogue with an AI assistant. The system asks guiding questions, gives advice and refines the search over multiple turns. As the conversation develops, it extracts relevant criteria and dynamically maps them to available platform filters.
Conversational search is especially helpful for users who are early in their buying journey or overwhelmed by the number of choices. A user might start with, “I need a family car for five people that is fuel efficient and under CHF 40,000.” From there, the assistant can help clarify needs, weigh trade-offs and gradually narrow down the options. This serves the advisee persona: undecided, open to exploration and looking for step-by-step guidance similar to speaking with a knowledgeable dealership advisor.
From natural language to vehicle results
Behind the scenes, both semantic and conversational search rely on the same core challenge: translating human intent into platform filters that can retrieve relevant vehicles.
In semantic search, the system uses natural language processing to extract available AutoScout24 filters directly from the user’s text prompt. Once those parameters are identified, the search engine filters the inventory and orders the matching vehicles using AutoScout24’s internal ranking system.
Conversational search adds another layer of complexity because the query is not fixed at one point in time. The AI analyses the ongoing dialogue, identifies how the user’s intent changes or becomes clearer, and continuously maps that evolving intent to the corresponding filters. The resulting candidate vehicles are then isolated and presented using AutoScout24’s ranking logic.
In both cases, the aim is not to replace the structured search engine, but to make it easier for users to express what they want. AI becomes a translation layer between everyday language and the structured vehicle data that powers the marketplace.
The hardest challenge is not matching specs, it is matching intent
Automotive search is technically complex because vehicle data is highly structured, but user intent is often not. Matching a brand, fuel type or mileage range is comparatively straightforward. Understanding what someone means by “budget-friendly”, “family car”, “sporty” or “car for work” is much harder.
Subjective terms vary from person to person. “Budget-friendly” may mean one price range for a first-time buyer and a completely different one for someone looking at premium electric vehicles. “Family car” could imply five seats, generous boot space, high safety ratings or all of these at once. “Sporty” might refer to horsepower, acceleration, design or driving feel.
The system therefore performs best when users provide clear and concrete criteria. The more ambiguous the language, the more the AI has to infer. A phrase like “car for work” is a good example: it could mean a reliable commuter car, a vehicle for carrying tools and equipment, or something suitable for visiting clients. Turning that kind of intent into useful filters requires careful interpretation and ongoing learning.
Why data quality and freshness matter for search relevance
Search relevance depends not only on the intelligence of the search system, but also on the quality of the listings it can search through. Much of the vehicle information on AutoScout24 starts with data provided by sellers. That data can vary in depth and structure from listing to listing.
AutoScout24 works to enrich and standardise listing information where possible, filling gaps and making details easier to compare. But the starting point still matters. A listing with rich details and clear photos is easier to match accurately than one with a lighter description. Better data gives the search system more reliable signals and gives users more confidence in the results they see.
Newness is just as important. Cars sell, prices change and listing details can be updated. Real-time updates help ensure that users are seeing accurate and relevant options rather than vehicles that have already moved on. In a fast-moving marketplace, relevance is not only about finding the right match — it is also about finding a match that is still available and correctly represented.
Measuring whether search is working
For AI-powered search experiences, direct user feedback is especially valuable. AutoScout24 uses signals such as thumbs up and thumbs down on results to understand whether the system has correctly interpreted what a user was looking for.
Seekers can also rate how well the translated filter results match their original search intent. This helps identify the gap between what people typed and what they actually meant. These feedback loops are important because semantic and conversational search are not only about technical retrieval quality. They are about whether the user feels understood.
Over time, this feedback helps reveal patterns: which types of queries work well, which are ambiguous, and where the system needs to improve. It also supports a more user-centred approach to search development, where improvements are guided by real interactions rather than assumptions alone.
Balancing precision, speed and scalability
Building search for a large marketplace means constantly balancing precision, speed and scalability. More precise understanding often requires more computation. The more layers of interpretation, mapping and ranking are added, the more demanding the search process becomes.
AutoScout24’s approach is to keep the fast path fast. Simple filter-based searches remain lightweight so that users receive results instantly.
Scalability depends on smart architectural choices. Efficient indexing, robust filtering and careful system design help ensure that additional intelligence does not slow down the platform as listings and traffic grow. The challenge is to make search smarter without compromising the responsiveness users expect from a modern marketplace.
What recent experiments are teaching us
Recent improvements have followed a gradual rollout approach. Conversational search is currently fully rolled out on web and to a percentage on native apps. For now, these experiences are scoped to the car category rather than campers or utility vehicles.
This staged rollout is deliberate. It allows the team to study real queries, analyse feedback and understand where AI-powered search creates the most value before expanding further. Rather than launching everywhere at once, the focus is on learning from actual user behaviour and improving the experience step by step.
One trend already standing out is the growing interest in electric vehicles. Users are asking increasingly nuanced questions about EVs, such as range, suitability for family use and practical day-to-day requirements. This shows how search behaviour reflects broader shifts in the automotive market and highlights where intelligent search can be especially useful.
The future of search on AutoScout24
The biggest opportunity for the future is to continue improving how well AutoScout24 understands what users really mean. Search is moving from a purely parameter-driven experience towards a more flexible system that can support different levels of certainty, knowledge and intent.
For expert buyers, classical search will remain essential because it provides control, precision and speed. For users who know their needs but not the technical filter combinations, semantic search offers a faster and more natural way to express intent. For users who are still exploring, conversational search can provide guidance and reduce the complexity of choosing the right vehicle.
As the team learns more from real-world usage, the next steps include refining the quality of intent understanding, expanding successful patterns to more categories and continuing to close the gap between what users say and what they mean. In the end, the goal is simple: help every seeker find the right vehicle more easily, whether they search like an analyst, an expressive searcher or an advisee.
Author
Raluca Lacatusu
Lead Product Manager
Automotive