Claude Vehicle Matching Assistants for Dealership Websites

An AI vehicle matching assistant built on Claude answers the question most dealership websites still cannot: "which of your 600 cars is right for me?" A multi-site franchised group listing new, approved-used and trade stock across eight locations presents buyers with filters for make, model, price and fuel type, and then leaves them to work out the rest. Buyers do not think in filters. They think in situations: a monthly budget of around £350, two child seats and a dog, a 40-mile commute into a clean air zone, occasional towing of a small caravan, and a nervousness about electric range in winter. Translating that into a shortlist takes an experienced salesperson twenty minutes on the phone, and the website offers nothing comparable.

The result shows up in the numbers dealer principals track. Website sessions are high, but the ratio of sessions to enquiries is low, and many enquiries are for the wrong car: a buyer picks the cheapest seven-seater in the list, discovers on the test drive that the boot disappears with the third row up, and the appointment ends without a deal. Meanwhile stock that would have suited them sits on a different site's forecourt, ageing past 60 days and heading for a price reduction. Matching errors are expensive at both ends: wasted appointments for the sales team and slow-turning stock for the group.

Claude, integrated into the dealership website and connected to the group's live stock, changes the conversation. It reads a buyer's needs in their own words, asks the one or two clarifying questions a good salesperson would ask, and returns a ranked shortlist drawn only from vehicles actually in stock, with a plain explanation of why each one fits and where it falls short. The shortlist leads straight to a test-drive booking at the nearest site. Salespeople receive the buyer's requirements already structured, and the first conversation starts from understanding rather than from "so what are you looking for?"

Where Traditional Vehicle Search Falls Short

Stock search on dealership websites is built around the data the DMS holds: make, model, derivative, year, mileage, colour, price, fuel and transmission. That data is accurate but narrow. It says nothing about whether the boot takes a pushchair, whether the rear doors open wide enough for an ISOFIX seat, or whether the real-world range covers a winter commute with the heater on. Buyers who do not already know which model they want are left guessing, and buyers who do know are often wrong about what suits them. The website's job of pre-qualifying stops at the price slider.

Manufacturer-supplied configurators and "help me choose" quizzes try to close the gap, but they are static decision trees that only cover new cars and cannot see the group's used stock. Group-level matching across sites is even weaker. A buyer in Exeter has no way to know that the ideal car is at the Taunton branch unless a salesperson happens to search group stock for them. Enquiries therefore cluster on the most visible listings, appointment quality suffers, and the group's slowest stock stays invisible no matter how well it would fit the people browsing the site.

What Claude Adds to a Vehicle Matching Website

• Claude interprets a buyer's situation, constraints and preferences from ordinary language and knows when to ask a clarifying question.

• It produces a structured requirements profile and a ranked, explained shortlist strictly from the vehicles in stock.

• It routes the buyer to a test drive at the right site and briefs the salesperson before the first call.

Interpreting Real Buying Situations

A buyer types "need something bigger than my Focus, three kids, budget about £18k or £300 a month, mostly school run and motorway to Bristol twice a week, must be automatic". Claude extracts the seating need, the budget in both forms, the mileage pattern, the transmission constraint and the implied preference for comfort on long runs. It notices what is missing, such as whether the buyer has a part-exchange or needs a tow bar, and asks only the questions that would change the shortlist.

The model also handles the soft signals. "Reliable" from a family buyer usually means low running costs and a warranty; "economical" from a high-mileage driver may point to diesel or a plug-in hybrid depending on charging access. Claude is given the dealership's own guidance on how to interpret such phrases, so the interpretation follows the sales team's judgement rather than a generic rulebook.

Structured Requirements and Explained Shortlists

Claude returns a fixed JSON profile: seats, body style, budget (cash and monthly), fuel preferences, must-haves, nice-to-haves, deal-breakers, driving pattern, location and part-exchange. Against that profile it scores each candidate vehicle supplied by the stock service and returns the top five with a fit score, the reasons it fits, and honest caveats such as "third-row seats reduce boot space to 200 litres" or "this car is at the Taunton site, 35 miles away".

Because the output is structured, the website can present the shortlist as cards, save it to the buyer's account, and pass the profile into the CRM. A buyer who understands why a car was suggested is more likely to book, and a salesperson who reads the same reasoning is better prepared.

Routing to the Right Site and Person

Every shortlist links to a test-drive booking that respects site opening hours, vehicle availability (already reserved, in preparation, at the bodyshop) and the diary of the sales team. Claude identifies whether the buyer is a retail customer, a business user asking about benefit-in-kind, or a Motability customer, and the booking routes accordingly.

Where the best-fit vehicle is at another site, the integration can offer a transfer or a video walk-round, and the group decides the rule. The salesperson receives the profile, the shortlist and the buyer's own words, and the DMS lead is created with the stock IDs attached rather than an empty note.

Best Use Cases for Claude Vehicle Matching

• Multi-site franchised groups gain group-wide matching that surfaces the right car regardless of which forecourt holds it.

• Used-car supermarkets with large mixed stock use matching to convert browsers who arrive without a model in mind.

• Electric-vehicle specialists use it to answer range, charging and cost questions accurately before the test drive.

Multi-Site Franchised Groups

A group with eight sites and several brands has a stock file that changes hourly and a sales team that rarely sees the whole picture. Claude matching works from the consolidated feed, so a buyer on the Ford site can be shown a suitable approved-used Kia at the sister dealership when the rules allow it. Brand rules are respected: the group decides whether cross-brand suggestions appear and how they are worded.

The accuracy benefit shows in appointment quality. When the shortlist is right, the test drive confirms rather than discovers, and the conversion from test drive to order rises. Aged stock across the group gets a fair hearing because the matcher scores on fit, not on which listing has the best photographs.

Used-Car Supermarkets

Supermarkets carry 1,000 or more vehicles and many buyers arrive with a budget and a vague idea. The site's filters produce 300 results, which is no better than none. Claude narrows that to five with reasons, and the buyer's answers to two clarifying questions do more than any facet panel. Speed matters here because supermarket buyers compare several sites in one evening.

The integration also captures the buyer's profile for retargeting with consent: if nothing fits today, the buyer can opt in to be told when a matching car arrives, and Claude re-runs the match against new stock each morning.

Electric Vehicle Specialists

EV buyers ask questions that depend on their circumstances: whether they can charge at home, how far the daily commute is, whether they need rapid charging on the motorway, and what the real range is in cold weather. Claude works from the dealership's own range and charging data for each vehicle and explains the trade-offs honestly, including the difference between WLTP figures and typical winter range.

This is where accuracy protects reputation. An EV sold to a buyer with no home charging and a 90-mile commute becomes a complaint and a review. A matcher that flags the mismatch before the test drive saves the sale from happening for the wrong reasons.

Core Features of a Claude Vehicle Matching Website

• A conversational intake that collects the buyer's situation in their own words, with optional guided prompts.

• A matching engine that pairs Claude's requirements profile with live stock and returns explained, ranked results.

• A booking, CRM and analytics layer that turns shortlists into test drives and shows which stock is being matched.

Intake Layer

The intake is a chat panel or a short guided form on the website, available on the homepage, the search page and each vehicle listing. Buyers can start with a sentence or with structured choices, and Claude fills the gaps by asking at most two or three questions. The intake respects the buyer's time: it never asks for contact details before showing a shortlist, and it remembers the profile within the session.

The intake layer also handles returning buyers. A saved profile can be reloaded, and changes such as a new budget or a different location re-run the match without starting over.

Intelligence Layer

Behind the panel, the backend retrieves candidate vehicles from the stock service using the hard constraints (budget, seats, transmission, fuel) and passes the candidates, the buyer's words and the dealership's interpretation rules to Claude. The model returns the profile, the ranked shortlist and explanations in strict JSON, which the backend validates before rendering. Vehicles that are reserved or sold since the last feed refresh are removed at this stage.

The layer keeps a record of every match: profile, candidates, result and buyer actions. That record is what allows the group to see how well the matcher performs and to refine the rules.

Action and Dashboard Layer

Each shortlisted car carries a test-drive button that books into the right site's diary, creates the lead in the DMS with the profile attached, and sends the buyer a confirmation. Sales managers see a dashboard of profiles, shortlists, bookings and outcomes, with filters by site and stock age.

Stock buyers gain something new: a view of what customers are asking for that the group does not have. If forty profiles in a month want an automatic seven-seater under £20,000 and the group holds two, that is a buying signal no portal provides.

Step-by-Step Integration Process

• Define what a good match means for this group, including the rules for cross-site and cross-brand suggestions.

• Design the intake and shortlist experience around real buyer conversations, not around the stock schema.

• Connect the backend to Claude, save every match, keep the sales team in control of bookings and measure results.

Step 1: Define Matching Goals, Categories and Business Rules

Bring the sales director, a top salesperson and the stock buyer together. Agree the target metrics: session-to-shortlist rate, shortlist-to-booking rate and test-drive-to-order rate. Write down the interpretation rules a good salesperson uses: what "reliable" implies, when to suggest a hybrid, when to mention a car at another site. Decide the deal-breakers the matcher must never override, such as budget ceilings and seat counts.

Define the shortlist format, the number of results and the tone of explanations. These decisions become the system prompt and the JSON schema.

Step 2: Design the User Journey for Buyers and Sales Teams

Sketch the buyer's route from the first sentence to the confirmation email, including the moment contact details are requested. Then sketch the salesperson's route: the lead arrives with a profile, the salesperson reviews the shortlist, calls the buyer and confirms the appointment. Both journeys should be tested with recent enquiries rewritten as if they had come through the matcher.

Decide how much the sales team can adjust. Some groups let salespeople edit the shortlist before the buyer sees it; most let the buyer see it immediately and give the salesperson a chance to add alternatives.

Step 3: Connect Your Website Backend to Claude

The backend, on Velo, Node or WordPress, exposes a matching endpoint. It fetches candidates from the stock service, assembles the prompt with the rules and the buyer's words, calls the Anthropic Messages API and validates the JSON. Stock data is refreshed from the DMS or the portal feed at the same cadence as the group's adverts, and reservation status is checked at booking time.

Timeouts and fallbacks are designed in. If the model is slow or returns invalid JSON, the buyer sees the conventional filtered search with a polite note, never a blank screen.

Step 4: Save Results, Trigger Workflows and Keep Humans in Control

Every profile, shortlist and booking is stored with timestamps and consent flags. Workflows follow: DMS lead creation, diary booking, confirmation messages and the daily re-match for buyers who opted in. Salespeople confirm every appointment and remain responsible for pricing, finance and part-exchange discussions, which the matcher deliberately does not attempt.

When a salesperson marks a suggestion as poor, the reason is recorded and reviewed.

Step 5: Measure Matching Quality and Improve Over Time

Compare booking and conversion rates for matched leads with conventional website leads. Review a weekly sample of shortlists with the sales team and adjust the interpretation rules. Watch the stock-age profile of matched vehicles: if the matcher is fairly surfacing older stock and those cars are selling, the business case is proven.

Over time the group can add insurance groups, running-cost estimates or warranty terms to tighten the matching further.

Security, Privacy, Cost Control and Long-Term Scalability

• Buyer profiles are personal data under UK GDPR and are stored with consent, purpose limits and retention rules.

• The matcher never discusses finance figures, so FCA consumer credit obligations stay with authorised staff.

• Costs scale with matches, not with sessions, and are kept low by pre-filtering candidates before calling Claude.

A requirements profile reveals family size, location, commute and budget, which is personal data even before a name is attached. The website should explain what is collected and why, request consent before saving a profile to an account or using it for re-matching, and delete profiles on a defined schedule. API calls are made through the business tier of the provider with no training on the group's data, and stock data sent to the model contains no customer information. The ICO's guidance on automated decision-making is relevant: the matcher recommends, it does not decide, and buyers can always browse the full stock.

Finance is deliberately outside the matcher's scope. It can show the advertised price and, if the group chooses, the approved representative monthly example, but it never calculates a personal quote, which keeps the group's FCA permissions and Consumer Duty responsibilities with trained staff. Cost is controlled by filtering candidates in the database first and sending Claude only the twenty or so that pass the hard constraints, by capping tokens per match and by caching the rules and stock context. Adding a site or a brand is a data change, and the same engine can later serve the aftersales site or a mobile app.

Davydov Consulting scopes and builds vehicle matching assistants on the client's existing platform, whether that is Wix Studio with Velo, a Node and Firebase stack on Google Cloud, WordPress or a native mobile app. We connect the stock feed and DMS, design the intake and shortlist experience with your sales team, and keep people responsible for every booking, price and finance conversation. If you would like to see how matching could work with your stock, we would be glad to discuss it.