Why RoadMatch exists
Car advice should start with the driver.
Most car searches begin with make and model. RoadMatch begins with the life the car has to fit—then shows its working.
Our position
A useful recommendation is not “buy this”. It is “this fits because—and here is what you give up.”
RoadMatch is an independent recommendation and referral concept for UK car buyers. We combine a structured questionnaire with editorial model data, then rank the models in our reviewed set against the answers provided.
The reviews are original, AI-assisted RoadMatch syntheses. Primary facts are checked against manufacturer, government and safety material before we compare attributed independent assessment from Autocar, Auto Express, Honest John and Carwow. Every review links the evidence used and flags material generation or trim differences. We do not reproduce publisher wording, photography or scores, and these are not first-hand RoadMatch road tests.
We do not scrape or republish marketplace listings. When a user wants to see live stock, we prepare a filtered link to Auto Trader and clearly identify any preferences that still need applying.
Editorial evidence standard
A complete review must be checkable.
Coverage begins with officially marketed UK passenger cars from 2010. Reviews are organised by generation, with material facelifts or powertrain changes split when the buying advice changes.
Browse published reviewsGeneration, model years and recommended version are stated
Specifications are checked against authoritative or manufacturer material
Safety claims link to the relevant Euro NCAP or official record where available
Independent editorial conclusions are compared across named UK publications
Source scope differences and uncertainty are disclosed
Image reuse rights and UK context are recorded
An editorial QA pass is required before publication
Principles before features
Four rules keep the product useful.
Explain every score
Each recommendation retains the contribution from budget, practicality, driving, ownership and evidence—not just a number.
Name the compromise
A high score never hides a tight rear seat, high insurance group or feature that depends on the individual trim.
Separate advice and inventory
Editorial recommendations live here. Current price, seller claims and availability stay with the marketplace.
Keep estimates dated
Used prices, running-cost ranges, insurance groups and review judgements are snapshots that need periodic review.
Recommendation methodology
Adjustable rules, visible reasoning.
Version one uses deterministic scoring. The weights are configuration, not buried in interface code, so they can be tested, tuned and audited.
Try the questionnaireCollect a practical profile
Budget, condition, body style, seats, doors, powertrain, mileage, driving environment, main-driver age, economy, costs, features, vehicle age, mileage cap, makes and search area.
Apply firm eligibility rules
Avoided makes, seating, doors, gearbox, must-have equipment, budget availability and age-related insurance safeguards are checked before ranking. An ineligible model cannot be rotated into the shortlist, and we return fewer cars rather than pad a result.
Score every eligible model
Each active dimension returns a 0–100 fit. A cost dimension carries no weight when the user selects “No preference”. The current maximum weight set is:
- budget
- 20%
- body Style
- 9%
- practicality
- 9%
- powertrain
- 11%
- driving
- 8%
- economy
- 7%
- ownership
- 8%
- features
- 7%
- age
- 4%
- make
- 3%
- reliability
- 6%
- driver Age
- 8%
Keep the reasoning visible—and the choice varied
The result stores the explanation for every dimension. Strong contributions become “why it fits”; low contributions become trade-offs. Reliability uses our desk-researched model assessment and a confidence label. A device-local seed can reorder only a narrow band of near-equal eligible cars, so a Toyota Aygo can get a fair turn beside a Skoda Citigo without promoting a materially worse match.
Hand off only supported filters
A separate marketplace adapter converts the model and profile into an allow-listed search object. Unsupported or uncertain filters remain visible for the user to apply manually.
Built to evolve
Rules first. AI and more marketplaces later.
The scoring strategy and marketplace adapter are deliberately separate. A future AI system can rerank an eligible rules-based shortlist, write richer explanations or ask useful follow-ups without overriding explicit budget, seating or avoid-list constraints.
Additional authorised marketplaces can implement the same adapter contract and declare which filters they support. The user keeps a consistent explanation even when the destination changes.