The car buying customer journey, and the stage no dealer can measure

Almost every published map of the car buying customer journey begins at the same place: a buyer who knows they want a specific vehicle and starts researching it. That map describes 29% of the market. The other 71% enter the process with no fixed vehicle in mind, and the decisions that matter most to a dealership happen inside that undecided stretch, on surfaces no store can measure.

The figures throughout come from the 2025 Cox Automotive Car Buyer Journey Study, the sixteenth annual edition, released January 2026 and based on 2,300 consumers who purchased a new or used vehicle in the preceding twelve months.

What the buying journey actually looks like

Five phases hold up against the data, and the useful part is their relative size rather than their sequence.

Undecided. A need exists and no vehicle has been chosen. Cox reports 71% of buyers entered with an open mind, and that only 29% were certain of a vehicle, down from 37% in 2020. This phase has been growing for five years.

Shortlist. A few models and a few stores survive. Cross-shopping is now the norm rather than the exception: 66% of buyers considered both new and used vehicles, up from 57% previously, and among used-vehicle buyers that figure reaches 74%.

Comparison. Specific units, prices and stores get evaluated across an average of 4.6 websites. This is the first phase where a dealership site is reliably involved, and by the time it starts, the set of candidate stores has already been drawn.

Contact. The buyer reaches one or two stores by form, phone or walk-in. This is where dealer measurement begins and where most dealer marketing spend is aimed.

Transaction. Still overwhelmingly in a building. 53% completed all required steps in person and 7% purchased entirely online, with 63% saying the ideal experience blends the two.

Read the sizes rather than the arrows. Dealer visibility begins at phase four. The selection happened in phases one and two.

The undecided phase is now the largest one

A buyer who already knows they want a specific trim of a specific model is running a shopping task. A buyer who knows only that the current vehicle will not survive another winter, that there are now three children, or that the payment has to come down is running a different task entirely, and it is a question-answering task rather than a search-and-compare one.

Cox found selection satisfaction improving even as certainty fell, with 62% of buyers highly satisfied with the selection available against 36% during the inventory shortage of 2022. So the undecided buyer is not frustrated. They are comfortably exploring, over a longer window, with more options considered than before.

Affordability is driving much of it. 62% of buyers said leasing or owning a vehicle is too costly, and Kelley Blue Book estimated the average new-vehicle suggested retail price peaked above $52,600 in December 2025. A buyer under that kind of price pressure does not start with a model. They start with a budget and a use case, and then ask what fits.

Nearly all dealership content is built for the shopping task. Model pages, inventory pages, offer pages and comparison pages all assume a chosen vehicle. The question-answering task is served, when it is served at all, by marketplaces and publishers.

The stage that leaves no trace

In its first year of tracking the behaviour, Cox found 19% of all buyers and 25% of new-vehicle buyers used AI websites such as ChatGPT or Copilot, or AI-generated overviews, during the shopping process. Those buyers reported the highest satisfaction levels in the study, and 84% of mostly-digital buyers who engaged an AI assistant were highly satisfied with the overall process.

Those two findings sit next to each other for a reason. The undecided buyer has an open-ended question, and an assistant answers open-ended questions better than a results page does. The behaviour and the phase match.

What makes this different from every previous shift in car shopping is the absence of evidence. When buyers moved to marketplaces, dealers could see referral traffic. When they moved to mobile, dealers could see device splits. When a buyer asks an assistant which three-row SUV suits a family of five on a fixed budget in a specific metro, the store that gets named receives no impression, session, referrer or line in any report.

If that answer names three dealers and the buyer searches one of them by name a week later, the store records a branded search with no discoverable origin. The event that decided the outcome is not underrepresented in the analytics. It is absent, and no amount of better attribution software recovers it, because the interaction never touched the store's property.

Where buyers actually spend the five visits

The study measured which categories of site buyers used, and the ordering is worth sitting with because it describes the competitive position of a dealership website precisely.

Third-party marketplaces reached 75% of buyers. Dealership sites reached 59%. Search engines reached 41%. Used-vehicle online retailers reached 30%, social media 26%, automaker sites 25%, and AI sites 12%, rising to 17% among new-vehicle buyers. The average buyer visited 4.6 sites in total.

Two things follow. First, a dealership site is a minority surface in its own category, reached by fewer buyers than the marketplaces it competes with for the same units. Second, the AI figure is the only one in that list that did not exist as a measurable category a year ago, and it is already ahead of nothing at all while sitting below every established channel. Treating 12% as small is a reasonable reading of a single year. Treating it as static is not, given that 83% of consumers in the same survey said AI will affect car buying in future and 63% of dealers said investing in AI tooling now is critical.

The competitive point is sharper than the volume point. A marketplace listing publishes price, mileage, trim and options as plain readable text, which is exactly the form an engine can retrieve and quote. Many dealer sites publish the same facts inside interactive components that no engine reads, a pattern covered in more detail in automotive digital retailing. When an assistant assembles an answer about available inventory in a market, it is working from the sources that made themselves readable.

What decides which store is in the shortlist

An engine naming a dealer is doing something narrower than ranking. It is retrieving specific claims it can attribute, and then repeating them. That changes what earns the mention.

Specificity beats volume. A page that states which models a store keeps in depth, what its service department actually performs, and what a named market area covers gives an engine something quotable. A page describing a commitment to customer satisfaction gives it nothing, and every store has one of those.

Readable facts beat rendered ones. Price, availability, trim and location have to exist as text and as structured data rather than only inside a script that runs after a click. This is the single most common failure and it is technical rather than editorial.

Review text beats review scores. A four-point-six average is not quotable. A review naming a vehicle, a repair and an outcome is. The distinction is covered at length in automotive reputation management, and it is the reason generic review requests produce unusable text.

One store gets named, not a group. A generated answer returns a business, so a multi-rooftop operator does not choose which of its stores appears. That is a structural problem for groups and is treated separately in dealer group AI search visibility.

Answering the questions that get asked before a model is chosen

If 71% of buyers arrive undecided, then the content that reaches them is comparative and situational, and almost no dealership publishes it.

What fits a specific use, not what is on the lot this week. How two trims differ once you live with them, in terms a buyer would notice. What ownership actually costs across fuel, insurance, maintenance and rate, which matters more than ever with 62% of buyers reporting affordability pressure. Whether leasing makes sense for a given situation, given that 29% of new-vehicle shoppers weighed leasing against buying, an all-time high.

These are ordinary questions with locally specific answers, and they are the ones asked at the point the candidate set is drawn. A store answering them is present when the shortlist forms. A store publishing only inventory and offers is arguing for a decision that has already been made.

None of this requires guessing at what an assistant wants. The questions are observable in search data, in the questions the sales floor already fields daily, and in the segments the store actually stocks. The publishing discipline behind it is automotive content strategy, and the measurement side is covered in vehicle search analytics.

What this changes, and what it does not

The transaction has not moved. Buyers still finish in a building, satisfaction with that experience is at a record high, and dealership satisfaction reached 76% in the latest survey. Nothing in this data suggests a store should reorganise around online transactions that 7% of buyers complete.

What has moved is selection. The set of stores a buyer will consider is now assembled during a phase the store cannot observe, increasingly with help from a system the store cannot query, on the basis of facts the store may not have published in readable form. That is a discovery problem wearing the costume of a marketing problem, and it does not respond to more spend on the channels that measure well.

VulcanAX works on the retrievability half of that: whether the pages are indexed, whether the facts are readable, and whether an engine asked a question the store should own can find and attribute an answer to it. Lead handling, CRM, media and the showroom process stay with the store. The boundary is that this operator can affect whether a rooftop is in the answer, and can do nothing about what happens once the buyer walks in.

FAQ