How AEO Gets Your Business Named When a Driver Asks Their Car Where to Go

The Question a Driver Actually Asks

A spoken in-car question is a filter, not a search: it names conditions a business either satisfies in machine-readable fact or does not. Google publishes the example itself, in its own help documentation for Gemini in the car. Hey Google, find a supercharger near a sushi restaurant that's open now. One ordinary sentence, spoken at speed, carrying four separate constraints at once.

Read the clauses apart. A supercharger is a charging-hardware check. Near is a distance check, computed against a moving position rather than a postcode. Sushi restaurant is a category check. Open now is a clock check, evaluated at the second the question lands. Four conditions, four different fields, all resolved before anything about quality enters the calculation.

The interaction model is not a command syntax. Google states that a driver can speak naturally without the use of traditional voice commands or phrases, with the assistant reached by saying Hey Google, by the microphone, or by a press and hold on the steering wheel. The question arrives in whatever words the driver already had.

That matters because the channel is now sustained rather than clipped. An on-road study of 32 licensed drivers found mean glance durations with Gemini Live staying well below the two-second safety threshold, then cognitive load holding stable across extended multi-turn conversations. A long spoken exchange is no longer an edge case at the wheel.

So the claim can be named precisely, because it is narrower than it first sounds. Filterable attributes gate eligibility before quality is ever compared. A brand can hold the best reputation in its category, the strongest recall, the largest media budget, then still leave the candidate set at the clock check, silently, in a fraction of a second.

Look down that third column. Every failure in it is operational rather than editorial. Nothing there is repaired by a better headline, a stronger offer, or a larger campaign. Each is a field on a record, held by somebody inside the business, either correct at the moment of the question or not. That is what machine readability means in a car.

What Changed, Stated Exactly

The dateline deserves stating without inflation. On 30 April 2026 Google described Gemini as starting to roll out in cars with Google built-in, in its own words as an upgrade from Google Assistant. The rollout starts with English in the United States, coming not only to new cars but also to existing ones through a software update.

Two details there carry the operational weight. The software-update path means the surface arrives in vehicles already sold, so the installed base changes without anybody buying anything. The help documentation also still presents Gemini or Google Assistant as a settings choice, so nothing here describes Assistant being removed.

Android Auto came earlier. On 20 November 2025 Google brought Gemini to Android Auto globally in 45 languages, stating the surface as available over 250 million cars on the road, with the assistant able to give insights from reviews then answer common questions about businesses. That last phrase is the commercial one.

The scale figures need holding apart carefully. Google's May 2025 post states over 250 million cars that support Android Auto, then more than 50 car models with Google built-in. The first is a vehicle count. The second is a model count. Google publishes no vehicle count for Google built-in, so none is stated here.

Conflating those two numbers is the most common error made about this surface, which is why the figure above keeps them in separate cards with separate units. A model count says how many nameplates ship the system. It says nothing about how many of those vehicles are on the road, nor how many drivers ever switch the assistant on.

Durability is the reason this is not a passing feature. Deloitte's analysis of software-defined vehicles, built on interviews with over 160 OEM executives, expects 81 percent of OEM fleets to be software-defined by 2030. A fleet that updates itself is a fleet whose assistant keeps changing long after the sale is booked.

Why a Spoken Question Narrows the Field

Voice narrows the field, which is a claim that needs evidence rather than assertion. A voice-only search study presented a ranked list of only the top four results, explicitly in contrast to the top ten used for text, because the voice channel is transient and linear. Four slots, read aloud, in a fixed order.

The reason is physical rather than editorial. A screen holds ten results in parallel, scannable at a glance, revisable by the eye at no cost. Speech is serial instead. Each result spoken aloud spends seconds of attention a driver does not have spare, so the presentation layer compresses the list long before anybody has asked it to.

The input side changes shape too. A field study of 60 participants found voice searches significantly more likely than text searches to be phrased as complete sentences, then significantly more likely to contain question words. Its conclusion was blunt: optimisation has to shift from keyword targeting to the syntax and semantics of spoken language.

The honest counter-evidence sits in the same literature. A study of response presentation found short responses reduced cognitive load then improved satisfaction on low-complexity tasks, yet on high-complexity tasks they led to more query issuance rather than less. Its stated conclusion is that no single response style is optimal in all situations.

So the correct word is narrowing, never singularity. Nothing in the evidence supports the idea that a car returns one answer to every question. What it supports is a shorter list, a longer conversation on the harder questions, then a candidate set filtered before any presentation logic runs at all. That is also where a spoken query and an AI answer now converge.

The distinction matters commercially, because the second question is where a difficult decision actually gets made. A brand present through a multi-turn exchange has more chances to be named than a single-shot model would ever allow. It also has more chances to be eliminated, on any attribute that fails a check partway through the conversation.

The Fields Google Actually Filters On

The fields are not a matter of speculation, because Google documents them. The Places API data fields reference exposes the operational record as discrete typed values: accessibilityOptions, evChargeOptions, parkingOptions, paymentOptions, currentOpeningHours, regularOpeningHours, curbsidePickup, delivery, takeout, rating, userRatingCount, goodForChildren, servesBreakfast.

Read that as an eligibility schema rather than a feature list. Each entry is a question a driver might ask in passing, answerable without reading a sentence. Whether the car park takes a van. Whether the charger fits. Whether the door is open at this exact minute. Whether the counter serves breakfast before a shift starts.

Google's own account of local ranking rests on relevance, distance then prominence, with prominence drawn partly from how many websites link to the business, partly from how many reviews it has. The same page states that businesses with complete and accurate info are more likely to show up in local search results.

The publishing side is thinner than most teams expect. Google's LocalBusiness structured-data guidance requires only name then address. Everything else is recommended, including aggregateRating, openingHoursSpecification, then geo at a precision of at least five decimal places. Required is a floor rather than a strategy.

Ownership is the column most groups cannot fill. The fields are typed, documented, then checkable, yet in a multi-location business each one usually belongs to a different function, or to nobody at all. A field with no named owner is a field that drifts quietly, until a driver somewhere hears the wrong answer about a real place.

The boundary belongs stated rather than buried. Publishing a field is not a promise that Google surfaces it. Google's documented gallery of supported structured-data features contains 25 entries, none of which is an in-car answer. No markup buys a spoken mention, so parity is a discipline rather than a lever.

The vocabulary itself already assumes facts that expire. Schema.org's LocalBusiness type declares only five properties of its own, inheriting the operational vocabulary from Place then Organization. Its amenityFeature is not free text either: it expects LocationFeatureSpecification, which adds hoursAvailable, validFrom then validThrough.

Read those three property names as an instruction about cadence. A vocabulary that ships a start date, an end date, then an availability window is a vocabulary designed for facts with a lifespan. Which is the same conclusion reached from a different direction in what markup actually earns a citation.

Your Operational Record Is Contested, and Partly Written by Strangers

The operational record is not a private document. In 2025 the Maps community suggested 80 million updates to business hours then contact information, while Google blocked 79 million inaccurate or unverified edits to Business Profiles across the same year. Both figures describe strangers writing into the record of a business they do not run.

State the limit plainly, because neither figure answers the obvious question. The first is a volume of suggestions. The second is a volume of blocks. Neither states how many wrong edits went live, nor for how long, nor on which fields. Google does not publish that figure, so nothing resembling it is estimated here.

The scale underneath explains the volume. Google Maps analyses over 300 million places, drawing on reviews from more than 500 million contributors, with arrival assistance now highlighting the building entrance, nearby parking, then which side of the street to be on. Arrival is a data problem too, not only a navigation one.

Part of the record is written by customers by design. Google states that operational attributes including accessibility are managed in the Business Profile, then notes that some attributes are populated from customer input rather than by the owner. A brand can be described, accurately or not, by people simply passing through.

The guidelines carry teeth as well. Google requires an accurate address or service area, plus regular customer-facing hours, capping a service area at roughly two hours of driving time from the business base. Failing the guidelines can result in changes to the information, or removal of the business information from Google.

Removal is the sentence to sit with. A brand with a wide marketing radius, a strong reputation, then a service area drawn to flatter the sales plan is not merely ranked lower. Its information can be changed, or taken down, on a rule nobody in the building has read. That is the argument behind treating the profile as infrastructure rather than a listing.

The DSF Eligibility-Before-Excellence Cascade

The DSF Eligibility-Before-Excellence Cascade orders this work into seven stages: four gates, one line, then the two stages most marketing budgets already fund generously. The order is not decorative. Each stage consumes the output of the one before it, so a failure early in the sequence cannot be repaired late in it.

Stage one is Eligibility, Existence. The place has to be a retrievable entity at all, held as one canonical record rather than several competing ones. Take a multi-location vehicle service group as the worked example. A site that moved across its own car park two years ago now has a legacy record, a franchise-era record, then the current one, all three retrievable.

That is entity conflation, the cheapest failure in the cascade to cause then the most expensive to notice. Every field maintained diligently on the current record is invisible on the other two. A driver asking for the nearest service bay can be routed to an address the group vacated, from a record nobody in the group monitors.

Stage two is Eligibility, Accuracy. Every filterable attribute has to match what actually happens on site. The group still lists on-site fast charging at a location where both bays were decommissioned last winter. A wrong fact eliminates faster than a missing one, because the driver arrives, finds nothing, then writes about it in the field the group cannot edit.

Stage three is Eligibility, Currency. The fact has to be true at the moment of the question, maintained at the cadence at which that fact changes. Hours move weekly at some sites, seasonally at others, instantly when a bay floods. A record swept once a quarter is wrong for most of the quarter on any field that moves faster than the sweep.

Stage four is Eligibility, Machine Form. The value has to be typed, sitting in a field or property read without inference, never a sentence a human must interpret. Free parking behind the building, written into a page paragraph, is not parkingOptions. It is prose that happens to contain a fact, which nothing in the retrieval path can filter on.

Stage five is the Precedence Line, the point at which the candidate set closes. It carries the most weight in the cascade because it is the only stage that explains why good marketing loses to worse marketing. Above the line the question is whether a place is eligible to be considered. Below it the question is which eligible places rank highest.

This is where a budget meets a fact it cannot argue with. Nothing that failed a gate is repairable by content, by brand, or by media spend. A campaign can lift a place that is eligible. It cannot enter a place that was already filtered out, because the filtering happened before the comparison the campaign was built to win.

Stage six is Excellence, Comparison, applying to survivors only. Among the places that cleared every gate, reputation then specialism separate the remainder. Google puts prominence partly on links, partly on reviews, so this is exactly where reputation work earns its keep. One stage earlier it earns nothing at all, however well it is executed.

Stage seven is Excellence, Naming. A short ranked set is read back, and everything filtered out earlier is never mentioned at all. This is the only stage a business does not control, which makes it a diagnosis rather than a target. A brand never named on a question it should own has failed at a numbered stage above, which is a finding instead of a mystery.

The cascade is deliberately unkind to the work most brands already pay for, because four of its seven stages are operational rather than creative. Digital Strategy Force Runs the Cascade With the Operations Team, fixes ownership field by field before a word is written, then holds each record current at the cadence its own facts change.

Why This Is a Data-Maintenance Problem, Not a Content Problem

The cadence mismatch is the whole diagnosis. Marketing is refreshed on a campaign calendar, quarterly at best, annually in practice. Operational facts move on a clock of their own. A bay closes on a Tuesday. A payment terminal changes vendor. A charger is decommissioned. None of those events has ever appeared on a content calendar anywhere.

So the first deliverable is a name against every field, not a plan. Hours belong to the site manager, with a holiday calendar owner above them. Access, parking then charging belong to facilities. Payment methods belong to the payments lead. Ratings belong to nobody inside the group at all, because customers are the ones writing them.

That last line is not a throwaway. Two of the fields a car filters on, rating then userRatingCount, are produced entirely outside the business, while Google states that some attributes are populated from customer input as well. A group can influence those fields. It cannot own them, so it should stop budgeting as though ownership were available.

The rule most dispatch brands have never checked is the service-area cap. Google limits a service area to roughly two hours of driving time from the business base. A mobile operation advertising statewide coverage out of a single depot is describing something its own profile is not permitted to claim, on a page nobody in marketing has read.

Structured data belongs here only as parity discipline. The typed values on the site mirror the operational record, so a parser reads the same fact a driver would be told out loud. Nothing in the markup carries a claim the operational record does not support. That is the entire brief, which is why no field-by-field tutorial follows it.

Entity consistency is the same problem one level up. If a location resolves to more than one record, every field maintained on the right record is invisible on the wrong one. Consistency is therefore not a nicety of naming. It is the precondition that makes maintaining any of the fields worth the effort.

None of this is a content problem, which is the uncomfortable part for anybody holding a content budget. A brand that publishes more articles has not moved a single filterable field. A brand that fixes ownership across the record has changed what a car is able to consider, having published nothing. The same logic decides which brand a model treats as the category leader.

What It Is Worth, and What Not to Claim

Sizing the stake calls for restraint, because no source measures revenue won through an in-car answer. What can be stated is the size of the ground it sits on. The United States Census Bureau puts e-commerce at 16.9 percent of total retail sales in the first quarter of 2026, 326.7 billion dollars of 1,929.0 billion, leaving 83.1 percent transacted outside e-commerce.

Two further figures set the frame. Food services and drinking places, the Census Bureau category name, totalled 603,906 million dollars for the first six months of 2026, up 3.8 percent on the same period a year earlier. In 2024, 69.2 percent of United States workers drove alone to work, unchanged from 2023, at a mean one-way travel time of 27.2 minutes.

Read those three together without reaching past them. Most retail revenue still changes hands somewhere a customer travels to. Most commuters travel alone, for roughly half an hour each way, with a hands-free assistant now installed in the vehicle. The overlap of those facts is where the question in this article actually gets asked.

The limits deserve equal billing. Deloitte's Global Automotive Consumer Study, covering more than 28,000 consumers in 27 countries, found local-language voice command support crucial for 43 percent of United States consumers, against 80 percent in China, 80 percent in India then 70 percent in Japan. The United States is the laggard on this measure.

So an American brand is building for a surface a minority of its own customers currently call crucial, while the same study found 52 percent of United States respondents would keep a vehicle longer given regular over-the-air updates. Appetite for the car that keeps changing is already the majority position. Appetite for the voice layer specifically is not.

Google states the caution itself, on the same help page that publishes the supercharger example. It is quoted here in full rather than paraphrased, because a business planning around this surface should plan around the vendor's own warning about it, not around a summary written by somebody selling the work.

That warning cuts two ways for a business. An assistant can state something inaccurate about a place whose record is entirely correct. It can also state something accurate that the brand never published, drawn from a stranger's edit or a review nobody inside the company has read. Neither outcome is controllable, so neither should be promised to a board.

Which sets the honest limit on measurement. No source in this article measures an in-car answer as a traffic source, a conversion path, or a named referrer. Nothing here should be sold as attributable revenue. What can be verified is the input side: whether every field a car filters on is correct, current, typed, then owned by somebody named.

So the decision on the desk is narrow, then unfamiliar. It is not a decision about content volume, nor about brand, nor about media weight. It is a decision about whether the operational record of every location is maintained as verifiable typed fact, at the cadence each fact changes, by a person who answers for it when it is wrong. Everything above the Precedence Line is that one question. Everything below it is work good brands already do well, on places that never entered the set.

FAQ — In-Car Answers

Next Steps — In-Car Answers

• ▶ Pull every location record and count how many distinct records resolve to each physical site. One canonical record per location is stage one, so nothing downstream is worth funding until that count is correct.
• ▶ Put a named owner against every filterable field, site by site. Hours to the site manager, access, parking then charging to facilities, payment methods to the payments lead, with ratings recorded as owned by nobody inside the business.
• ▶ Set a maintenance cadence per field rather than one sweep for all of them. A field that changes weekly cannot be held current by a quarterly review, however thorough that review is on paper.
• ▶ Check the service area of every mobile or dispatch operation against the roughly two-hour driving-time cap. A radius drawn to flatter a sales plan risks changes to the information, or removal of it.
• ▶ Audit the structured data as parity only. Every typed value mirrors the operational record exactly, carries no claim the record does not support, then is corrected in the same change as the record itself.

A brand with locations customers drive to does not need more content to be named in a car. It needs an operational record held as verifiable typed fact, current at the cadence each fact changes, owned field by field by people who answer for it. Digital Strategy Force builds that record with the operations team, sets the ownership before anything is published, then keeps it correct as the sites themselves change. Speak With the Answer Engine Optimization Team.