Local & Industries · 9 min read · July 15, 2026
Service bookings from the chat window: how AI decides which shop to name
Drivers have stopped typing keywords and started asking questions: "Where can I get my inspection done near me this week?" ChatGPT is at 900 million weekly users, and Google's AI Overviews reach more than two billion people a month, so a lot of those questions get answered before anyone ever sees a list of shops. Generative engine optimization is the work of making sure the models know your service department, describe its brands, hours and prices correctly, and name it when someone asks.
Why the service question now gets settled in the chat window
Your customers' search behaviour has changed shape. The old query was three words in a search box, "brake shop near me," followed by a scan of ten blue links. The new one is a whole situation: "My Golf is due for its 60,000-mile service, which independent shop near me is honest and won't void the warranty?" ChatGPT, Gemini or Perplexity hand back a recommendation rather than a list: two or three named shops, with a reason attached to each. Every other shop in town is simply absent from that answer, and therefore absent from the decision.
This lands harder on the service side than on sales. A car purchase gets researched for weeks across a dozen tabs; a service appointment is settled in one sitting, locally, usually on a phone. Those fast local decisions are exactly the ones moving into the chat window. When a language model names three shops for "AC recharge," "tyre change" or "timing belt replacement" and yours is not one of them, you lose the appointment before the phone rings or the booking form opens.
The hard part, and the real break from classic SEO: there is no page two. On a results page, position eight still gets clicked now and then. In a generated answer there are two or three slots and then nothing. Ahrefs found that only about 6 to 8 percent of the URLs ChatGPT cites also rank in Google's top ten, and roughly 80 percent do not rank in the top 100 at all. Being findable on the web is not the same thing as being the source a model builds its answer from.
What generative engine optimization means for a dealership
Generative Engine Optimization, or GEO, is the practice of being understood correctly by a language model and then named by it. For a dealership that means the model knows more than the fact that you exist: that you service all makes, that you work to the manufacturer's maintenance schedule, that you offer collection and delivery, that your air-conditioning equipment is certified for R-1234yf, and that inspection slots are available the same week. Those facts have to sit on the open web in a form a machine can lift without guessing.
The difference from ranking is granularity. Search ranks pages; a model extracts statements. "We take care of your vehicle" gives it nothing to work with. "We carry out manufacturer-schedule servicing for VW, Audi, Skoda and Seat using original parts, with your factory warranty preserved" gives it a sentence it can quote. Sentences like the second one are what end up inside recommendations.
GEO does not replace the marketing you already do, it feeds on it. Your website, your Google Business Profile, review sites and trade directories are the material these systems retrieve from. The job is to make those sources consistent and dense with facts, so you get filed as a dependable address for service work rather than a name the model is unsure about.
The questions drivers really type into an assistant
To get named, you have to know how drivers actually phrase things. Real prompts look like this: "The dealership quoted me £680 for the 60,000-mile service, can I get it cheaper without losing the warranty?", "Which shop near me works on electric-vehicle high-voltage systems?" or "I failed my inspection and have a list of faults, who can fix it this week?" These are not keywords. They are situations, with a budget, a deadline and a worry attached.
That is where your opening is. A page that answers the situation, explaining why servicing at an independent specialist on the manufacturer's schedule keeps the warranty intact, what the price difference to the franchised dealer usually looks like and what paperwork you hand over, matches that first prompt almost word for word. Given a choice between a page that answers the question and a service menu that merely lists it, a model cites the one that answers it.
Collect these questions on purpose. Your service advisers hear them all day at the counter: "Do I need an appointment for an AC recharge?", "Do you do wheel alignment?", "How long does a tyre change take?" Every question that repeats is a candidate for its own plainly answered section on your site, and another handle for AI visibility.
Structured data: what the machines can read
Retrieval systems reward clarity, which makes markup your cheapest technical lever. With Schema.org markup for AutoRepair, LocalBusiness and Service, your opening hours, your service list, your location and your ratings become machine-readable instead of merely visible. Be straight about the ceiling, though. Google has said since 2018 that schema is not a direct ranking factor, and its current guidance states that no special markup and no AI-specific file are required to appear in AI Overviews. Schema buys extractability, which means fewer invented prices and fewer wrong hours. It does not buy rank.
In practice: give every service its own named entry. Not one vague "workshop" block, but separate items for scheduled servicing, inspection preparation, brake work, air-conditioning maintenance, tyre storage and accident repair, each with a short factual description and a price range where you can give one. That granularity is exactly what gets pulled when someone asks about a single job.
Then keep it consistent. When your name, address and phone number differ even slightly between your site, your Google profile and the directories, a machine is left choosing between versions, and it chooses badly. Columbia Journalism Review's Tow Center ran 1,600 queries across eight AI search tools and found more than 60 percent of the answers about a source were simply wrong, usually stated with total confidence. One identical set of facts everywhere is unglamorous, and it is the floor under every AI recommendation you will ever get.
Reviews and mentions: your strongest lever
Models do not recommend out of thin air. They lean on signals that look like corroboration, and this is the best-evidenced lever anyone has found: an Ahrefs analysis of roughly 75,000 brands found the frequency of brand mentions across the web correlated with AI citation at about 0.66, some three times stronger than backlinks at 0.22. Reviews are the part of that signal you can influence directly. A shop with a long run of recent, wordy reviews saying "got me an inspection slot the same week," "the estimate matched the invoice" or "talked me out of a timing belt I didn't need yet" is handing over exactly the evidence a model looks for. Stars alone carry nothing. The sentences are what gets quoted.
So steer what customers write. After an air-conditioning job or a big service, ask for a review and name the job in the ask. Over a year that produces a review profile that does not just say "great service," but repeats the actual terms people ask about. Those terms are what tie your name to that job when a model goes looking.
Mentions count beyond reviews: the local paper covering your new EV bay, an entry in your trade association's directory, your listing as an authorised service partner on the manufacturer's own site. Every credible outside source that connects your name to a specific competence raises the odds you get named for it.
From chat to booked appointment: mind the gap
Being named is half the job. When an assistant sends someone to your page, the path to a booked slot has to be short. A live service scheduler showing real open slots for servicing, tyre changes and inspections turns a recommendation into a repair order. A contact form promising "we will get back to you" hands the customer straight back. SparkToro and Similarweb measured around 68 percent of US Google searches ending with no click to any website at all, so the visitors who do arrive are worth catching properly.
Make the booking speak the same language as the question. Someone who arrived asking about the 60,000-mile service should find that exact job on the landing page, with duration, price range and the next free slot. That continuity between the answer, the page and the calendar is what converts visibility into cars on the ramp.
Measure the path, roughly. Tag your analytics so you can see arrivals from assistants and answer engines, and treat the trend as the signal rather than the decimal, because referral data from these tools is still patchy. Then watch the physical evidence: a schedule that fills earlier in the week and bays that run level instead of lurching. That is the outcome you were actually buying.
Mistakes that keep your service department invisible
The most common mistake is a sales-only site. Plenty of dealership sites lead with this month's new-car offers and leave the service department as a phone number in the footer. A model reads that the way anyone would: you sell cars, somebody else repairs them. If you want to be named for service questions, service needs the same content weight as sales, with its own pages, its own facts and its own answers.
The second is marketing language with nothing inside it. "Your satisfaction is our motivation" is unusable to a model because there is nothing in it to verify. Trade it for claims that can be checked: which makes, which jobs, which certifications, what the wait is, what it costs. Google's guidance goes further and warns against writing separate content "for AI" at all, so one honest, specific page beats two vague ones.
The third is drift. A Google profile carrying last year's opening hours, two different phone numbers, a scheduler nobody has touched since spring: each inconsistency gives a machine a reason to name someone else instead. Age cuts the other way too. Ahrefs found the pages ChatGPT cites have a median age of around 500 days, so this is maintenance rather than a launch.
A 90-day plan you can run
Start with the questions. For two weeks, have every adviser write down the service questions they get asked, then pool the list. That list is your content plan. Take the twenty that repeat most and write one short, honest, specific answer for each into your service pages. You are now covering the prompts people actually type.
In parallel, fix the plumbing: clean schema for the business and for each individual service, one identical set of contact details everywhere, a scheduler that genuinely shows free slots. Then the trust signals, which means a standing habit of asking for a review that names the job, plus a pass through your directory and manufacturer listings to correct whatever has gone stale.
At the three-month mark, audit it by asking the models yourself. Put your customers' real questions, with your town in them, to ChatGPT, Gemini and Perplexity, and note both whether you appear and whether what they say is right. Two caveats keep this honest: Profound found only about 18 percent of ChatGPT conversations trigger a web search at all, and answers vary from run to run. Repeat the check monthly instead of reading one result as a verdict. Wherever you are missing, you have found a question your site has not answered yet.
Questions we get asked
As an independent shop, do I lose out to the franchised dealers in AI answers?
Usually the reverse. A model names the source that answers the question best, not the biggest sign on the building, and the hierarchy it uses is not the search hierarchy: Ahrefs found only about 38 percent of the URLs cited in AI Overviews also rank in the traditional top ten, down from roughly three-quarters a year earlier. Questions like "can an independent shop service my car without voiding the warranty" play directly to you. Answer that situation plainly, name the makes, the jobs and the price ranges, keep the reviews coming, and size stops being the deciding factor.
Do I need to rebuild the dealership website for this?
No. It is almost always enough to give the service side real content and make the facts machine-readable: a page per service, specific wording instead of promotional wording, clean schema, and a workshop calendar people can actually book in. Your existing site is the foundation. GEO adds structure and substance to it rather than replacing it, and Google's own guidance is explicit that no AI-specific files or markup are required.
How would I even know whether an assistant is recommending us?
Ask it. Put your customers' real questions, with your town in them, to ChatGPT, Gemini and Perplexity, for example "reliable shop for AC service in your city," and check both whether you appear and whether the details are correct. Do it monthly, signed out, and keep the answers so you can see movement. Alongside that, watch the schedule: bays filling earlier in the week with no other explanation is the outcome the sampling is only a proxy for.
Read on