DealerSpark
A free working guide for franchised dealers
What artificial intelligence actually is right now, where it genuinely changes dealership work, how to avoid getting fleeced by vendors — and a 90-day plan to put it to work in your store.
How to use this guide: read Parts 1 and 2 this week and actually do the exercises — that's the fast-track. Parts 3–6 are reference material for when you're evaluating a tool, writing your staff policy, or sitting across from a vendor. Part 7 is your store's plan; Part 8 is a framework to bring to your state association. Part 9 pops the hood on AI agents for those who want to go deeper, and Part 10 answers the questions every dealer asks. Part 11 is the shelf at the end: a starter tool stack, a geekier look at the engines, and where to keep learning. Nothing here requires technical background.
Print-and-pass-around pages: three one-page PDFs pulled straight from this guide, made for a managers' meeting or a 20-group binder.
You're reading the single-page version. The same guide is also published one part per page, which is easier to share and bookmark: start at the guide's front page.
Part 1
Forget robots and science fiction. The AI that matters to your store in 2026 is four things, and you can understand all four — including why they fail — without a single line of code.
A large language model (the technology behind ChatGPT, Claude, and Gemini) is a prediction machine. It read a staggering share of everything humans have written — books, websites, manuals, forums — and learned, statistically, what word tends to come next in any given context. When you ask it something, it isn't looking up an answer in a database; it's composing an answer, word by word, from those learned patterns.
Here's the showroom version: it's like a 20-year veteran who has worked every desk in the store and read every trade publication ever printed — but who woke up this morning with no memory of your store, your customers, or today's date. Brilliant on patterns, blank on specifics. Everything it knows about your situation, you have to tell it in the conversation.
Two consequences fall straight out of that, and they explain nearly every AI success and failure you'll see:
ChatGPT, Claude, Gemini — general-purpose assistants you type (or talk) to. This is where you personally start, because it's $20 a month and pays for itself the first week. Part 2 is a full 30-day curriculum for exactly this.
Your CRM, DMS, phone system, and website platform are all shipping AI features: call summaries, suggested replies, lead scoring, appointment booking. Some genuinely work. Some are a checkbox on a renewal pitch. Part 5 gives you the interrogation kit.
Buyers increasingly ask AI engines the questions they used to type into Google, and the AI answers by citing a handful of sources. Either your store is one of them or it isn't. Part 4 covers this in depth, because it's the area where early movers can win their market outright.
The first three are AI you choose. This one walks in the door whether you choose it or not: buyers now use the same assistants to research vehicles, sanity-check your price, decode F&I products, and draft their side of the negotiation. A few have AI agents emailing a dozen stores for quotes on their behalf. Part 3 covers what that looks like on your floor and how to sell to a customer who has a coach in their pocket.
Dealers ask this constantly, and the honest answer is: it matters far less than just picking one and using it daily. Still, here's the field as of late 2026:
| Tool | Made by | Where it shines for a dealer |
|---|---|---|
| ChatGPT | OpenAI | The most widely used; strong all-rounder; your staff has probably already tried it. |
| Claude | Anthropic | Excellent with long documents (contracts, OEM programs, regulations) and produces writing that sounds less like AI. |
| Gemini | Built into Google Workspace — if your store runs on Gmail/Docs, it's already in your tools. | |
| Copilot | Microsoft | Built into Microsoft 365 — same logic if you live in Outlook and Excel. |
| Perplexity | Perplexity | AI search with visible citations — useful for research, and one of the engines you'll audit your own visibility on in Part 4. |
Pick one, buy the roughly-$20/month paid tier (free tiers are throttled and weaker), and commit to using it every working day for a month. That's the entire decision.
The one mental model to keep: today's AI is a very fast, very well-read assistant with no judgment about your business and no accountability. It multiplies the person using it. It replaces nobody's signature — the dealer principal still owns the deal, the ad, and the disclosure. And when you evaluate any of it, think like an employer, not a software buyer: don't ask what features it has — ask what job you'd hire it to do. Part 5 turns that instinct into a working test.
Part 2
You will not learn AI from a course, a conference keynote, or this guide. You'll learn it by pointing it at work you already have to do, every day, for a month. Here's the curriculum.
Long prompt-engineering guides are mostly noise. Three habits cover 90% of it:
The paid assistants have grown well past a blank chat box. Three features, available in some form in ChatGPT, Claude, and Gemini, are worth an afternoon in your second or third week — the names differ by product, the ideas don't:
Eight working prompts, one per chair in the store. Replace the bracketed parts, paste the relevant material where indicated, and adjust the output until it sounds like you.
Dealer principal / GM
Dealer principal — reputation
Sales manager
Sales manager — training
Worth naming what this prompt is: one rep, coaching themselves, once. That's the single-player version of AI role-play. The team version — every rep practicing daily against voiced customers, with scoring a manager can actually see — is the "sales training" row in Part 3, and per the disclosure there, that one's ours.
Service manager
Service manager — declined work
Marketing / BDC
Office / HR
One rule before you start: don't paste customer names, SSNs, credit applications, or deal jackets into a public AI tool. Use real work, not customer data. Part 6 has the full traffic-light table for what's safe to share where.
Part 3
Every department has an AI pitch aimed at it right now. Here's an honest map of what's delivering in stores today versus what's still mostly a demo — followed by what to pilot, what to measure, and roughly what it costs. Read every row through the employer's lens from Part 1: not "what are the features," but "what job would I be hiring this to do — and can it do the job?"
| Department | What AI does today | Status |
|---|---|---|
| BDC / lead handling | Instant, around-the-clock first response to internet leads; appointment setting; follow-up sequences that don't go stale. The best tools hand off to a human the moment the conversation gets real. | Delivering |
| Service drive | AI phone agents that book appointments after hours and on overflow, automatic RO summaries, predictive maintenance reminders based on mileage and history. Every missed service call is measurable money. | Delivering |
| Used-car pricing & appraisal | Market-based pricing tools have used machine learning for years — the most mature category. Newer tools add condition assessment from photos. Trust the data, keep a human on the final number. | Delivering |
| Marketing | Ad copy, email campaigns, social posts, and video scripts drafted in minutes instead of days. Quality depends entirely on the brief. Watch for AI-generated sameness — your voice is the differentiator. | Delivering |
| Sales training | AI voice role-play — reps practice objections, phone-ups, and closes against a simulated customer that talks back, with feedback on every rep, every day, instead of one workshop a quarter. Many reps are already coaching themselves with generic ChatGPT advice (Part 6 covers why that's a signal, not a solution). (Disclosure: this is DealerSpark's own product line — live AI voice coaching for dealerships, running in stores today — we're citing the category because it works, and you should hold us to the Part 5 questions like anyone else.) | Delivering |
| AI search visibility (GEO/AEO) | Measuring who AI engines cite and recommend in your market, then working the fixes — structured data, crawler access, quotable content, third-party profiles. The best platforms pair the scoreboard with a built-in AI agent that walks your team through each fix, so the store learns how AI works by using it. (Disclosure: this is DealerSpark's GEO platform, live in stores today — and the feedback dealers give most isn't about the scoreboard, it's that working alongside the agent is how their team finally understood AI. Part 4 covers the whole discipline, including the DIY version.) | Delivering |
| "Agentic" suites from your DMS / CRM | The big platform vendors spent 2026 announcing AI "agents" by job title — a salesperson agent, an F&I agent, an accounts-payable agent — bundled into the systems you already run. Some will be genuinely good, because they sit on your real data. But a bundle isn't a verdict: evaluate each agent as its own hire (Part 5's operator's test), pilot one at a time against a baseline, and ask for the tool list and guardrails in writing (Part 9). "It's included in your renewal" is a price, not a result. | Emerging |
| Back office / accounting | Document handling — invoice matching, schedule reconciliation prep, drafting collection letters, summarizing statements. Useful as an assistant to your office manager; nowhere near replacing controls or review. | Emerging |
| F&I | Menu-presentation training and objection practice are useful. Anything touching credit decisions, deal structure, or disclosures needs compliance review first — this is where AI mistakes get expensive. | Proceed carefully |
| "Fully autonomous AI salesperson" | The pitch that AI closes deals end-to-end without staff. Customers notice, trust drops, and when it misquotes a price you own the quote. Not there — maybe not ever for franchise retail. | Mostly hype |
The pattern: AI delivers wherever the job is fast response, repetitive communication, or pattern-finding in data. It's hype wherever the pitch is "remove the human from a trust decision." Car buying is still a trust decision.
Everything in that table is AI you deploy. The shift that arrived faster than most stores expected is the AI your customer deploys. Cox Automotive's 2026 AI in Auto Retail Tracker has found, wave after wave, that about six in ten in-market shoppers say they are using or will use AI to shop for their next vehicle, with AI tools approaching parity with the shopping sites as a research source. The same study found 82% of dealers already use AI somewhere in the store — but industry coverage of it reports only about 29% have adjusted how they sell to an AI-equipped shopper. That gap is the opportunity. Here's what it looks like on the floor:
What to do about it — none of which requires buying anything:
This is the cleanest AI pilot in the industry because the baseline is zero — a call that rings out at 7pm books nothing. An AI phone agent that answers, checks the schedule, and books the appointment converts pure loss into ROs. The math is illustrative but easy: if your store misses even 5 bookable calls a week and your average customer-pay RO runs a few hundred dollars, recovering half of them pays for a tool costing several hundred to $1,500 a month many times over. What to measure: after-hours/overflow calls answered, appointments booked from them, show rate, RO value. Your phone system already logs the "before."
Speed-to-lead has been preached for twenty years; AI finally makes sub-minute response at 9pm on Sunday real. What to measure: median response time before/after, contact rate, appointment-set rate, and — important — whether customers are being annoyed (read the transcripts weekly; you'll learn a lot about your tool). Typical cost: mid-hundreds to low-thousands per month depending on volume. Insist on the pilot terms in Part 5.
Sales training got one row in the table above, and it deserves the same treatment as the two pilots — because this category is filling up fast and the products are less alike than their demos suggest. Five things to pin down before you sign with anyone, in the order they matter. (Disclosure, again: DealerSpark sells in this category. The checklist is written so you can run it against us as hard as against anyone else.)
What to measure in a training pilot: practice sessions per rep per week (usage is the first thing that fails), score movement on two or three named skills, and one business number you already track, such as appointment-show rate, closing percentage on a specific objection, or F&I product penetration. Write the baseline down first, same as any other pilot.
For a look at how the current products differ, DealerSpark keeps a running comparison: the AI sales and F&I coaching platforms for car dealerships in 2026. It's our post, so read it as a vendor's view and check it against this list.
Pricing honesty: the ranges above are typical of late 2026 and shift constantly. Treat any number in this guide as a sanity check, not a quote — and treat any vendor who won't give you a simple all-in monthly number as a red flag (Part 5).
Part 4
This is the shift most dealers haven't seen coming, and the one where early movers can win their market outright.
For twenty years the playbook was Google: rank in the map pack, buy the ads, manage the reviews. That still matters — but a growing share of buyers now start by asking an AI a question and reading a single synthesized answer instead of ten blue links. Google itself now puts an AI-written answer above the links for many searches. The AI builds that answer by citing a handful of sources it trusts.
When someone asks "best place to buy a used truck near Billings," the engine runs a search behind the scenes, reads the pages it finds, and composes an answer from the ones that are readable (its crawler isn't blocked), legible (the page states facts plainly — structured data, real prices, real hours, actual answers), and corroborated (other sources it trusts say consistent things about you). Notice what's not on that list: your ad budget. Ads now exist inside AI answers (more on that below), but they sit beside the answer, labeled — you cannot buy your way into the citation itself the way you buy the top of Google. You earn it with your website and your footprint — which is precisely why a prepared single-point dealer in Kalispell can beat a lazy group store.
DealerSpark measures this weekly in live dealer markets — real buyer-intent questions probed across ChatGPT, Perplexity, and Gemini, with every cited source captured. The pattern is stark: third-party sites — Cars.com, Edmunds, CarGurus, Reddit, Carfax, local news — capture the overwhelming majority of citations, and individual dealership websites are cited in the low single digits of percent. The AI usually knows dealers exist (stores get mentioned by name far more often than cited), but it builds its answers from everyone's website except the dealer's. We've also found dealer website platforms whose bot protection was silently blocking AI crawlers entirely — the store was invisible to AI engines and nobody knew.
Why this is an opportunity, not just a threat: almost no dealer is doing anything about this yet. In a small or mid-size market, the store that makes its website legible to AI engines can become the source that gets cited for its whole market while competitors are still optimizing for 2019 Google. And the results are graded, not binary — stores go from absent, to mentioned-by-name, to cited — so progress is visible month over month.
New since this guide first went out: OpenAI began showing ads in ChatGPT in 2026 — on the free and lowest-cost tiers, with paid plans staying ad-free — and has opened a self-serve ad platform that local businesses can buy on. Google runs ads in and around its AI answers too. Expect your agency or ad rep to pitch both. Two things to hold onto:
This area — called GEO (generative engine optimization) or AEO (answer engine optimization); same discipline, two names — is the part of the AI landscape most specific to auto retail and the least served by existing vendors. It's also measurable, which makes it a natural topic for the association conversation in Part 8: member visibility across a state's markets, tracked quarter over quarter.
One more thing worth knowing about how this work gets done well: visibility platforms worth having don't just hand you a report — they pair the measurement with an AI agent that walks your team through each fix, explains why it matters, and executes the pieces you approve. That structure has a second-order benefit dealers consistently report: it's the best AI training a store can get, because the team learns how AI works by working with one on their own business. (This is DealerSpark's GEO platform — live in stores today, and where the citation numbers above come from. The DIY audit earlier in this part is the honest starter version of the same method — the difference is dose and rigor. The DIY version is 8 questions, two engines, once. The platform builds its question set from your store's profile and market area, runs roughly a thousand buyer-intent questions — long tail included — across three frontier engines every week, captures every cited source, reads what the engines actually say about your reputation, and tracks each fix to done. Run the DIY version this week regardless; it's how you learn what the full instrument is measuring.)
A promise you should demand from anyone in this space, including us: nobody can guarantee an AI engine will cite you — the engines sell labeled ads now, but they don't sell the citation. What can be promised is the work (measurement, fixes, tickets to your website vendor) and honest reporting of movement. Anyone guaranteeing "#1 in ChatGPT" is selling you the 2026 version of "#1 on Google" snake oil.
Part 5
Every vendor deck you see this year will say AI on it. These eight questions separate the tools from the stickers. A good vendor answers all eight crisply; a repackaged vendor gets vague by question three.
If the answer is a feature list you had in 2022 with "AI-powered" in front of it, it's a sticker.
Canned demos hide everything. Any serious vendor can run a pilot on your real leads or ROs.
Every AI system makes mistakes. A trustworthy vendor has an error story: escalation to humans, review queues, correction logs. "It doesn't make mistakes" ends the meeting.
Get it in writing. Your customer and pricing data should not be improving a product your competitor buys.
Undisclosed AI conversations are a trust risk and, increasingly, a legal one. You want disclosure built in.
"Engagement" is not a result. Appointments set, calls answered, hours saved — numbers you can audit against your own systems.
Spoiler: you are — regulators treat claims made by your tools as your claims. So the real question is what the vendor does to prevent it and how they indemnify you.
Month-to-month after a pilot, your data exportable, no hostage-taking. AI tools are evolving too fast for three-year lock-ins.
Print it: the eight questions and five contract terms on one page (PDF). Bring it to the next vendor meeting.
The eight questions interrogate the vendor. Once a tool survives them, switch lenses entirely: stop evaluating it like software and start evaluating it like labor. This isn't a slogan — it's how we run our own companies. Before DealerSpark sold anything to a dealer, our founder spent three years building an AI workforce inside a nationwide collateral-recovery firm — recorded calls, sensitive financial data, lender compliance, real consequences for mistakes. That company runs about 14 human employees today, with roughly two dozen AI employees working alongside them. Some have big jobs, some have tiny part-time jobs — and every one of them is managed with the same six questions you'd ask about a human hire:
| Question | What you're asking |
|---|---|
| Job | What, exactly, is it responsible for? (The narrower the job, the better the work — no human runs the whole store either.) |
| Quality | Is the work good enough for this job? Perfect is the wrong standard — you've never managed a perfect human. Reliable at the level the job requires is the standard. |
| Output | How much work actually gets done? Count it the way you'd count a person's. |
| Supervision | How much oversight does it need? Higher-risk job = tighter controls and more human review (Parts 6 and 9). |
| Wages | What are you paying it, all-in, per month? |
| ROI | What is the work worth against those wages? |
Notice what's not on the list: parameters, benchmarks, model names — the things AI vendors love to talk about. You're not hiring artificial intelligence. You're hiring somebody to do a job, and the whole evaluation comes down to one question: can it do the job? This lens works whether you're buying a vendor's agent, building your own (Part 9), or deciding whether a $20 assistant seat is earning its keep.
Here's the operator's test as a worksheet. Pick one job you're thinking of handing to an AI tool, fill in your own numbers, and it works out what you're paying per unit of usable work and whether the hire earns its wages. The starting numbers are an illustration (an after-hours service-call agent), not a benchmark. Replace every one of them.
Everything is per month unless it says otherwise. Nothing you type here is sent anywhere; the math runs in your browser.
6 · ROI
How it figures: AI total cost is its wages plus the supervision time (hours per week × 4.33 weeks × the reviewer's hourly cost). Good units are output × the quality share. Net is the value of the good units minus the AI's total cost. The human figure is there for comparison only; for a lot of AI jobs the honest baseline isn't a person, it's the work not getting done at all, which is why the value-per-unit line matters more than the wage comparison.
Part 6
None of this requires a law degree, but it does require house rules before your staff starts pasting things into chatbots — because they already are, whether you've sanctioned it or not.
Before the rules, an uncomfortable inventory. Walk any dealership today and some of this is going on, on personal phones and free accounts, with the best of intentions and zero oversight:
None of this makes your people bad actors — it makes them early adopters with no rules. The wrong response is a ban, which just drives it further underground. The right response is the rest of this part: clear rules, paid accounts with data protections for the sanctioned uses, and a one-page policy that tells everyone where the lines are. You can't govern what you pretend isn't happening.
About that self-serve sales training: a rep asking ChatGPT for closing advice is showing you exactly the hunger for coaching your store isn't feeding. But what a generic assistant returns is the averaged advice of every sales book ever written — with none of your process, your inventory, your market, or your word tracks in it. It isn't wrong; it's generic, and it's identical to what the rep across town is getting from the same prompt. The fix isn't to stop them — it's to give the practice structure: the role-play prompt in Part 2 grounded in your store's context, and when you want every rep practicing daily with scoring a manager can see, the platform version in Part 3.
The FTC's position is simple and dealers should tattoo it somewhere: there is no AI exemption from consumer-protection law. A deceptive price claim, a fake review, an invented offer — the FTC treats an AI-generated one exactly like a human-written one, and it has brought enforcement actions against companies both for deceptive claims made by AI tools and for deceptive claims about what their AI could do. "The chatbot said it, not us" is not a defense. Your tools' claims are your claims.
The Safeguards Rule already covers you. Dealerships that arrange financing are financial institutions under GLBA, and customer financial data is exactly what must not leak into unsecured tools. An employee pasting a credit application into a free chatbot is a data-handling incident, full stop. Your written information security program should name AI tools explicitly.
Your state may have its own privacy law. Around twenty states now have a comprehensive consumer privacy law, and the details differ in ways that matter to a dealer: how many consumers' data you must handle before the law applies, whether you get a grace period to fix a violation, the penalty per violation, and whether the financial-institution exemption covers your whole business or only the GLBA-covered data. The trend is toward tighter. Montana is a useful example: its law took effect in October 2024, and amendments effective October 2025 cut the threshold in half (to 25,000 consumers), ended the cure period, added civil penalties of up to $7,500 per violation, and narrowed the exemption to GLBA-covered data. Translation for a dealer in a state like that: your credit and deal data is still GLBA territory, but marketing lists, website tracking, and AI chat logs likely are not — and more stores, especially groups, clear the threshold than did a year ago. Whether your state has a law, and where you sit under it, is a thirty-minute conversation with your attorney or your state association's counsel (see Part 8).
Calls and texts have their own tripwire. The FCC has confirmed that AI-generated voices in outbound calls fall under the TCPA's robocall rules — consent requirements apply. Any AI tool that calls or texts customers outbound needs its consent story checked before you turn it on, not after. The opt-out side moved in 2026 too: the FCC rewrote its consent-revocation rules, effective November 25, 2026, changing how a customer's "stop" must be honored across your calling and texting. A separate FCC proposal to require AI disclosure at the start of a call is still pending, not final. You don't need to master the details — you need every vendor that calls or texts for you to explain, in writing, how they capture consent and honor opt-outs under the current rules.
Your OEM has opinions too. Franchise agreements, brand standards, and co-op advertising rules all apply to AI-generated content exactly as they do to human-made content — and OEMs are starting to issue AI-specific guidance. Loop in your OEM rep before deploying anything customer-facing that carries the brand.
Who writes the AI rules is still being fought over. A December 2025 executive order pushed for one national AI framework and directed federal agencies to challenge state AI laws; Congress has debated overriding those laws but, as of this update, has not passed anything that does. So state laws remain in force and the picture will keep shifting. Don't wait for it to settle: the four rules below hold up under any outcome, because they're built on things that aren't in dispute — protect customer data, don't deceive, disclose, verify.
| Data | Public chatbot (free account) | Paid business AI account | Contracted vendor tool |
|---|---|---|---|
| Marketing copy, inventory descriptions, public info | Fine | Fine | Fine |
| Internal docs: schedules, meeting notes, aggregated numbers | Avoid | Fine | Fine |
| Customer names + contact info | Never | Only if your data terms cover it | If contracted for it |
| Credit apps, SSNs, deal jackets, DMS exports | Never | Never | Only under a security-reviewed contract |
Print it: the staff AI policy as a one-page PDF, with the signature line.
Fill in the brackets, have your attorney spend thirty minutes on it, and put it in front of every employee. Done is better than perfect: the risk isn't an imperfect policy, it's the six employees already using free chatbots with no rules at all.
I have read and understood this policy. Name ______________________ Date ____________
Part 7
Committed beats sophisticated. Here's the sequence for a dealer principal starting from zero — each step small enough to actually happen.
Print it: the 30/90-day roadmap as a one-page PDF.
You use AI daily and can tell a real tool from a sticker. Two managers run their departments with it. One pilot has a verdict backed by your own before/after numbers. Your staff has signed a policy. You know exactly who AI engines cite in your market and whether it's you. Total spend: a few hundred dollars in subscriptions plus one pilot. That's a fast-track — and it compounds from there.
Part 8
Some of this work is wasteful to repeat forty times in forty stores — that much is obvious. But how an association acts is a decision for its board, its members, and its counsel, not for a vendor writing a guide. So this part is a framework, not a playbook: five models for making AI knowledge a shared asset, offered as discussion material for a board meeting. What to adopt, how to structure it, and where the legal lines sit are the association's calls to make.
Read this part with counsel in mind: trade associations operate under antitrust rules that don't apply to individual stores — programs involving shared competitive information or any group engagement with vendors are exactly what association counsel exists to structure, and every established association already runs that discipline on everything from surveys to standards. Nothing below is novel in kind; each model has decades of precedent in association work. The counsel review isn't a hurdle for these ideas — it's part of the design.
Every member interrogating the same vendors alone is pure waste. One model: a members-only folder where a store that pilots an AI tool can file a one-pager — what it was, the baseline, what moved, would you keep it. Ten honest one-pagers beat every conference booth in existence. The structure matters and is standard association practice: counsel shapes the template, contributions are individual experience and measured results, and every store's vendor decisions remain entirely its own.
Cost: near zero. Value: members stop paying tuition on each other's mistakes.
The Part 4 audit, run across the state's main markets on a fixed question set, quarterly, published to members: who's cited, who's mentioned, who's absent, which third-party sites dominate each market. It's built on public data — anyone can ask ChatGPT the same questions — and watching a rival climb the board will do more to move members than any seminar.
Measurable, repeatable, and specific to your state — the kind of benefit no national player is offering. (It's also the work DealerSpark does commercially — and rigor is what separates a snapshot members will act on from a science-fair project: the same question set every quarter, sources captured, every market scored the same way. The DIY method can prove the appetite; a real program is a decision for the association to make on its own terms.)
The Part 6 policy and the state-law questions, taken to association counsel once, with the reviewed version made available to every member. Forty stores get governance for the price of one legal consult — and a model policy distributed as a resource, adopted store by store, is bread-and-butter association work.
Also the fastest possible member win — deliverable in a month.
No vendor keynotes. Members present their own pilots — the numbers, the surprises, the one they killed. Add a live demo of the visibility audit on one of the state's markets, on screen, so every dealer watches the AI recommend (or ignore) stores they know.
Peer evidence converts skeptics that presentations never will.
Much of Part 4 is gated behind website platforms dealers don't control, and a lone store's support ticket is easy to answer vaguely. One model that keeps every decision in the store: the association publishes a well-formed technical question set — "which AI crawlers can access member sites, what does the robots.txt allow, what structured data can be deployed and how" — that each member sends its own provider, individually, on its own letterhead. Forty identical, well-informed tickets are hard to answer vaguely; collecting the answers back into member information is classic association work. The same information-gathering lens applies to OEM AI guidance as it emerges. Anything beyond that — engaging providers as a group — is precisely the kind of step an association designs with counsel or doesn't take.
The leverage here is shared knowledge and a common vocabulary — each rooftop still acts alone.
If a board adopted some version of this: quarter one, model policy reviewed and made available; a first visibility snapshot. Quarter two, the experience library seeded with the first pilot one-pagers; the platform question set published. Quarter three, a convention session taught from member results; snapshot two shows the first trend lines. Quarter four, every member has access to a policy, a baseline, and peer-tested vendor intelligence. Which pieces, in what order, under what structure — that's the board's conversation; this framework is just meant to make it a short one.
Part 9
Everything up to here you can use without knowing how the machinery works. This part is for the dealer who wants to go deeper — because you're evaluating a serious deployment, because a vendor is throwing the word "agent" at you, or because you're the kind of operator who pops the hood. Plain English throughout, and it ends with a working example you can poke at right now.
A chatbot answers. An agent does. The difference is that an agent can take actions — check a schedule, book the appointment, send the confirmation text, create the CRM task — and string those actions together to finish a job, not just describe one. When your service AI answers a call at 9pm, looks up Thursday's open slots, books 7:30am, and texts the customer a confirmation, that's an agent running a loop: listen, decide, act, check the result, act again.
That loop is also why agents deserve more scrutiny than chatbots. A chatbot that's wrong says something dumb. An agent that's wrong does something dumb — books the wrong day, quotes the wrong price, texts the wrong customer. Capability and risk arrive together, which is why half of this part is about guardrails.
Every AI agent you will ever be pitched — voice, chat, or invisible back-office — is built from the same six components. Learn these and no demo will ever snow you again. Notice how naturally they read as an employee file — a job description, granted keys, supervision rules: an agent is Part 5's hire-it-don't-buy-it idea made literal.
| Component | What it is | The dealership translation |
|---|---|---|
| Engine (the LLM) | The rented prediction engine from Part 1 — OpenAI, Anthropic, Google, or a budget engine for simpler jobs. | The Cummins under the hood. The vendor almost never built it, and doesn't need to have — judge the truck, not the engine. |
| Instructions (the system prompt) | A written job description the agent reads before every single conversation: who it is, what it does, what it refuses, how it talks. | Your best walkaround script and your employee handbook, compressed into one page the agent can't ignore. Ask a vendor: "Can I read the instructions my agent runs on?" Good ones say yes. |
| Knowledge base | Your facts — hours, services, pricing menus, policies, inventory — stored where the agent looks them up at answer time (the trade calls this retrieval, or RAG). | The difference between a new hire guessing and a new hire checking the board. Retrieval beats "we trained it on your data": facts stay current (edit the document, the agent updates), and you can see exactly what it's working from. |
| Tools | The specific actions it's allowed to take: read the schedule, create an appointment, send a text, look up a part. Each tool is granted, not assumed. | The keys you hand it. An agent with read-the-schedule can't double-book you; an agent with send-any-text can embarrass you. The tool list IS the risk list — ask for it in writing. |
| Memory | What it retains: this conversation only (session memory), or history across visits ("you were in for brakes in March"). | Cross-visit memory is powerful and it is customer data — everything in Part 6 applies. Ask where it's stored, who can see it, and how a customer gets it deleted. |
| Guardrails | The limits wrapped around all of the above. | Next section — this one gets its own. |
Guardrails aren't training wheels — they're brakes. Brakes are what let a truck go fast. An agent with no guardrails isn't more capable, it's just uninsurable. Here's the full set; every serious deployment should have most of these, and a vendor should be able to tell you which ones theirs has without checking with engineering:
A voice agent — the thing answering your service line — is a chat agent wearing a headset. Three extra parts are bolted on: speech-to-text (ears), the same LLM loop in the middle (brain), and text-to-speech (voice). That pipeline runs in real time, which raises the bar in ways a demo can hide:
When an agent gives a wrong answer about your store, the cause is almost never the engine — it's the knowledge base: stale hours, an outdated menu price, a policy nobody wrote down. Two rules from the trade, both boring, both decisive: garbage in, gospel out — the agent states whatever it's given with total confidence, so feed it only what's true (start small: hours, services, pricing menus, policies, directions, current specials — the 40 questions your BDC answers every week); and every knowledge base needs an owner — one named person who updates it when the shuttle policy changes, the same way someone owns the website today. An agent reading last quarter's service specials isn't an AI problem; it's a nobody-owned-the-update problem.
The chat bubble on this guide — Spark — is a live agent built exactly the way this part describes, and we'll open the hood as a demonstration: the engine is rented (an inexpensive one, deliberately — this job is answering questions from one guide, not writing legal briefs; matching engine cost to the job is itself a lesson). The instructions scope it to dealership AI and tell it to decline everything else — ask it to write you a Python script or do your kid's homework and watch the refusal. The knowledge base is this guide. The tool list is empty — it can't book, text, or spend, because its job doesn't need those keys. The guardrails: reply-length caps, per-visitor rate limits, a daily ceiling, no PII, disclosure in its header, and standing orders to never guarantee AI-search rankings — try to get it to promise you "#1 in ChatGPT" and see what it says. That's not a diagram in a vendor deck; it's running right now, and you're welcome to red-team it yourself. When a vendor pitches you an agent, ask them to open the hood the same way. The ones who will are the ones to shortlist.
For most stores, most of the time: buy — a vendor's agent for a bounded job (service calls, lead response), interrogated with Part 5 and audited against this part's anatomy. Build — realistically, have someone build for you — when the job is specific to how you operate and the off-the-shelf options don't fit; a scoped agent like Spark is days of work for someone competent, not months, and you own every word of the instructions. Wait is the right call only for high-stakes surfaces like F&I and anything touching credit — and even there, "wait" means "pilot the safe stuff now so you're fluent when the high-stakes tools mature," not "check back in 2028."
Part 10
Asked in every 20 group and every association meeting. Answered without hedging.
No — and the vendors implying it will are the ones to walk away from. A vehicle purchase is a trust decision, and customers still want a person accountable for it. What AI does replace is the drudgery around selling: the 9pm lead response, the follow-up nobody sent, the fifth rewrite of the same email. The realistic risk to your people isn't AI — it's salespeople and managers who use AI outworking the ones who don't. The future isn't a dealership without people; it's great dealership people with an AI workforce behind them.
By being the store whose numbers hold up. A buyer coached by AI is a buyer who did homework — that's a serious customer, not an adversary. Give one honest out-the-door figure in writing, ask to see what the AI told them, and correct it with a source where it's wrong (it often is on incentives and local inventory). The stores getting hurt are the ones whose process relied on the customer not knowing something. Part 3 has the full playbook, including an exercise that lets your reps read the other side's script.
Maybe, later, as a measured test — not as a substitute for being cited. The ads are labeled and sit apart from the answer; the AI's actual recommendation still comes from sources it trusts, and no ad buys that. Fix what the engines read about you first (Part 4), then pilot ads with a small budget and a named baseline like anything else.
More than you'd think, and the direction only goes one way. AI answers are already embedded in ordinary Google results — your customers are reading AI-generated summaries whether they know it or not. And your next generation of buyers uses these tools as a matter of course. The rural angle actually favors you: fewer competitors in your market means fewer stores for the AI to cite — being the one prepared dealer in a small market is far easier than being one of forty in a big metro.
Fair question, wrong pattern. The fads never had daily users doing real work; AI assistants do — ChatGPT alone reported roughly 900 million weekly users in early 2026 and has reportedly passed a billion since — and the technology is being wired into the phones, search engines, and office software your customers and staff already use. The honest caveat: plenty of individual AI products are absolutely fads — that's what Part 5 is for. The technology is staying; many of the vendors won't.
Less than one bad hire. Personally: ~$20/month. Two deputized managers: ~$50/month. Your first department pilot: several hundred to $1,500/month, against a measurable number, cancelable if it doesn't move. Visibility work: the baseline audit is free and takes 45 minutes (Part 4). A platform is a different instrument, not a convenience — instead of 8 questions in a spreadsheet, it builds the question set from your store's actual profile and market, runs on the order of a thousand of them (including the long-tail questions nobody would type by hand) across three frontier engines, logs every answer and source, and tracks it week over week. Do the free version first either way; it's how you'll know whether any paid version is measuring something real. The expensive path is the other one — multi-year contracts signed at a trade show for tools nobody measured.
The tools will be better and cheaper in two years — that part of waiting is real. What you can't get back: the AI-visibility ground rivals claim in your market (citations compound — engines keep citing who they've cited), two years of staff fluency, and two years of pilots that told you what works in your store. Waiting doesn't pause the race; it just runs it without you. The good news: in most markets, almost no dealer has started. That advantage has a shelf life.
Trust it the way you trust a bright new hire: excellent first drafts, real insight, occasional confident nonsense. Verify anything that matters — especially numbers, laws, and anything going in front of a customer. The stores that get burned aren't the ones using AI; they're the ones using it without review.
OEMs are moving too — several have AI programs and approved-vendor lists in motion, and brand standards apply to AI-made content like anything else. Practical rule: internal uses (Part 2) are your business; customer-facing, brand-carrying deployments deserve a heads-up to your rep. And when OEM AI mandates arrive — they will — the dealers who already have baselines and policies will negotiate from knowledge instead of fear. That's also an association conversation (Part 8, item 5).
In this order: your own 30 days of hands-on use (nothing inoculates against hype like fluency), your 20 group and fellow association members' pilot results, association resources from Part 8, and then vendors — interrogated per Part 5. Anyone who leads with a contract instead of a baseline measurement is selling, not helping. That standard applies to DealerSpark too; hold us to it. We wrote the Part 5 gauntlet because we're happy to sit through it — pilot against a named baseline, month-to-month terms, your data exportable, and we'll open the hood on the agent the way Part 9 describes. Anyone can print the questions; shortlist whoever volunteers the answers.
Part 11
Everything before this part you can do with one chat subscription. This part is for two readers: the dealer who wants a sane starter package, and the resourceful person in your store who's already building things — Part 6 told you they exist; this is their sanctioned path. Tools are named because vagueness helps nobody, but the categories matter more than the brands: names will change, the shape of the stack won't.
| Tool | What it does | The data rule |
|---|---|---|
| Paid assistant seats (ChatGPT, Claude, or Gemini — Part 1's table) | The workhorse for everything in Part 2. Business-tier accounts, not personal free ones — the data protections are the point, not the features. | Follow the traffic-light table in Part 6. Business tier moves you from the "never" column to "only if your data terms cover it." |
| An AI meeting note-taker (Fathom, Otter, or the one already built into Teams/Zoom) | Manager meetings, vendor calls, 20-group prep — recorded, summarized, action items extracted. The fastest "where did my admin time go" win after the assistant itself. | It's a recording device: tell attendees, and keep it out of any meeting where customer deals are discussed by name. |
| A document notebook (Google's Gemini Notebook — the tool known as NotebookLM until its July 2026 rename — is the current standard) | Load your OEM program bulletins, franchise agreement, and policy manuals; ask questions and get answers cited to the page. Retrieval from Part 9, productized — it only answers from what you gave it. | Internal documents, fine. Customer files, no. |
Total for a dealer principal and two deputized managers: on the order of $100 a month. That's the whole Tier 1 decision.
Automation platforms connect your tools so multi-step chores run themselves: a lead arrives → enrich it → draft the reply → post to the CRM → ping the manager. Two honest flavors:
The rule that keeps Tier 2 out of trouble: an automation is an agent with tools — it acts without asking. Everything in Parts 6 and 9 applies: name an owner, log what it does, gate anything customer-facing behind a human, and inventory these the way you'd inventory keys. An automation nobody remembers building, quietly running on a personal account, is the vibe-coded side tool from Part 6 wearing a nicer shirt.
Part 6 warned about vibe-coded side tools on personal laptops. Here's what the sanctioned version looks like. AI coding tools — Cursor (an editor with AI built in) and Claude Code (an agent that writes and runs code from plain-English instructions) — have made real internal software a days-not-months proposition for one capable person: the aged-inventory report nobody will build for you, the DMS-export cleaner, a scoped agent like Spark (Part 9). The difference between an asset and a liability is entirely procedural: it lives on store-controlled hardware and accounts, someone besides the builder knows it exists, anything touching customer data gets the Part 6 rules, and it has an owner — the same one-named-person rule as the knowledge base in Part 9. Handled that way, the employee who was your shadow-IT risk becomes your most valuable technical asset. Same person, same skills; the only variable is whether you gave them a sanctioned place to work.
Part 1 gave you the engine analogy; here's the next layer down, because Tier 2 and 3 work makes you choose engines directly:
Ranked the same way Part 10 ranks help — closest to hands-on first:
What's deliberately not on that list: paid "AI for business" masterclasses and social-media AI gurus. The field moves too fast for courses to stay current, and "where do I learn this" is precisely where the fleecing happens — the same instinct that sells AI stickers (Part 5) sells AI seminars. Everything worth knowing at the dealer level is free or already in your dues.
Last thought on the stack: keep it boring. The stores winning with AI don't have the most tools — they have three or four, owned and measured, with real work routed through them daily. The tool collection is the hobby; the routed work is the fast-track.
This guide is reviewed every quarter and revised whenever the ground shifts. Each revision is listed here.