Visible where shoppers ask.

1. Foundation: the entity

• One entity string. Name, legal entity, address, phone, email and domain, byte-identical on the site, in the schema, on Google Business Profile and in every directory the engines read.
• NAP audit. Every place the store's details appear, compared and corrected. One mismatched suite number splits a dealership into two entities.
• Google Business Profile. Sales, service and parts profiles claimed, hours and pin correct, inventory, photos and posts kept current, held to the same string.
• Crawler access, verified. The AI crawlers from OpenAI, Anthropic, Perplexity and Google allowed in and confirmed by IP-checked logs, not by a user-agent string.

2. The site the engines can read

• Answer-first pages. The first sentence says who, where and what, because that sentence is most of what an engine reads.
• Schema that says what things are. AutoDealer for the rooftop, Product, Car and Offer with VIN and payment on every vehicle, FAQ on the answer pages.
• Technical hygiene. One canonical, one title, one description per page; stable vehicle URLs for the life of the car; filters kept out of the crawl; a sold-vehicle lifecycle that never 404s on sale day.
• Speed that holds with everything on. Hero image priority, zero layout shift, third-party tools loaded as quiet buttons rather than overlays.

3. Content written to be named

• A surface plan of forty or more substantive pages per rooftop: model and trim hubs, comparison pages sized to local driving, finance and fee explainers, service content with real prices, city pages that say something true, and a journal.
• The store's name inside the claim sentences, never beside them. This is the difference between being cited and being mentioned, and it is a writing decision.
• Written from the store's own facts, not filler: the inventory, the market, the prices, the people. AI drafts in the Back of House, a person publishes.
• Nothing fabricated, no superlative without a source. The engines cross-check, and so do regulators.

4. Reputation, by the rule

• Ask every customer, or ask none. The federal rule bans gating by sentiment and conditioned incentives. Asking everyone is explicitly safe, and it is also what produces volume.
• No salesperson names in reviews. A pattern the engines and Google both treat as manipulation.
• Velocity tracked. Recency and volume are what the engines weigh; two hundred reviews and a steady monthly rate are the working targets.
• Proof in every medium. Photos, video, written outcomes, before the claim. Engines and shoppers both cross-check a recommendation against what others say.

5. Measured per engine

• Presence rate, not rank. A fixed panel of the questions shoppers in that market ask, run against ChatGPT, Claude, Perplexity and Google AI Overviews on a schedule, reported per engine with a confidence interval.
• Cited, mentioned and recommended reported separately, because they have different causes and different fixes.
• Keyword rankings too, alongside the AI numbers, never instead of them.
• A plan after every run, kept with the run, so the next run shows whether following it worked.

The SparkBeacon method →

What Spark Noir Creative does not sell

• An llms.txt file as a feature. No major engine has publicly confirmed reading one. It costs nothing and may help; it is not a substitute for pages the crawlers are verified to reach.
• A rank inside an AI answer. Position in an AI answer is noise from one day to the next. Presence across many runs is stable, so that is what is reported.
• A blended AI score. The same store scores two to three times differently between engines. One number would hide the one that matters.
• "Cited" sold as "recommended." Being a source and being the store an assistant names for a stranger are different rungs. Spark Noir Creative reports which one a result is.