How to pitch an AI-visibility audit to a local-business client

A client who's happy with their Google position doesn't think they need anything else from you. The conversation that changes that isn't "AI search is the future" — it's showing them, specifically, what happens when someone asks ChatGPT the exact question their next customer is going to ask.

Start with the question, not the pitch

Before you open a deck, run one check. Ask an AI model something a real customer would ask — "best [their category] in [their town]" — and see what comes back. If your client's business isn't in the answer, you have a concrete, specific fact to open with, not a hypothetical. If it is, you have a different opening: showing them who's sitting alongside them, and why that competitor might be there.

RankSignal's free scan (ranksignal.co.uk/scan) does exactly this in about 30 seconds, no login — it's built for this exact moment in a sales conversation, run live on a call or screen-shared beforehand.

Frame it as coverage, not replacement

Clients who've invested in Google rankings can hear "you also need AI visibility" as a sunk-cost threat — like the SEO work they paid for wasn't enough. It's more accurate, and lands better, to frame it as a second channel: Google rank answers "where do I show up when someone searches," AI visibility answers "what does the assistant say about me when someone doesn't search at all, they just ask." Different question, same customer.

What to actually show them

• The literal prompt you asked and the literal answer that came back — not a score, not a summary of a score. The realness of it is what lands.
• Whether they were named at all. A flat "not mentioned" is a clearer motivator than a vague percentage.
• Who was named instead, if anyone. That's the competitor conversation, and it's usually the moment a client leans forward.

Don't oversell what one check proves

One prompt, one model, one moment in time is a snapshot — useful for a pitch, not a strategy. Be upfront that a single answer can vary between asks, and that what you're proposing is tracking this properly over time, across the models that matter, the same way you'd never judge a Google ranking off one search either.