How does ChatGPT decide which local companies to name?
Three inputs combine, and knowing them tells you what's fixable and how fast:
- Model knowledge. The model's training data includes snapshots of the public web. Companies with years of consistent presence — site, listings, reviews, coverage — may be "known" without any lookup. Slow to change; rewards longevity and consistency.
- Live web search. For local recommendations, ChatGPT usually searches the web mid-answer and summarizes what it retrieves: your service pages, review platforms, "best [trade] in [city]" roundups. Changeable in weeks — this is the main battleground.
- Business-listing data. Local answers increasingly draw on maps-style structured data: categories, hours, ratings, review counts. Your Google Business Profile and its equivalents feed this layer directly.
Notice what's not on the list: any form you can fill in, any fee you can pay, any trick. The evidence is the mechanism — which is why the audit below measures, and the work section fixes evidence.
Why a fixed prompt list? Because AI demand is invisible
Classic search comes with a meter attached. Google Ads keyword data (US figures, pulled September 2026) can tell you that "best plumber near me" runs about 14,800 searches a month, "best roofing company near me" and "best hvac company near me" about 8,100 each, "best electrician near me" about 3,600 — and "garage door repair near me" a staggering 165,000. You can see the demand, so you can rank-track against it and know what a position is worth.
Now try to get the same numbers for ChatGPT. You can't — and neither can any vendor. As of September 2026 there is no keyword-volume dataset for assistant prompts: no tool reports how many homeowners in your city asked an assistant who should replace their water heater last month, because the AI companies don't publish it. The demand is real — the mechanism section above is how it gets answered — but it's invisible to measurement. Any vendor quoting you "AI search volume" for specific prompts is quoting a number that does not exist.
That asymmetry is the entire reason this audit is built the way it is. When you can't measure the demand side, the only rigorous option left is to fix the sample: a constant set of prompts, phrased the way homeowners phrase them, run identically every time. You're no longer asking "how many people search this?" — unanswerable — but "when someone does ask, am I in the answer?", which is answerable, repeatable, and honest. That's the instrument below.
The 25-prompt AI visibility audit
Five intents, five phrasings each. Replace [trade],
[job], [city], [zip], and
[company] with your specifics, keep the wording otherwise
identical, and run each prompt in a fresh conversation. Do it in
ChatGPT, Gemini, and Google (noting AI Overviews) — same 25 prompts
in each.
Intent 1 — Hiring: "who should I hire?"
| # | Prompt |
|---|---|
| 1 | Who are the best [trade] companies in [city]? |
| 2 | I need to hire a [trade] contractor in [city]. Who do you recommend and why? |
| 3 | Give me a shortlist of three reputable [trade] companies near [zip]. |
| 4 | Which [trade] company in [city] is best for a [job]? |
| 5 | I'm new to [city]. How do I find a trustworthy [trade] company here? |
Intent 2 — Emergency: "who do I call right now?"
| # | Prompt |
|---|---|
| 6 | Emergency — I have a [urgent problem] in [city] right now. Who should I call? |
| 7 | Which [trade] companies in [city] offer 24/7 emergency service? |
| 8 | Who is the fastest-responding [trade] company near [zip]? |
| 9 | It's the weekend and I need a [trade] contractor in [city] today. Options? |
| 10 | Who handles same-day [job] in [city]? |
Intent 3 — Comparison: "help me choose between options"
| # | Prompt |
|---|---|
| 11 | Compare the top-rated [trade] companies near [zip]. |
| 12 | Compare [company] with other [trade] companies in [city]. |
| 13 | I have quotes from three [trade] companies in [city]. How should I decide between them? |
| 14 | Which [city] [trade] companies have the best reviews for [job]? |
| 15 | Rank the [trade] companies in [city] by reputation. |
Intent 4 — Reputation: "can I trust this company?"
| # | Prompt |
|---|---|
| 16 | Is [company] a reputable [trade] company? |
| 17 | What do reviews say about [company] in [city]? |
| 18 | What are the pros and cons of hiring [company]? |
| 19 | Has [company] had complaints or negative reviews? |
| 20 | Tell me what you know about [company], the [trade] company in [city]. |
Intent 5 — Cost research: "what should this cost, and who's fair?"
| # | Prompt |
|---|---|
| 21 | How much does a [job] cost in [city], and which companies are fairly priced? |
| 22 | What's a fair quote for [job] in [city]? Which local companies should I get quotes from? |
| 23 | Who offers free estimates for [job] in [city]? |
| 24 | Is [company]'s pricing for [job] reasonable for the [city] market? |
| 25 | How do I avoid overpaying for [job] in [city], and which companies have transparent pricing? |
How do you score it?
- One point per mention. Your company named in an answer = 1 point for that prompt, per tool. Maximum 25 per tool.
- Record what was said. For every mention, copy the sentence. Mark it accurate/inaccurate and positive/neutral/negative. An inaccurate mention is a finding, not a win.
- Record who else was named. The competitors appearing repeatedly are your real AI-visibility rivals — often not the ones you expect.
- Log the run conditions. Date, tool and version if shown, your location or the location you specified, exact phrasings. Answers vary; the log is what makes comparison honest.
- Re-run quarterly, identically. Same prompts, same tools, fresh conversations. The trend line is the deliverable.
What scores mean — rough bands, not gospel
Most established local companies first score in the low single digits per tool; zero is common and not a crisis — it's a baseline. Scores concentrated in Reputation prompts (16–20) with none in Hiring prompts (1–5) usually mean the AI can verify you exist but has no reason to recommend you: typically a review-strength and citable-content gap rather than a data problem.
What actually moves the score?
Four workstreams, in the order most local companies should attack them:
- Entity consistency. One canonical name, service list, and service area across your website, Google Business Profile, and every directory that mentions you — dead listings and old business names cleaned up. Contradictions make you unsafe to recommend.
- Google Business Profile completeness. Categories, services, hours, photos, attributes — the structured layer local AI answers draw from. Full scope on the GBP page.
- Review volume, recency, and responses. The dominant input to every "reputable?" and "best?" answer. A steady cadence of real reviews outperforms any burst.
- Citable pages. Cost, process, and comparison content engines can quote — the craft covered in the AEO guide, built as part of home services SEO.
The honest limits
As of September 2026: nobody controls AI answers. They vary by phrasing, session, user location, and model version, and they change when models are updated — sometimes overnight, in either direction. No vendor (including us) can guarantee specific mentions. What the audit gives you is the thing guarantees pretend to be: a repeatable measurement, tied to work with a plausible mechanism, judged by its trend. That's the strongest honest claim available in this market — distrust anyone offering a stronger one.
Prefer the work done for you — the audit across all three surfaces, plus the remediation itself? That's what AI search optimization covers, and the pricing page explains how engagements are scoped.