Before You Optimize What AI Says, Find Out What People Are Asking It
Introducing Answerability Maps — a research series from TrueSignal. Eight maps are live now, starting with the 264 questions consumers ask about an HVAC company. Browse all maps →
Almost every business owner I talk to lately has some version of the same worry: AI is telling customers and prospects things about my company, and I have no idea what.
They're right to think about it. But the space around that worry has gotten noisy fast. AI visibility audits, GEO, AEO, agencies pitching optimization packages, and a dozen contradictory frameworks for "showing up in ChatGPT." Most owners and marketers I meet aren't skeptical of the shift — they're overwhelmed by it. They know something changed. They don't know where to start.
Here's where I'd start, and I think it's the step almost everyone skips: before you try to optimize anything, understand what people are actually asking AI about a business like yours. What comes up every single time. What comes up often. And the specific, long-tail questions that only some people ask — but that decide the job when they do. You can't answer questions you've never seen. And right now, almost nobody has seen the full list.
So we built it.
What an Answerability Map is
An Answerability Map is a complete, structured inventory of the questions consumers put to AI systems about a specific type of business they're considering hiring. Not discovery questions ("who's near me") — consideration questions. The questions asked about you, once you're on the shortlist.
And it's rarely one question. People ask AI something, get an answer, and keep going: is this company any good, do they handle emergencies, how do they compare to the other name on the list, which one would you pick. It's less like a search and more like an interview — and the follow-up questions are surprisingly consistent.
Here's how we mapped it. We ran four AI systems — Claude, ChatGPT, Gemini, and Perplexity — through independent rounds of question generation for HVAC, three runs per system. The raw output was roughly 2,355 question phrasings. We collapsed those into clusters: two questions worded differently are the same question if they'd be answered by the same facts. What survived was 264 unique questions.
Then we tiered them by agreement. If all four systems surfaced a question independently, every time, it's Core — there are 18 of those. Questions that showed up across two or three systems are Common — 35. And 211 questions appeared in only one system: the Long tail, where things get specific. Does this company service heat pumps? Do they hold an EPA 608 certification? Will they honor a warranty on a unit someone else installed?
One honest note on method, because we label our sources or we're nobody. This maps what four AI systems converge on as the canonical question set — it is not a log of observed consumer conversations, and nobody selling AI visibility has one of those either. But here's why the convergence is the thing worth mapping: these are the systems generating the answers. When ChatGPT's model of an HVAC decision centers on warranty terms and emergency availability, it structures its recommendations around warranty terms and emergency availability — whether or not any individual customer typed those words. The map isn't a proxy for the terrain a business competes on. It largely is the terrain. That's why we tier by cross-system agreement instead of inventing volume estimates, and why the full question list, methodology included, is free to download on the map page.
Three things the map shows
1. The questions run on facts, not feelings
Read the 18 Core questions and a pattern jumps out: they're the questions you'd ask in a job interview. How long have you been doing this. Are you licensed and insured. Do you handle emergencies. What do you charge for a service call. Who vouches for your work.
Consumers aren't asking AI for vibes. They're asking for verifiable specifics — and the AI system relaying the answer needs somewhere to find them. Most HVAC company websites answer almost none of these questions in a form a machine can read. The map makes that gap visible, question by question.
2. Comparison questions are where businesses win or lose
A meaningful slice of the map is comparative: Is this company cheaper than the other one? Which is more experienced? Whose warranty is better? Which would you pick?
When AI referees a comparison, it needs differentiating facts. If both companies present as similar star ratings and similar website copy, the comparison collapses — there's nothing concrete to say. The business with a published, verifiable operating history becomes the strong half of the answer. The other becomes the weak half. Not because it's worse. Because it's less legible.
3. 78% of the questions are answerable — by the business
We classified every question on the map three ways: answerable from a business's verified operating records, answerable by facts the business can formally declare, or beyond any data. The result: 78% of what consumers ask AI about an HVAC company can be answered with the company's own data.
Sit with the other side of that number too. 22% of the questions — what do customers say, is the owner honest, what would my job cost exactly — no data layer can touch, and we publish that share just as plainly. But the 78% is the opportunity: the majority of a conversation that decides who gets hired, answerable with information the business already has, currently published nowhere.
Where this goes
Answerability Maps now cover eight verticals: HVAC, Dental, Roofing, Plumbing, Pest Control, Career College, Medspa, and Marketing Agency. Each one maps the full question set consumers put to AI, tiered by cross-system agreement, classified by answerability, free to explore. Methodology travels with every map, and we'll refresh them as the AI systems evolve.
The pattern holds across all eight. The answerable share ranges from 64% (dental) to 78% (HVAC) — in every vertical, the majority of what AI says about a business is answerable with data the business already has, currently published nowhere.
If you're overwhelmed by the whole AI optimization conversation, this is the place to stand. Don't start with tactics. Start by reading what people ask about a business like yours — then look at how many of those questions you currently answer anywhere a machine can read. That gap is the work. Everything else is downstream of it.