How they ask

How do people actually ask an AI for a local business?

The short answerIn three shapes. The recommendation ask wants curation and expects two or three names. The problem ask describes a symptom rather than a service, so the model must work out the trade first. The verification ask happens after someone already has your name — and it is the one most businesses lose without ever finding out.

Branded, Claude30 of 30 probes

Branded, ChatGPT18 of 26 probes

Unbranded, Gemini0 of 64 runs

The problem ask — a symptom, not a service

Why it matters

They are three different competitions, and they have different winners.

The shape decides who wins

These are not three phrasings of one question. They are three different competitions with different winners.

A recommendation ask is a contest between businesses in a category. A problem ask is first a classification task and only then a contest. A verification ask is not a contest at all — it is a single business being checked, and the only outcome is confirmed or not confirmed. Being strong at one says very little about the others.

Three asks, three different contests

Verification is the one that quietly costs you

It is worth dwelling on, because it is the one that already affects every business with any marketing at all.

Somebody sees your van, gets a postcard, is handed your name by a neighbour. Twenty years ago they called. Now they ask an assistant whether you are any good. If the answer is confident and positive, your existing marketing just worked. If it is vague — "I don't have specific information about that business" — the referral quietly dies, and nothing in your analytics records that it happened.

The referral quietly dies, and nothing in your analytics records that it happened.

What it actually is

What we actually measure across the three shapes.

What the data says about each

We probe these weekly, so this is measurement rather than theory.

Verification asks are winnable now. Across the live sites we run, asked about a business by name, Claude named it in 30 of 30 probes, ChatGPT in 18 of 26, Gemini in 18 of 26.

Open discovery asks are much harder. On unbranded questions like "best HVAC company near West Palm Beach", the same businesses are named far less often, and on Gemini they were not named at all across 64 runs.

We would rather tell you that than sell you the easier story.

Asked by name — measured 3 Aug 2026

How it works

What answering each one requires.

Answering all three

The verification ask needs an authoritative record about you specifically — enough structured detail that a model can answer confidently instead of hedging.

The problem ask needs your reviews analysed by problem, not just by service, so that "water heater making a banging noise" connects to you rather than only "plumber" connecting to you.

The recommendation ask needs everything the other two need, plus being discoverable in the first place — indexed, submitted, and present in the sources a model reaches for. It is the hardest of the three and the one we are most careful not to overclaim.

What answering each ask requires

Which of these should I care about most?

Verification, if you do any outbound marketing at all. It is the ask that decides whether the money you already spend converts, and it is the one we can most clearly demonstrate.

Do people really describe symptoms instead of services?

Constantly — it is most of the natural-language advantage. 'Strange smell from the AC vents, who do I call' is a real probed prompt, not a hypothetical.

How do you know what people ask?

We probe a panel of prompts per client weekly against Claude, ChatGPT and Gemini, and record what came back. The prompts on each industry page are real rows from that, with their dates.

Answers come from this site. Anything it can’t answer goes to Gus.

Read the argument

Start, $20/month Book a call View a live site

Not affiliated with Google or any AI platform.Push Button Inc. · Vero Beach, Florida · Privacy · Terms