How we measure this

Twice a week we ask four AI assistants a fixed set of questions a real customer might ask, and we check whether your business is named in the answer. We keep the answers verbatim, so you can read them yourself rather than taking a number on trust.

We do not teach AI about your business, and nobody can. Sending information to an AI model through its API does not change what that model tells anyone else — it processes your request and keeps nothing. What actually decides whether you get named is what the assistants find when they look: your Google Business Profile, Bing Places, Apple Maps, directory listings, and whether their crawlers can reach your website at all. That is the part we work on.

ChatGPT, Gemini, Perplexity and Claude, through each provider’s official API with web and local grounding switched on. We use the sanctioned APIs rather than automating the chat apps, because automating those breaches the providers’ terms.

Microsoft Copilot is not included, and we would rather say so than imply otherwise. There is no official API that returns Copilot’s consumer answers. Copilot is grounded on the Bing index, so we track your presence there through Bing’s own webmaster tools instead — that is a related signal, not the same measurement, and we label it as such.

20 questions, in two groups that mean different things:

• 12 that never mention you by name — “who are the best plumbers in Moreleta Park?” This is the hard one, and the one that matters. Being suggested when nobody asked for you is what actually brings work in.
• 8 that name you directly — “do you know Joe Plumbing in Moreleta Park?” Being recalled when asked about is much easier than being recommended unprompted, so we report these separately and never average the two into one figure.

The first group is shared with every business of your type in your suburb. We run those questions once and score them for everyone at the same time. That keeps the price down — and it is also more accurate, because you and your competitors are measured against exactly the same answers rather than separate rolls of the dice.

By fixed rules, never by asking another AI. We look for your business name, any alternate names you have confirmed, your phone number, and your website address — including in the sources the assistant cited.

The rules are deliberately strict rather than generous. We match on whole names, so “Ace Plumbing” is not counted when the answer actually said “Palace Plumbing”. If we are not sure, we count it as not mentioned. That means we may occasionally undercount you — which we would much rather do than tell you that you are being recommended when you are not.

AI answers are not fixed. Ask the same question twice and you can get different businesses back. So a single run is a sample, not a verdict — and a percentage on its own would be presenting chance as fact.

Every rate we show carries a range showing where the true figure most likely sits. We lead with the plain count — “named in 6 of 20 answers” — because that is the honest shape of what we know.

The same caution applies to change over time. We only call a week-on-week movement real when the two ranges do not overlap. Small swings are usually noise, and we will not dress one up as an improvement.

Sometimes an assistant is unavailable, or a run stops early. When that happens we record it as a gap — we never count it as a week you were not mentioned. A gap is our problem, and reporting it as your result would be telling you your visibility fell when in fact our own check did not finish.

If too much of a week is missing, we say the week was incomplete rather than showing you a figure built on half a run.

• Not a guarantee. Nobody can guarantee that an AI assistant will recommend a particular business, and anyone who tells you otherwise is selling you something they cannot deliver.
• Not identical to what you see in the app. We measure through official APIs. The consumer apps can answer differently — this is a close, consistent proxy, not the same surface.
• Not a traffic forecast. AI assistants currently send a small share of web traffic, though a fast-growing and unusually high-intent one. We think it is worth getting right early. We are not going to pretend it is your main source of customers today.

When a customer scans one of our review QR codes, they are shown every review platform the business has set up — Google, Hellopeter, Facebook — in the same order, whoever they are and whatever they thought of the job. There is no question asked first, and no version of the page that shows fewer options to someone who seems unhappy.

That is not restraint on our part, it is the law of the platforms. Steering happy customers to Google and unhappy ones to a private form is called review gating. Google prohibits it, and a business caught doing it can have its profile publicly flagged — a far worse outcome than never asking for a review at all. We have built it so that it cannot happen by accident: the code that decides which links to show is given only the business's own saved links, and has no way to find out who is asking.

The page also offers a private message straight to the business, always below the public options and never instead of them. If you leave one, it goes to the business and is not published anywhere by us.

One thing no tool can do, including this one: no API anywhere creates a review. Not Google, not Hellopeter. Every link takes you to that platform's own form, on your own phone, and what you write there is between you and them.

If we change how a mention is detected, the numbers move without anything about your business changing. When that happens we mark the break in your history rather than drawing a line through it and calling it progress.