When someone asks AI what to buy, does it say your name?

No, and nobody can. We don't control these systems and we have no relationship with the companies that run them.

What we do is measure what they currently say, work out what's driving it, and tell you what to change. Whether that works is influenced by a lot of things outside anyone's control. Our Method page is honest about all of it.

Core covers four: ChatGPT, Google Gemini, Perplexity and Microsoft Copilot. Pro and Global cover seven, adding Claude, Meta AI and Grok. Meta AI matters more here than the global numbers suggest, because in Singapore and Malaysia it sits inside WhatsApp and Instagram.

We stop at seven rather than padding the number, and we only track assistants your customers can actually reach from your market. We don't count China-market chat tools: they would raise the number on this page without describing anyone who buys from you.

We don't track Google's AI Overviews, and you should treat that as a real gap rather than a footnote. Google prohibits automated querying and offers no API for it, so anyone claiming to track it at scale is either doing it by hand or not doing it. We would rather say that than pretend.

What it means for you: if a shopper types your category into Google and reads the AI answer above the results, we did not see that. Gemini covers Google's assistant, and the two draw on overlapping sources, so a change that moves one often moves the other. But they are not the same surface and we won't claim they are. If Google search is where most of your discovery happens, factor that in before you buy.

What nobody can tell you, including us: exactly what share of your buyers uses each one. There is no reliable public data on which AI assistants Singapore shoppers use to decide what to buy, and any firm quoting you a precise percentage is guessing. We would rather say so.

That uncertainty is the argument for measuring several rather than betting on one. In our own readings, two assistants asked the identical question in the same session have never yet agreed on more than half the brands they name. If you only measured the one you assumed mattered, you would have a confident answer and a false one.

Your report shows how each assistant behaved for your category, separately, never averaged. Which ones matter varies more by category than most people expect, and working that out is part of the job.

Partly us, partly you, and the split is deliberate rather than a gap we're hoping you won't notice.

On Pro and Global we write. Not a list of suggestions: a finished piece of content each month, plus page rewrite briefs specific enough for whoever runs your site to act on without a meeting. Global is three pieces a month. That is execution, not advice.

What we never do is touch your store. No admin access, no theme edits, no publishing on your behalf. That's a firm line. A firm that can break your checkout at 2am should not also be the firm grading its own work, and we won't own your storefront.

Some of it only you can do. Getting into a Singapore roundup is a relationship, not a task. We tell you exactly which publications the assistants keep citing for your category and which are realistically reachable, ranked. Someone at your end still has to send the email.

If you want the whole thing done for you, say so when you enquire. That's a scoped engagement, not a plan on this page, and we'll tell you honestly if it isn't worth it.

More, and here is the arithmetic rather than a vague reassurance.

A question is a unique prompt, written for your category. What varies is how many times each one is asked.

Pro, 150 questions, one market. 135 of them go through official APIs and are asked three times each, in separate sessions, across the assistants that offer one. The remaining 15 are typed by hand into the consumer apps, once each, because a person is doing the typing. That is roughly 1,700 individual answers recorded per month, not 150.

Core, 75 questions. 65 through APIs at three runs each, 10 hand-run once. Roughly 600 answers a month.

Why three runs: these systems are non-deterministic, so a single observation tells you almost nothing about whether you genuinely sit in an answer or occasionally appear in one. You get "named in 2 of 3 runs", not "named".

Your report shows both numbers, and never blends them into one figure.

A fair question, because we make a point of the difference ourselves. The consumer apps layer their own retrieval, shopping data and local availability on top of the model, so they and the API genuinely return different answers.

Every figure in your report is labelled with where it came from. Hand-run answers from the consumer apps are reported separately from API answers, always, and never averaged together into a single score.

The split by design: Core is 10 hand-run of 75. Pro is 15 of 150. Global is 20 in your primary market. Two of the seven assistants, Copilot and Meta AI, have no API at all, so those are hand-run whatever the plan.

So the honest description is: the hand-run sample is the high-fidelity reading, showing what a customer actually sees, including product cards, local retailers and SGD prices. The API provides breadth and month-to-month movement at a volume no person could type. Both are useful. They are not the same measurement, and we don't pretend they are.

If a finding matters to a decision you are about to make, ask and we will confirm which half it came from.

A finished, publishable article of roughly 800 to 1,200 words, written for one specific gap we found in your reading, and delivered as a document you paste into your own site.

It is not a blog post about a topic we liked. Each one targets a question the assistants answered without naming you, and it is structured for how they read: a direct answer near the top rather than buried after an introduction, specifics they can lift (prices, materials, sizes, who it suits and who it doesn't), plain headings that match how the question is actually asked, and no filler.

You get the plain text and the suggested title, headings and meta description. We do not publish it, and we never touch your site. Your developer or your platform editor puts it live, which takes about ten minutes.

Core gets one a quarter. Pro gets one a month. Global gets three a month, one per market plus a shared one.

Honest limit: publishing it does not guarantee an assistant will start naming you, and if a question's answers carry no citations at all we will tell you that publishing is unlikely to move it rather than write you something that can't work.

You could ask an assistant a few questions yourself, yes. We'd rather you did that than paid us for something you don't believe in. But the half-hour version and the report are not the same thing, and the difference is not effort.

Asking from your own account tells you what the assistant says to you. These systems personalise from your history, and at least one of them uses it to shape the searches behind the answer. We have asked the same question signed in and signed out and got a different leading brand and a different set of cited sources. If you've been researching your own category for months, your account is the least neutral place to ask.

One reading is a number, not a trend. Answers move on their own. Without a fixed question set asked the same way every month, you can't tell a real change from noise, and you will read progress into randomness.

The answer isn't the finding. Knowing you're absent is the easy half. The work is establishing which specific pages and publications the assistants keep drawing on for your category, which of those you could realistically get into, and in what order. That takes reading dozens of answers side by side, not five.

Do the half-hour version first. If it tells you everything you needed, you've saved S$799 and we'd rather you kept it.

A fair question, and the honest answer is that month one is the most valuable report you will get from us. It's the one that tells you things you didn't know.

But a reading from March doesn't describe September. Your competitors are publishing. New brands launch into your category every month. A single roundup going live can push you out of an answer you were in, and nothing you did caused it. The assistants themselves keep changing which sources they trust. None of that is noise, and none of it announces itself.

So month two onwards answers two questions: what changed in your category, and was any of it you? Those are not answerable from one reading, and they are the questions that decide whether you keep spending on the actions we ranked, or stop.

If you're not acting on the reports, the honest advice is to cancel and come back when you are. It's month-to-month and one click, deliberately. A report nobody acts on is a cost with no return, and we'd rather lose the subscription than take money for that.

Because we're carrying part of it, and we're specific about which part.

You see the work before you pay. The free check is a smaller version of the real report, on your own brand, with your competitors named. Nothing is hypothetical.

The price you start on is the price you keep. Early clients took a chance on an unproven firm, so their price never rises for as long as they stay subscribed, even after we raise it for everyone else.

Cancel in one click, any month. No contract, no minimum term, no notice period.

Every finding is checkable. Each report carries the exact question, the date and time, and the answer verbatim. You can retype any question yourself and see whether we reported it straight. Very few things you buy can be audited that cheaply.

What we won't do is invent testimonials or a "trusted by" strip. When there are named clients, they'll be here.

Because these systems are non-deterministic. Ask the same question twice and you can get two different answers, with no change in the world.

That's why we don't treat small month-to-month movement as progress, and we say so in the report when we think we're looking at noise. Sustained movement over several months means something. A single month usually doesn't.

No. Not your site, not your analytics, not your ad accounts, not your platform. We never ask, and there's nothing to connect.

Everything we measure comes from asking public questions to public AI assistants.

Related, but not the same. SEO is about ranking in a list of links. This is about whether an assistant names you when someone asks it what to buy, where there is no list, just an answer with a few brands in it.

The mechanics overlap. The measurement doesn't exist anywhere in your analytics.

One click, from the billing link on every invoice and every report. There is no notice period, no fee, and nobody you have to email.

It takes effect at the end of the month you've paid for, and you still get that month's report.