Almost every brand in this industry has a detailed view of its Google position and no view at all of what an assistant says when someone asks the same question. That gap is odd, because closing the crude version of it is nearly effortless.
What follows is the manual method. It takes an afternoon, costs nothing, and will tell you more than you currently know. It also has real limits, and we will be specific about those at the end rather than pretending otherwise.
Start with the questions, not with your brand
The instinct is to type your own brand name and see what comes back. Resist it. That answers a question almost nobody asks. Someone who already knows your name is not the customer you are worried about.
The questions that matter are the ones asked by people who do not know you yet, phrased the way people actually phrase things. Write twenty of them, spread across five shapes.
| Shape | What it tests | Roughly how it sounds |
|---|---|---|
| Open category | Whether you are a default name at all | Who are the good options for [category] in [market] |
| Attribute-led | Whether you own any specific quality | Which [category] brands are best for [specific attribute] |
| Alternatives | Whether you are in a competitor’s orbit | What are alternatives to [competitor] |
| Head to head | How you are characterised when compared | [Your brand] or [competitor], which is better for [use] |
| Buyer-stage | Whether you appear in a considered shortlist | What should I look for when choosing a [category] provider |
Run them properly, which mostly means running them repeatedly
Ask each question at least three times, in a fresh session each time. This is the step people skip and it is the step that produces the actual insight.
A brand that appears in one answer out of three is in a completely different position from one that appears in all three, and both look identical if you only ask once. Volatility is information. A question where the names change every time is a question nobody has claimed, which makes it the cheapest ground available to you.
Do it on more than one assistant. ChatGPT, Claude, Perplexity and Gemini do not agree with each other, and Google AI Overviews behaves differently again because it sits on top of an existing index. Treating them as interchangeable will give you a confident and wrong picture.
Record five things, every time
A spreadsheet is fine. What matters is recording the same fields consistently, because the value compounds when you repeat this next month.
- The exact question, word for word, because small changes in phrasing move answers.
- The engine and the date.
- Every brand named, in order. Order is not incidental. First mention carries disproportionate weight with a reader.
- Whether you were named at all, which is the number you will end up reporting.
- Any sources cited. This is the most valuable column and the one most people leave out.
That last column is where the manual exercise stops being a curiosity and starts being a plan. After twenty questions you will notice the same handful of domains appearing over and over. That set is what is actually deciding your category, and if you are not in it, you now know precisely what to work on.
Five ways people get this wrong
Asking about themselves. Covered above and worth repeating, because it is the most common mistake by a distance.
Asking once. One answer is an anecdote. You cannot distinguish a stable position from a coin flip.
Leading the question. Asking whether your brand is good is not a test. The model will generally be agreeable, and you will learn nothing except that it is polite.
Testing from one place. Answers differ by market, and for this industry they differ enormously. We wrote about why that happens separately.
Recording the verdict and not the reasoning. Knowing you were absent is mildly depressing. Knowing which three sources put your competitor there is actionable.
Where the manual method runs out
We would rather say this plainly than let you discover it in month three.
Sample size. One person asking twenty questions three times is sixty data points. That is enough to spot the obvious and not enough to be confident about anything subtle, or to defend a budget request.
Markets. You can only easily test from where you are. Testing properly across several markets by hand is slow and difficult to keep consistent.
Benchmarking. Being named in nine answers means nothing without knowing what typical looks like in your category. Absolute numbers are not a position.
Consistency over time. The value comes from repeating this identically. Manual processes drift, questions get reworded, someone leaves, and the comparison quietly stops being a comparison.
None of which is an argument against doing it. Do it. It is the cheapest way to find out whether you have a problem, and if the answer is no, you have saved yourself a purchase. If the answer is yes, you will have a much better conversation about what to do next.
Common questions
How many questions do I need?
Twenty is enough to learn something and few enough to finish. The value is in running each of them several times rather than running two hundred once, because a single answer tells you nothing about whether a position is stable.
Should I use a logged-in account or a fresh session?
A fresh session, every time. A logged-in assistant that has watched you research your own brand for months is not answering the question a stranger asks. Personalisation is the single fastest way to convince yourself you are doing better than you are.
How often should I repeat it?
Monthly is a reasonable rhythm for a manual check. Answers move as sources change and as models are updated, so a single snapshot ages quickly. What matters more than frequency is that you record it the same way each time.