AnswersThe question people ask first

Why does a competitor come up first when you ask an assistant what to buy?

Almost always because they clear a step you do not, and there are only five steps. The assistant has to be able to read you, name you, describe you accurately, rank you, and tell the buyer where to get you. Whichever of those you fail first is the answer, and it caps every one after it however good the rest are.

What has to happen before an assistant recommends anybody?

Five things, in order. It has to be able to read you, name you in an answer, describe you accurately, put you at or near the top, and tell the buyer where to get you. Each one depends on the one before it.

An assistant answering what to buy is doing something closer to summarising a literature than looking anything up. It draws on what it has read about a category, and on whatever it retrieves at the moment of asking, and it produces a shortlist.

So a brand can fail at five distinguishable points. Nothing readable about you exists, so you are not a candidate. You are readable but never come up. You come up and what is said about you is wrong. You come up accurately and sit fourth. Or you are recommended and the buyer is sent somewhere they cannot complete the purchase.

The order matters more than the list, because the earliest failure caps everything after it. A brand nothing describes cannot be ranked well, so improving the pages that would have ranked it changes nothing at all until the earlier gap closes.

Is the competitor actually better?

Usually not in any way the assistant can see. What it can see is what has been written about each of you, and a brand that appears in the comparison articles, forum threads and round-ups it reads will be named by it whether or not the product is stronger.

This is the part that surprises people and it is worth sitting with. An assistant has no access to your product. It has access to what has been published about your product, filtered through what it happened to read and what it retrieves when asked. So the competitive advantage it responds to is coverage rather than quality.

In practice a small number of pages decide most of a category: a few comparison articles, a couple of community threads, a retailer listing, sometimes a video. Whoever is described well on those is described well by the assistants. It is a narrower and more attackable set than a search ranking, which is the good news buried in this.

Which also means the fix is rarely on your own website. That is the uncomfortable half and it is why plans that only ever say fix your pages tend not to move anything.

What if you are mentioned but they are the one recommended?

That is a different failure from being absent and it needs a different fix. Being named and passed over means the assistant knows you exist and something in what it read made somebody else the safer answer.

An answer can do four things to a brand: recommend it, name it and recommend somebody else, describe it wrongly, or leave it out. Counting mentions collapses all four into one number, which is why a mention count can rise while sales do not move.

Named but not chosen is demand you already paid to create going to somebody else, and it usually has a specific cause you can find: a missing specification the assistant needed to be confident, a claim about you that a source got wrong, a price or availability signal it could not confirm, or a rival with clearer third-party evidence for the same claim.

That is a much better position to be in than absence, because the hard part is already done. You are in the consideration set and the work is narrower.

How do you tell which of the five it is?

Ask the questions your buyers ask, repeatedly, across the assistants they use, and record what comes back. The pattern in the answers tells you which step fails, and repetition is what separates a real gap from one bad reply.

You can start this yourself in ten minutes. Write down the ten questions a buyer would actually type, ask each one, and record which brands come back and in what order. Do it more than once, because you will get different answers, and the difference is the point.

What you are looking for is which failure repeats. Never appearing at all is a different diagnosis from appearing fourth every time, and appearing with a wrong claim attached is a third. Once you can name which one it is, the work is obvious and usually smaller than expected.

The reason to measure rather than sample by hand is not rigour for its own sake. It is that assistants give different answers to the same question, so a handful of tries cannot tell you whether something changed or you got unlucky. That is also what makes a competitor look dominant on a Tuesday and absent on a Thursday.

What does the fix actually look like?

Getting described accurately in the specific places the assistants read when they answer your category. Which is a short list, and it can be named rather than guessed at.

The work divides into three, and which one you need depends on the step that fails. Being present in what gets read, which is comparison pages, round-ups, community threads and retailer listings. Being described correctly there and on your own pages, which is where a wrong claim gets corrected. And being findable at the moment of purchase, which is availability, price and where to buy.

What makes it tractable is that the list of places is short and knowable. For most categories a few dozen sources account for most of what assistants quote, and the same handful come up again and again. Once you know which ones name your competitor and not you, the plan writes itself.