Your B2B website stopped producing enquiries

Not because those buyers chose a competitor. Because they asked an AI system who to consider, got two or three names, and contacted those. You were never in the conversation, so there was no visit to record and no enquiry to lose.

That is a measurable thing rather than a theory. You find out by asking the questions your buyers ask, in their words, across the systems they use, and writing down which companies get named. If your name appears in two answers out of forty, you have your explanation. If it appears in thirty, the cause is somewhere else — and you have ruled out an expensive guess.

Did the traffic actually fall, or only the enquiries?

Check which one before diagnosing either. They are different problems with different causes, and a quarter of wasted work usually starts with treating them as the same thing.

If sessions are steady and enquiries are down, something on the site or in the offer has changed. If sessions have fallen on the pages that used to produce enquiries, while brand searches hold steady, that is a different shape entirely — it means the people who did not already know your name have stopped arriving.

We are not going to give you a percentage here. Published figures on traffic loss measure other companies' websites, in other categories, against question sets that have nothing to do with yours. Quoting one at you would be the opposite of what this article is for.

Why the fall is uneven

Google's generative features are built on its core Search ranking systems, and use retrieval-augmented generation and query fan-out to surface material from the Search index. One buyer question becomes several behind the scenes, and the answer is assembled from what those searches return.

The practical effect is that questions with a short factual answer get answered in place. Questions that end in a decision still send people to a website. So the loss concentrates on exactly the pages that used to do your introducing.

A packaging machinery manufacturer whose best-performing page was "how to choose a case packer for glass jars" loses more than one whose enquiries arrived through its own name. The first page answered a question. The second was a destination.

Source: Optimizing your website for generative AI features on Google Search, Google Search Central.

Why your reporting still looks healthy

Because every tool you own measures visits, and this loss happens before a visit.

A buyer who reads three supplier names and contacts two leaves no trace with the third. No session, no referrer, no search term, no bounce rate, nothing in the CRM. There is no record of an enquiry that was never made, so the dashboard reports a quiet quarter with no cause attached to it.

You do not lose the deal. You never learn it existed.

This is why the first sign is usually a conversation between a sales director and a finance director in which neither can explain the number.

But we still rank on page one

You can, and still be absent from the answer a buyer reads. Ranking measures where your page sits in a list of links. Being named measures whether a company appears in an assembled answer. They are related, because the answer is assembled from indexed pages, but they are not the same measurement and a healthy one does not guarantee the other.

That distinction is the whole subject of what AI visibility actually means, which also sets out why it is four separate readings rather than one.

Is it us, or is it the market?

It may well be the market. Capital equipment runs in cycles, currency moves, a competitor opens a plant closer to your customers, and a slow quarter can have four causes at once.

Measurement does not tell you the market is fine. It tells you whether you are present in the answers your buyers are reading, which is one candidate cause you can actually do something about. If you are present in most of them and enquiries have still fallen, look elsewhere — and you have narrowed the search rather than guessed at it.

Ruling a cause out is worth as much as confirming one. It is usually cheaper too.

What actually tells you

Three readings, taken from the same set of questions.

AI Visibility is how often you appear at all in answers to the questions your buyers ask. Citation share is the narrower figure underneath it: the share of tested questions in which your company is named. AI Position is where you sit relative to the suppliers named alongside you, because being third in a list of three is not the result being first is. All three are defined on Baseline, the step of our method that produces them.

None of this requires a tool. Thirty questions your buyers actually ask, put to the six systems in a logged-out session, with the answers written down, will tell you more in an afternoon than a year of dashboards. If you would rather have it done properly — agreed questions, competitors measured alongside you, the whole thing in writing — that is what the AI Visibility Audit is.

Why competitors are named and you are not

Usually not a judgement about quality. More often it is that the information about them is easier to retrieve, more consistent across sources, and corroborated somewhere other than their own website. That has its own article: why your competitors appear in ChatGPT answers and you do not.

What this article cannot tell you

It cannot tell you how much you have lost. Nobody can count enquiries that were never made, and any figure presented as that count was modelled rather than measured.

It cannot tell you that AI answers are the cause of your particular quarter. It says the cause is testable, which is a smaller and more useful claim.

And a single reading is a snapshot. AI systems are probabilistic and change continuously — the same question can return a different answer a week later with nothing having changed at either end. One test tells you whether you are absent today. A fixed set of questions, re-run on a schedule, is what tells you whether that is moving.

What to do this week

Write down thirty questions your buyers ask before they know who to call. Not keywords — questions, in their words, the kind that arrive in the first phone call. Put each one to ChatGPT, Google AI Overviews, Google AI Mode, Gemini, Perplexity and Claude, logged out. Write down every company named.

Then count how many of the forty answers include you. Whatever that number is, it is the first real thing you will have learned about this, and it is more than your analytics will ever tell you.