The person who has to explain the number

Why you cannot see it happening

When someone asks ChatGPT which company to use and it names three competitors instead of you, nothing is recorded anywhere you can look. No impression. No session. No referrer. No bounce. The moment that decided the deal left no trace in any dashboard you own.

That is what makes this so frustrating to diagnose from the inside. The numbers fall, every individual explanation you test comes back clean, and the actual cause is happening in a place your instruments do not reach. Plenty of companies conclude their content got worse, or blame a Google update, and spend a year fixing the wrong thing.

What we actually do about it

We start from the commercial objective, not the engines: which buyer, which product, which territory, and what a good enquiry looks like to your team. Then we go and ask. Your buyers' real questions, put to search and to ChatGPT, Claude, Perplexity and Google AI Overviews, sampled repeatedly because the answers vary run to run. That shows how often you get named, who gets named instead, and the pages the engines read to decide.

Those are diagnostics. The work that follows (pages, product data, quote and demo flows, technical fixes) runs as a budgeted programme, and it is judged on the enquiries your team accepts. A citation is not a lead, and we do not report one as if it were. The stages and prices are on the pricing page.

You can run the whole method yourself. We wrote it up step by step, free, and what to do with the results. If you would rather check our reasoning before anything else, our evidence register grades every claim we make in this field, including the ones that cost us money to admit, and the FAQ answers the sceptical questions directly.

The four explanations people reach for first

Almost every conversation that starts with “traffic is down and we cannot explain it” passes through the same four theories before it gets to this one. They are worth ruling out in order, because three of them are cheap to check and the fourth is where most of the wasted year goes.

• “It is a Google update.” Easy to test: compare your impressions and your average position separately. A core update usually moves position. The pattern we see more often is impressions holding or rising while clicks fall, which is not a ranking problem — it is the answer being given before the click.
• “Our content got worse.” Occasionally true, usually not. If the same pages that stopped converting still rank where they ranked, the pages did not get worse; the surface in front of them changed.
• “The market softened.” Check whether your named competitors are visible in the answers your buyers get. If they are and you are not, demand did not fall — it got routed.
• “We need more content.” The expensive one. More pages at the same depth mostly produces more pages at the same depth. The evidence that rewriting for quotability wins citations is weak enough that we grade it contested, and the thing that dominates it — being retrievable at all — is usually a configuration problem rather than an editorial one.

None of these rules out the others, and the point of measuring is that you stop guessing between them. A run across the engines tells you in one pass whether your absence is total, partial, or category-wide — three different problems with three different price tags.

What this is not

It is not a replacement for search work. For Google's own surfaces, Google's stated position is that optimising for generative features is largely still optimising for search, and on that narrow point we think they are mostly right. The part that is genuinely separate is the engines that do not run on Google's ranking at all — and those are invisible to every rank tracker you already pay for. That distinction is the whole argument, and we set it out with sources on the evidence register rather than asserting it.

It is also not a fast fix with a date attached. We could not find primary evidence for how long this takes to show effect, so we do not quote one. What we can show is our own position, measured on a frozen prompt set and published on our numbers, including the run where it went down.

If you want it framed for your role

The underlying work does not change by job title, but the commercial question you own does. These pages put it in those terms. If you would rather start from your market, see industrial suppliers or technical B2B and APIs.