Search referrals to finance content fell 23% in one quarter.

Two numbers in that summary point in opposite directions, and the gap between them is the finding.

Guide and how-to content fell more than 20%. Opinion and commentary rose 4%.

The pattern is not about quality

A how-to guide answers a question that has a correct answer. Once a model can produce that answer, the click stops being necessary — not because the guide was bad, but because the reader's need was informational and it has been met.

Opinion is not substitutable in the same way. A summary of someone's argument is a worse product than the argument, and readers who want a view want whose view it is.

So the sorting is by replaceability. That reframes the question for a professional firm from "how do we rank" to "what do we publish that a summary cannot replace".

What survives replacement

Three kinds, and all three are harder to produce than a guide, which is precisely why they hold.

A judgement with a name on it. Not what the rules say — what you would do, and why, and what you would refuse. A model can summarise the regulation; it cannot tell a reader what a broker in Edinburgh thinks about it.

A number nobody else has. First-party data is unreplaceable by definition. Our own infrastructure bill is on this site for that reason, and so is the list of figures we removed.

A tool that does something. A calculator cannot be summarised into uselessness, because the reader has to put their own numbers in. That is the whole argument for publishing arithmetic rather than conclusions.

The uncomfortable corollary

Most professional-services content marketing is guides. "What is the Renters' Rights Act". "Five things to know about Consumer Duty". That is the category falling fastest, and the response most agencies will recommend is producing more of it, faster.

The honest advice is the opposite: publish fewer and make each one carry something that is yours. A guide with your judgement, your refusals and one number you generated is a different object from a guide, even if it looks similar in a content calendar.

What this does not tell you

It measures publishers, not professional firms. A brokerage's website and a news site earn traffic differently, and a 23% fall in referrals to finance journalism does not mean a 23% fall in people finding a broker.

The forecast is a straight-line projection from a six-month window, which is the weakest kind of forecast; "pageviews halve by Q3 2027" assumes a trajectory continues, and trajectories rarely do exactly. We would not plan against the date. We would plan against the direction, which is corroborated by the underlying behaviour in the note on AI-search visibility.

And it says nothing about whether a model can reach your pages at all, which is a separate and more immediate problem since a crawler default changed in September. The visibility checker asks that question, and the retainer is where the answer stays fixed.