Clicks fell 42%. What survived tells you what to write next
The headline figure is bad enough to be quoted everywhere and specific enough to be useful almost nowhere. The number underneath it is the one that should change what you write.
Search Engine Land reported the analysis on 12 March 2026, based on Define Media Group’s study of 64 websites tracked through Search Console:
Organic search clicks have fallen 42% since AI Overviews began expanding in Google Search.
Danny Goodwin, Google AI Overviews cut search clicks 42%: Report, Search Engine Land
And, in the same dataset, the finding that matters more:
Breaking news traffic grew 103% from November 2024 through early 2026.
Danny Goodwin, Search Engine Land, citing Define Media Group
Losses concentrated in evergreen and informational content. Gains in material that was time-sensitive. The traffic did not evaporate uniformly — it moved, and it moved according to a rule you can act on.
What the rule appears to be
An assistant answers a question by summarising what it already knows. If the answer is stable — what a term means, how a process generally works, which options exist — the summary is as good as your page, and the visit does not happen. If the answer is not yet settled, or changes weekly, or requires knowing what happened yesterday, the model has to send the reader somewhere.
Put crudely: explaining loses, and being current wins.
That is uncomfortable for most B2B content strategies, because most B2B content is explanatory by design. “What is X”, “the complete guide to Y”, “X vs Y” — pages written to catch someone at the start of a problem. Those are precisely the pages an assistant can replace with a paragraph.
What survives, for a company that is not a publisher
A B2B studio is not going to win on breaking news. But the underlying property — content the model cannot generate from what it already has — is available in three forms:
Your own numbers. What you actually found in forty audits. What a migration actually cost. What the failure rate of a pattern actually is in your practice. No model has this, because it exists only in your files.
Your own position. Not “here are the six options” but “here is the one we choose, and here is when we would not”. A model will happily list the six options. It will not commit to the fifth on your behalf, and a reader who wants a decision rather than a menu has to come to a person.
Your own cases. Specific, dated, with the constraint that made the project hard. This is the material that gets cited rather than summarised, because it is a source rather than a restatement.
What to stop doing
Writing the definitional page. If a competent assistant can answer it in a paragraph, that paragraph is now the competition, and you will not out-rank it with a longer version of the same paragraph.
This is not an argument for publishing less. It is an argument for the same effort going into fewer, heavier pieces that could not have been written by someone who had not done the work.
The caveat worth stating
Sixty-four websites is a real dataset but a narrow one, and it skews toward publishers, whose traffic mix is nothing like a B2B services company’s. The direction is credible; the magnitude may not transfer. Treat 42% as evidence that the ground moved, not as a forecast for your own site — and now that Search Console reports AI impressions, you can measure your own version of it instead of borrowing someone else’s.
Source: Danny Goodwin, “Google AI Overviews cut search clicks 42%: Report”, Search Engine Land, 12 March 2026, reporting analysis by Define Media Group.
