Chartbeat data on small publishers shows search referral traffic down around sixty percent over two years. Medium publishers are down roughly forty-seven percent. The mechanism is simple: AI Overviews and chat assistants answer the question without the click.
The conclusion people draw from this is usually wrong. It is not that content stopped working. It is that a specific kind of content stopped working, and most blogs are made entirely of that kind.
The test
Ask one question about any post you have published: could a model produce a satisfying answer to this query by reading the top ten results? If yes, that post will not earn a click, no matter how well written. Summarising what is already known is the thing models do best and it is what most marketing blogs consist of.
Four types that hold up
Original data. Numbers that exist nowhere else because you generated them. A survey of your customers, benchmarks from your own platform, a pricing analysis nobody has run. A model cannot invent your data, so it has to cite you.
Working tools. A calculator, a checker, an interactive assessment. A model can describe how to calculate something but cannot be the calculator, so it links to one. This is the most durable format and the least used by Indian marketing teams.
Named first-hand experience. What happened when you actually did the thing, with specifics, attributed to a real person. Not a summary of best practice, an account of a particular attempt including what went wrong.
Genuinely current information. Regulatory changes, pricing updates, platform policy shifts. Training data is stale by definition and models are cautious about time-sensitive claims. For anything with a live deadline this is a real advantage.
What to do with your archive
Sort every post into two buckets: a model could write this, and a model could not. Stop updating the first bucket and stop producing more of it. Take the time you were spending there and put it into one piece from the second bucket per month.
One survey with real numbers will out-earn thirty explainers. That is uncomfortable if your content operation is built for volume, which is exactly why most teams keep publishing explainers.