Long-form still earns, but not for the reason you think

Long-form content still pays, and it no longer pays in sessions. Judge it on sessions and you will cancel the one format that still earns.

The click economics changed hard. SparkToro, using Similarweb clickstream data for US desktop and mobile web, put zero-click at 68.01% of Google searches across January to April 2026, up from 60.45% in 2024. Searches producing at least one click fell 9.51 percentage points over two years, a 22.9% relative decline. Chartbeat measured global Google search traffic down 33% year on year across more than 2,500 publisher sites, with smaller sites losing around 60% of search referrals over two years against 22% for large ones.

The click did not disappear. It got rationed.

The panic is pointed at the wrong surface

AI Mode was 0.34% of searches in that same January to April 2026 window. A third of one per cent. The strategic anxiety about it is running roughly a hundred times ahead of its query share.

The damage is coming from AI Overviews sitting on ordinary results pages, which is a different problem with a different answer. Ahrefs measured a 58% click reduction for the top-ranking result when an Overview appears. Seer Interactive, across 53 brands and 5.47 million queries, found organic click-through with an Overview present running about 2.1% when the brand's page was cited inside it and about 0.9% when it was not, against about 3.3% on results pages with no Overview at all.

That split is the whole strategy. On the same query, being cited is worth roughly double being present.

Why depth is what gets cited

Retrieval works on passages, and it fans out: one typed query becomes several related sub-queries run at the same time. A page that answers six sub-questions properly has six chances to be the retrieved passage. A page that answers one has one.

That is the actual mechanism behind long-form winning, and it also tells you when long-form loses. Length that comes from six real sub-questions is retrievable surface area. Length that comes from a 2,000-word target is padding, and padding pushes the retrievable passages further down a page that now has a worse reason to exist.

So set length by counting, not by feel. List the questions a buyer genuinely asks about the topic, out loud, in a sales call. Answer each in a self-contained section. If that comes to 700 words, publish 700 words. If it comes to 2,600, publish 2,600.

The traffic that does arrive converts differently

AI referral volume is small and behaves nothing like search traffic. Semrush puts AI-referred visitors at roughly 4.4 times the conversion rate of organic. Seer's client data shows ChatGPT-referred traffic converting near 16%, against about 1.8% for Google organic. ChatGPT referral volume grew 206% year on year from January 2025 to January 2026.

Now the honest part. It is still around 1% or less of most sites' sessions. Anyone building a 2026 plan on AI referral volume is building on a rounding error with a good conversion rate. Treat it as a leading indicator and a very high-quality trickle, not a channel.

The reason it converts so well is worth understanding, though. Someone arriving from an AI answer has already had their basic questions handled and clicked through for the detail. They are late-stage by construction. Which means the page they land on should be the deep one, not the overview.

The value nobody measures

Ask your sales team which pages they send to prospects. It will be the long ones. The detailed methodology page, the technical comparison, the project write-up with the numbers in it.

That page might pull 60 sessions a month and sit in the bottom quartile of every content report you run. It is also in the email that precedes half your closed deals. If your reporting cannot see that, your reporting will recommend deleting it.

Fix it cheaply. Add a UTM to the links sales actually sends, or just ask them each quarter which five pages they used and why. Then report those pages on assisted revenue rather than sessions. This one change saves more good content than any technical SEO work you will do this year.

What long-form should look like now

Answer first. The opening paragraph states the position in plain words, because that paragraph is what gets extracted and it is what a skimming buyer reads before deciding whether to keep going.

Sections that stand alone. Each one answers a full sub-question without depending on the section before it.

Specifics that could only come from you. Numbers from your own jobs, the standard you work to, the case where the usual advice fails. Generic depth is not depth, it is volume.

One honest limitation, stated. It reads as true and it pre-empts the objection a buyer was going to raise anyway.

No history section. No definition preamble on a page written for people who already know the term.

When short wins

Plenty of the time. A pricing page, a specification, a straight answer to a factual question, a process explanation with four steps. Stretching those to long-form makes them worse and buries the answer.

The rule is not that long beats short. It is that the page should be exactly as long as the number of real questions it answers, and most brands are still writing to a word count handed down in 2019.

Depth stopped being a traffic play. It became a citation and closing play, and those are worth more per unit than the sessions ever were.