Is the click dead? What changes when Google answers on its own

The most solid measurement available today compared 300,000 keywords — half with an AI answer at the top, half without — using aggregated Search Console data.¹ In April 2025, the click gap was 34.5%.² When the same calculation was rerun with December data, the gap had climbed to 58%.¹ In eight months, the loss nearly doubled.

The figure is a correlation between two similar groups of queries, not a controlled experiment. And it measures position one — whoever sits at position five was already getting thin clicks before AI answers showed up. If your own spreadsheet shows a smaller drop, your query mix is simply different from the average.

What nearly every report agrees on: the drop is real, it's large, and it accelerated.

AI answers don't appear on every query, and when they do, they don't hit every query equally. The pattern is consistent: the more a query asks for a fact, the more the AI resolves it on its own; the more it asks for a decision, a price comparison, or a vendor, the more the user keeps clicking. That changes the math on which content still pays off.

The practical read of that table is uncomfortable but useful: pure top-of-funnel content — glossary entries, what is pages, definitions — stopped being a good traffic investment. It still serves another purpose: getting cited. And citation is a different currency from the click.

This is where the logic flips, and almost nobody has adjusted yet. The question is no longer how do I rank higher but how do I become the passage the system lifts. Those are different criteria, and the second one is easier to meet than the first — because it depends on structure, which is entirely under your control.

Word count doesn't help. An analysis of 174,048 pages and 1.6 million cited URLs found a correlation of 0.04 between word count and AI-answer citation — which, in statistics, is the same as none.³ The size that matters is the size of each passage: a 2025 analysis of 10,000 AI citations found that passages of 40 to 75 words were cited 3.1 times more often than longer ones.⁴

What matters is shape. Self-contained blocks, passages that make sense read outside the page, are what AI systems can actually use: your content has to hold up when it's pulled out in chunks of 50 to 150 words.⁵ It makes mechanical sense: the system lifts a passage, not the whole page. If your passage only works alongside the paragraph before it, there's nothing to lift.

And hitting it in one place isn't enough. Between AI Mode and Google's own AI Overviews, only 13% of citations overlap.⁶ They're two audiences that barely draw on the same sources.

The most expensive error of the moment is reacting to the traffic drop by publishing more. More posts, more words, more variations of the same question. That raises cost without touching the cause: the problem sits in format, not in volume. More text with the same structure produces no more than before: nothing.

The second mistake is abandoning informational content entirely because it doesn't bring clicks anymore. That's throwing out the new currency along with the old. An article the AI cites puts your company name inside the answer, in front of someone who's deciding — without spending a cent on media. It gives you presence, the step that comes before the click.

The third mistake is assuming a tool fixes this. No plugin turns a poorly structured paragraph into a citable snippet, because structure is an editorial decision, not a software setting.

No full-site rewrite, no new hires, no new tools. The steps below are ordered by speed of return: the first two tackle what matters most; the last ones are habits, not projects. Work through them in order and stop when you run out of time — getting halfway already changes the result.

1. Pick the five pages that used to bring the most traffic and lost the most. Search Console shows this for free: compare impressions and clicks from the last three months against the same period last year.
2. Open each one with a 40-to-75-word answer. Before any subheading, answer the title's question in full. Someone who reads only that should leave with what they came for.
3. Break up the long paragraphs. No block over 150 words, one idea per paragraph, each one comprehensible out of context.
4. Turn one comparison into a table. A table is the format machines lift most often and people scan most readily — both at once.
5. Add three real, frequently asked questions at the end, each answered in one or two complete sentences. Use the questions your clients actually ask on the phone, not the ones you imagine.
6. Cite the source of every number. With a link, and with the date you checked it. That's what sets your page apart from generic mass-produced text — and it's exactly what a system looks for before repeating a claim.
7. Measure citation, not just position. Once a month, ask the AI tools what they answer about your topic and note whether your name shows up. It's rudimentary, it's manual, and it's more informative than any dashboard.

None of these steps needs a budget.

Take the page that lost the most traffic in the past year and apply only steps 2 and 3 — the 40-to-75-word opener and the paragraph breaks. Then note the date. That date is what will tell you, once the following weeks' numbers come in, whether the format change made a difference in your case, the only one that matters.