People Also Ask is being swallowed by AI Overviews

In a little over a year, the related questions box, the one that used to send people to a list of links, started returning a generated answer almost every time it shows up.

Limits of this source. The number circulates from a LinkedIn post, with no methodology report. Independent outlets repeat the same three data points, which confirms what was said without filling the methodology gap. The sample is declared as English only, so there is no basis to extend the reading to other languages or markets. There is also a competing measurement, by Saeed Khosravi with Allintitle data, that claims 100% as early as August 2026, and nobody has explained the gap between the two.

For almost a decade, the related questions box was the cheapest door into informational content. Entire teams built their editorial calendar from it, because it handed over the real question in the real language and still sent traffic to whoever answered well. It was the closest thing SEO had to a free lunch, and we all know how free lunches end.

What the PAA absorption curve shows, and what it does not

Three published data points support one claim and only one: absorption went from marginal to almost total in a little over a year, with most of the jump concentrated at the end of the period. Going from 86% to 97% in a single month is the strongest number in the set, because it measures a short and recent interval.

What these three points do not support is any reading about the path between them. There is no published monthly series, so claiming acceleration, a pullback or a phased test would mean inventing the shape of a curve from its two ends. I looked for the series in four outlets and in the primary source, and found nothing.

What PAA absorption changes in content work

• Topic research based on PAA is still a valid source of questions and has lost value as a traffic forecast, because the click now lands somewhere else.
• Content that only answers the question loses its job. What keeps value is the answer that needs context, first-party data or judgment, which is exactly where the generated answer falls short.
• The goal of ranking inside the questions box gives way to the goal of being cited inside the answer, with work on making each passage stand on its own.
• Reports that still count PAA presence as a success metric need a review before the next reporting cycle, with AI visibility tracking sitting next to the search report.

How to write for the question that now gets answered on screen

If people keep asking the question and the system now assembles the answer, the text with a chance of being used is the one that solves the question in a passage that works on its own. Three writing adjustments make most of the difference.

• The question becomes a subheading, phrased the way people actually ask it, and the answer comes in the first paragraph below it, with no warm-up.
• The answer passage names its subject and brings the source along with the data, because the system may extract that paragraph alone.
• After the direct answer comes what generated answers usually leave out: the condition under which the rule fails, the example from your own market, the number with a date.

The third adjustment is what justifies the visit once the short answer is already on screen, and it is what separates the cited page from the page the system merely read. To know whether it worked, the metric is citation, with the same set of questions repeated over the weeks, as described in AI search KPIs.

What I would do with the portfolio this week

List how many pages in the portfolio depend mostly on simple informational traffic, and measure how impressions and clicks on those pages moved throughout 2026. The comparison that matters is between that behavior and the jump reported at the end of the period.

The payoff of this measurement is producing something that has not been published: a series for your own market and language, with a declared methodology. If the two curves agree, the quarter’s priority moves from production to differentiation. If they do not, that is a finding too, and it is worth more than repeating someone else’s number in yet another slide deck.