What Does the Shift to AI Overviews in People Also Ask Mean for SEO?

What changed in People Also Ask?

Google now answers almost every People Also Ask question with an AI Overview instead of a quoted excerpt from one page. Data from AlsoAsked, built from 19.2 million English-language People Also Ask results and reported by Search Engine Roundtable on 2026-09-09, found AI Overviews inside 97% of PAA answers sampled in the first week of September 2026. One month earlier, in August 2026, the share sat at 86%. About fourteen months before that, around July 2025, AI Overviews touched roughly 12% of PAA answers. This has gone from a minor presence to the default format in a little over a year.

People Also Ask has sat inside Google's results since 2015: a stack of expandable questions that reveal a short answer when clicked. That short answer used to be a single sentence pulled from one ranked page, with that page credited underneath. It became one of the most closely watched has in SEO because it was cheap to win and easy to track. Match a question well enough, and Google rewarded the page with a box, a box that carried a click attached to it.

The mechanics underneath the box have changed. Google now synthesizes an answer from several sources at once, using the same generative layer that produces the AI Overview above the organic results, according to Brafton's analysis published 2026-09-18. The question on the page has not moved. What used to be one page's sentence, credited and clickable, is now a blend the model assembles on the spot, with no guarantee that any single source gets named.

The trend line is the real story, not any single snapshot. Going from 12% to 86% to 97% across roughly fourteen months is not a limited test running on a slice of queries. It reads like a completed rollout. Marketers who still picture PAA as a traditional SERP feature, one attributed answer at a time, are working from a version of the page that no longer matches what a searcher sees.

Take a query like 'what is a citation in AI search.' Before this shift, PAA might have pulled a one-line definition from a glossary page and credited that page underneath the question. Now Google is more likely to write its own definition on the spot, stitched from several pages at once, with no single page guaranteed the byline.

1. The question text on the page has not changed.
2. The answer beneath it is now synthesized from multiple sources instead of quoted from one.
3. A named source and a link are no longer guaranteed.
4. The same generative layer that powers the AI Overview above organic results now powers PAA.

Who does this change affect?

Anyone who built a page to win a single question feels this hardest. That tactic depended on Google needing a source to quote, and Google increasingly does not need one. A page written to answer 'what is X' in 200 words, with no other reason to exist, loses the mechanism that made it worth building: an attributed click.

Four groups sit closest to the change. Content teams that track PAA wins as a KPI will watch that number keep falling, not because their pages got worse but because the format stopped crediting single pages at all. SEO tools and dashboards that still report PAA presence as a ranking win are measuring something that no longer means what it used to. Brands that rely on question-based long-tail queries, things like 'how much does X cost' or 'what is the difference between X and Y', are the ones most likely to see their answer folded into a synthesized box instead of linked out. And comparison or 'best X' pages built to win one question lose the traffic path that funded them.

The loss is not even across the board. Pages that already went past the bare-minimum answer, ones with original data, a comparison table, or a number nobody else has published, are more likely to be the source an AI Overview draws from and names. Thin pages lose. Pages with something worth quoting have a better shot at staying in the answer, even if the click itself gets rarer.

This mirrors what happened when Google's AI Overview rolled out across standard search results starting in 2024: sites with generic, single-purpose pages saw the sharpest declines, while sites with depth held up better. PAA is now running the same test at the level of a single question instead of a whole results page.

Local and service businesses feel a narrower version of the same shift. A 'how much does X cost in [city]' question used to reward one local page with a snippet and a click. Now that answer is more likely to get folded into a synthesized box, which means the businesses that show up are the ones with a specific, current number to offer, not the ones that merely repeat the question in a title.

1. Several of your top pages by traffic are built around one question each.
2. Your SEO dashboard reports PAA appearances as a core KPI.
3. Your content briefs stop at answering the question, without adding data, comparison, or opinion.
4. Your organic traffic on question-shaped queries has dropped while your rankings held steady.

How should you respond?

Stop building pages to win a question, and start building pages worth citing. The old game matched a query to a title. The new game gives the model a reason to name a source when it writes the answer, because the model can generate a plain restatement of a question without any help.

PAA still does a job worth keeping. It maps the exact phrasing searchers use and it surfaces the follow-up questions a single article would not anticipate on its own. Keep mining it for that. What changes is what gets built in response. Instead of a short page answering one PAA question, build a page that answers the question and then goes further, with an original stat, a dated comparison, or a number pulled from your own data rather than a rewrite of the top three results.

Track a different number alongside the old ones. PAA appearance rate tells you whether Google borrowed your sentence for a box that may not name you. What matters now is whether ChatGPT, Perplexity, Gemini, and AI Overviews cite you by name when someone asks the underlying question, what Omnicite calls citation share. A page can lose every PAA slot and still be the most-cited source for its topic if it is the page the models keep pulling from.

Practically, start by auditing existing question-and-answer pages for what they add beyond the question itself, and fold the thin ones into fuller resources instead of leaving one page per question. Watch brand mentions across AI answers the same way you would watch rankings, because a citation with no link is still worth more than a snippet nobody reads. Neither habit requires abandoning PAA research. Both require treating the box underneath the question as a competitor for attention, not a guaranteed outcome.

None of this calls for panic or a rebuild of every page overnight. Google still needs sources to draw from, even when it stops naming them individually inside a PAA box. Sites disappear from that particular box before they disappear from AI answers generally, and the two are worth tracking separately.

1. Keep mining PAA for the exact phrasing and follow-up questions searchers use.
2. Consolidate thin, one-question pages into fuller resources with original data or comparison.
3. Add a dated stat, table, or data point to any page built mainly to answer one question.
4. Track citation share across ChatGPT, Perplexity, Gemini, and AI Overviews, not just PAA appearance rate.
5. Build for the underlying question instead of trying to game a single AI feature.

Sources

Source: Search Engine Roundtable

AI Overviews appeared in 97% of English-language PAA answers sampled in the first week of September 2026, up from 86% in August 2026 and about 12% roughly fourteen months earlier. Search Engine Roundtable, 2026-09-09

Source: Brafton

Google is increasingly using AI Overviews to answer People Also Ask questions, and marketers should build content that goes beyond the bare answer to keep visibility. Brafton, 2026-09-18