Google's People Also Ask feature may be undergoing a much deeper transformation than a normal SERP redesign. According to new third-party tracking, the expandable questions that have long exposed short web-derived answers are now overwhelmingly being answered through Google's AI Overview system.
Search Engine Roundtable reported on September 9 that AlsoAsked analyzed a sample of 19.2 million English-language People Also Ask questions collected during 2026. According to AlsoAsked founder Mark Williams-Cook, 97% of PAA answers in the first week of September were classified as AI Overviews, up sharply from 86% in August.
If the measurement accurately reflects Google's broader English-language PAA ecosystem, the change is significant. People Also Ask would no longer be best understood as a collection of expandable featured-snippet-like answers. It would increasingly function as a distributed network of small generative answers embedded directly inside ordinary search results.
There is an important qualification from the outset: 97% is not an official Google statistic. Google has not announced that 97% of PAA answers are now AI Overviews or confirmed a universal migration. The figure comes from AlsoAsked's large observational dataset and should be treated as third-party measurement of what its system saw, not as a formal product rollout percentage.
AlsoAsked says the shift accelerated dramatically in August and September
The scale of the reported change is what makes the data difficult to dismiss as an isolated interface test. Search Engine Roundtable says Williams-Cook examined an almost 20-million-question dataset after seeing separate evidence that People Also Ask answers were becoming AI-generated.
AlsoAsked's sample consisted of 19.2 million English-language questions collected in 2026. The share classified as AI Overviews reached 86% in August and 97% during the first week of September.
Williams-Cook said the trajectory suggested the percentage could soon approach 100%. That is an inference from the observed trend, not confirmation that Google intends to convert every PAA answer to an AI Overview.
A separate dataset from Allintitle, cited by Search Engine Roundtable, reportedly showed PAA answers reaching 100% AI-generated beginning in August. The two measurements use different systems and should not be treated as one combined study, but they point in the same general direction: AI-generated answers are rapidly becoming dominant inside People Also Ask.
AlsoAsked can explicitly identify AI Overviews in PAA data
The measurement is more concrete than simply looking for answers that “sound like AI.” AlsoAsked's current export documentation shows that its paid datasets include an “is AI overview” field, represented as a boolean value distinguishing conventional web-result answers from AI Overviews.
The exports can also include the URLs and page titles cited by the AI Overview. That means the platform is tracking not only the related question but the type of answer Google returns and, where applicable, the sources attached to the generated response.
This does not independently validate every classification in the 19.2 million-question analysis, and AlsoAsked has not published a peer-reviewed methodology for the September percentage. It does establish that AIO detection is a formal part of the company's current PAA data model rather than a manual estimate based on a handful of screenshots.
For SEO teams, that distinction matters because the reported change can be monitored programmatically. PAA is becoming measurable as an AI-search surface rather than only as a traditional SERP feature.
People Also Ask has historically been a map of Google's question graph
People Also Ask has been valuable to search marketers because it reveals how Google connects one query to adjacent questions. Expanding a question often generates additional related questions, creating a visible representation of how user intent branches from a starting topic.
AlsoAsked itself describes PAA data as a way to understand Google's view of intent and the questions likely to matter after an initial search. The company has built its product around collecting those relationships and turning them into question graphs for research and content planning.
Traditionally, the answer beneath a PAA question often resembled a featured snippet: Google extracted a passage from a webpage, displayed it prominently and linked back to the source.
An AI Overview changes the nature of that answer. Instead of selecting one passage from one page, Google can synthesize a response from multiple sources and present a generated explanation with citations.
The question can stay the same while the answer system changes completely
This is why the reported 97% figure matters more than a cosmetic label. From the user's perspective, People Also Ask can still look familiar: a list of expandable questions beneath the main results. Underneath that interface, however, the answer-generation mechanism may have changed.
A conventional PAA answer is primarily an extraction problem. Google identifies a page and extracts a passage it believes answers the question. A generative PAA answer is a retrieval-and-synthesis problem. Google can retrieve several sources, decide which claims matter and compose a new answer.
Those are different visibility systems. A publisher that previously won a PAA result by providing the extractable paragraph may now be competing to become one of several sources used in an AI-generated answer.
The question graph remains useful, but the optimization target shifts from “win this snippet” toward “be a credible source for the answer Google constructs.”
PAA optimization and AI Overview optimization may be converging
SEO teams have traditionally treated People Also Ask, featured snippets and AI Overviews as separate SERP features. If almost all PAA answers are now AIOs in a large English-language sample, that separation becomes harder to maintain.
The content qualities that support generative retrieval—clear entities, explicit factual statements, strong topical context, reliable sourcing and passages that answer specific questions—can increasingly matter inside PAA as well.
This does not mean there is one universal “AIO optimization formula.” Google has not published a special ranking checklist for PAA-generated answers, and the sources selected for one AI surface can differ from those used elsewhere.
But it does mean that PAA research can feed directly into GEO strategy. The questions reveal the information users may ask next, while the generated answers reveal which sources Google currently trusts enough to synthesize for those questions.
The change could multiply the number of AI answers on a single SERP
Discussion of AI Overviews often focuses on the large generated answer that can appear near the top of a Google result page. People Also Ask introduces a different scale problem.
A SERP can contain several PAA questions, and expanding one can expose additional questions. If each expansion triggers or reveals an AI Overview, a single search journey can contain multiple generative answers rather than one headline AIO.
That creates many more moments in which Google can answer a user's follow-up without requiring a conventional organic click. The effect may be especially important for informational searches where users move through a chain of increasingly specific questions.
From a publisher perspective, the relevant unit is no longer simply “does this query trigger an AI Overview?” It can also be “how many AI-generated follow-up answers are available within the same search session?”
The SEO opportunity also multiplies
More generative answers do not only create zero-click risk. They create more citation opportunities.
A page may never become the primary organic result for a broad query but still provide the best evidence for one narrow PAA question. If Google generates an answer for that question, the page can potentially receive source visibility in a part of the SERP that would previously have belonged to a single snippet winner.
This makes long-tail question coverage more strategically interesting. Content does not need to be organized as a mechanical FAQ containing every imaginable query, but it should answer the important subquestions users genuinely have about the topic.
The strongest content can become useful at several layers: the main organic result, a conventional snippet, an AI Overview and one or more AI-generated PAA expansions.
Question research may become a proxy for query fan-out
One reason People Also Ask remains valuable in an AI-search environment is that it exposes related intent directly from Google. Generative systems frequently decompose complex questions into subquestions or fan out searches to gather supporting information.
PAA is not the same thing as Google's internal AI query fan-out system, and it would be a mistake to claim that every visible PAA question is literally used as a hidden AIO subquery. But both reflect the broader problem of expanding one user need into a network of related information requirements.
AlsoAsked's own February 2026 research on PAA coverage and organic rankings found that, among top-five organic positions in its study, pages answering more related PAA questions tended to rank higher. The researchers explicitly cautioned that correlation does not make PAA coverage a standalone ranking factor.
For content strategy, the more durable lesson is that related questions expose semantic coverage. A page that can answer the important branches of a topic is useful whether Google surfaces those branches through organic ranking, PAA or generative search.
Do not turn 97% into “Google has officially replaced PAA”
The most important editorial restraint is avoiding a stronger claim than the data supports. AlsoAsked measured 97% of answers as AIOs in its English-language sample during the first week of September. That is not equivalent to Google announcing that 97% of all People Also Ask answers worldwide have been converted.
Search results vary by country, language, location, device, query and experiment. AlsoAsked itself notes in its documentation that PAA results can differ according to region and language settings.
The dataset is extremely large, which makes it valuable evidence of a major shift, but sample size does not eliminate questions about representativeness. The distribution of searches collected through AlsoAsked may differ from the distribution of all Google searches.
The defensible conclusion is therefore that AI Overviews appear to have become dominant in the PAA answers observed by one of the largest third-party PAA datasets. Whether Google has reached the same percentage globally requires broader measurement or official confirmation.
The speed of the increase may be as important as the absolute number
Even if future independent measurements produce a lower absolute percentage, the movement from 86% in August to 97% in the first week of September suggests a rapid deployment pattern inside AlsoAsked's dataset.
Search Engine Roundtable notes that roughly 14 months earlier, the same data source had observed only about 12% of PAA answers being AI Overviews. Moving from a minority of answers to nearly all observed answers in little more than a year would represent a substantial architectural transition.
That pace matters for SEO operations because historical assumptions can become obsolete before annual strategies are updated. A team that still treats PAA as a conventional featured-snippet opportunity may be optimizing for a result format that has already changed underneath it.
SERP-feature monitoring needs to record not only whether People Also Ask appears but what kind of answer appears after expansion.
Source tracking inside PAA is becoming more important
If PAA answers are generated through AI Overviews, the source panel becomes a new competitive dataset. Instead of asking only which domain “owns” a PAA answer, analysts can inspect which domains Google cites across thousands of related questions.
That can reveal specialist publishers that repeatedly earn citations even when they do not dominate traditional rankings. It can also expose categories where large authoritative domains capture most generative visibility.
Because AlsoAsked exports AI Overview URLs and page titles, the workflow can move from manual SERP inspection to structured analysis. Teams can group citations by domain, topic, question tier and intent, then compare those patterns with organic ranking data.
The key is not to assume that citation frequency equals traffic. A source can be cited without receiving a click, and Google does not expose the complete user-behavior data needed to translate PAA citations into visits.
Content teams should stop treating PAA as an FAQ generator
One of the weakest uses of People Also Ask data has always been copying a list of questions into an article and answering each with a thin paragraph. The rise of AI-generated PAA answers makes that approach even less compelling.
If Google can synthesize a concise answer from several sources, merely reproducing the question as an H2 is not a competitive advantage. The source needs to contribute something worth retrieving: precise facts, original data, first-hand experience, authoritative explanation, useful comparisons or unusually clear treatment of a difficult subtopic.
PAA data is better used as a map of information needs. Writers can cluster related questions, identify the underlying concern and build a coherent section that resolves the concern more completely than a collection of disconnected FAQ fragments.
This also produces better human-readable content. The goal is topical completeness, not keyword-shaped formatting.
AI-generated PAA answers raise another attribution question
A conventional PAA snippet makes attribution relatively simple. One source provides the displayed passage, and the link is usually closely associated with the text.
Generative answers can cite several sources while synthesizing them into one response. That makes it harder for a user to know which source supports which statement and harder for a publisher to understand how much of its work influenced the final answer.
The distinction between citation and utilization becomes important. A domain may appear among the cited sources but contribute little to the generated wording, while another source may underpin several central claims.
As PAA becomes more generative, source-quality analysis should therefore move beyond counting citations. The stronger question is whether the system accurately represents the source and preserves important qualifications.
PAA can now be part of an AI visibility benchmark
For brands tracking generative visibility, People Also Ask deserves to become a separate measurement surface. A useful benchmark can start with the important questions associated with the brand's topics, then record which questions generate AIO answers and which sources those answers cite.
That dataset can be compared with standard AI Overview tracking for the root query. The same domain may perform differently in the main AIO and the PAA expansions because the questions require different evidence.
Teams can also monitor change over time. If Google's answer source changes after content is updated, that is more informative than checking a single screenshot. Repeated sampling is particularly important because generative results can vary between runs.
The result is a more realistic picture of search visibility as a network of answers rather than one ranked list.
The bigger shift is from search features to answer infrastructure
People Also Ask was once easy to categorize: it was a SERP feature containing related questions. AI Overviews were another feature. Featured snippets were another. Organic results were the underlying list.
Those boundaries are becoming less useful. Google's generative answer system can increasingly appear inside multiple parts of the search experience, turning AI from one box at the top of the page into an answer layer distributed throughout the SERP.
The reported 97% figure is a striking illustration of that transition. It does not prove that every PAA answer everywhere is now generated by AI, and Google has not officially confirmed the percentage. But a 19.2 million-question third-party dataset showing 97% AIO classification is large enough that SEO teams should not treat the change as a curiosity.
If the trajectory continues, optimizing for People Also Ask and optimizing for AI Overviews will increasingly become parts of the same problem: understand the questions users ask, publish evidence worth retrieving, make important information explicit and monitor whether Google's generative layer chooses your content when it builds the answer.