Almost all People Also Ask answers are now AI Overviews

Google's People Also Ask box is now answered almost entirely by AI Overviews. Search Engine Roundtable reported on 9 September 2026 that two independent measurements put the share of AI-generated answers inside that box at or near total, and that Google labels those responses as AI Overviews. Neither measurement came from Google, and both cover English-language queries only.

The two readings, and who produced them

Saeed Khosravi of Allintitle posted on LinkedIn that data from the company's Also Ask Miner tool showed that, starting in August 2026, the People Also Ask answers it collected were 100% AI generated. Search Engine Roundtable quoted that post in its report on 9 September 2026.

Search Engine Roundtable then asked Mark Williams-Cook of AlsoAsked for a reading from a bigger sample. Williams-Cook said AlsoAsked looked at 19.2 million English-language queries made in 2026 and found a sharp month-on-month rise: 97% of People Also Ask answers in the first week of September 2026 were AI Overviews, up from 86% in August 2026. He added, "It seems highly likely we are on track for 100% shortly."

The gap between the two numbers is a sampling difference rather than a disagreement. Allintitle's figure describes what Also Ask Miner captured. The AlsoAsked figure describes a much larger pool of English queries, which is why it lands slightly below the round number. Both point the same direction, and neither figure came from Google.

How fast the change happened

The number that gives this its shape is the old one. Search Engine Roundtable noted that fourteen months before the September 2026 reading, the equivalent AlsoAsked figure sat at about 12%. At that point an AI-generated answer inside People Also Ask was the exception, and most expansions still showed a passage lifted from a ranking page.

The step from 86% in August 2026 to 97% in the first week of September 2026 happened inside a single month. That matters more than the headline percentage for anyone planning work in quarters. A format that changes eleven points in four weeks is not a stable base to plan a year of question-led content against, and the same speed cuts both ways: whatever the mix looks like on the day you read this, it can look different by the time a content batch ships.

The figures reported by Search Engine Roundtable on 9 September 2026 line up like this.

What a People Also Ask result is now

People Also Ask is the expandable list of related questions that Google shows inside a results page. For most of its life, opening one of those rows revealed a short passage taken from a page that ranked for the question, with a link to that page underneath it. The mechanism was extraction: Google picked text that already existed somewhere and showed it.

That is no longer what the box does in the samples described above. Search Engine Roundtable pointed out in its 9 September 2026 report that Google labels these responses as AI Overviews inside the People Also Ask section, which is the same treatment the standalone AI Overview gets at the top of a results page. The answer text is generated rather than quoted, and the generated text can draw on more than one page.

For a publisher, the difference is the relationship between the words on the screen and the page that supplied them. An extracted passage is a piece of one document, so ranking for the question and writing a clean answer paragraph were roughly the same job. A generated answer is assembled, so the page's role changes from being the answer to being one of the inputs to the answer. Search Engine Roundtable did not report how many links each generated response carries or how those links are placed, so that part stays open.

Who this affects, and who it does not

The sites with most exposure are the ones whose organic traffic is concentrated on question queries: how-to libraries, support and documentation hubs, glossaries, comparison pages, medical and financial explainers, and blog programmes built around long-tail questions. Those pages compete for a slot that has changed format underneath them.

Exposure is much lower for queries that rarely trigger a People Also Ask block at all. Brand and navigational searches, transactional searches with a clear commercial intent, and local searches where the map pack dominates the screen behave differently. A site whose demand sits mostly in those buckets can read this story as context rather than as a problem to solve this quarter.

There is a third group that no data here covers: sites in languages other than English. Both samples described by Search Engine Roundtable are English. That is a real limit, not a technicality, and it is the single most important caveat for a Thai-market reader.

What the report did not say

Search Engine Roundtable did not report whether the same pattern holds for non-English queries, and it did not report anything about Thailand or any other specific country. Both measurements are described as English-language samples.

The report did not say what happened to clicks. A change in how an answer is rendered and a change in how often people click through are separate questions, and no click or traffic figure appears in the report. Anyone converting 97% into a traffic forecast is adding an assumption that the source does not supply.

Google did not confirm the numbers. Both figures come from third-party tools, published by the people who run them, and Search Engine Roundtable reported no Google comment. The report also gives no breakdown by topic, by query type, or by device, so statements about which categories moved first are not supported by what was published.

What a marketer can check without waiting for more data

None of these steps require a tool purchase, and all of them produce a record that is specific to one site rather than to a global average.

• Pull the query report in Search Console for the last sixteen months and isolate question-shaped queries, including Thai patterns such as "วิธี", "ทำไม" and "คืออะไร". Compare impressions against clicks month by month, holding position roughly constant.
• Identify the pages that historically earned clicks from those queries and check whether the click rate at the same average position has moved.
• Sample the live results page for the ten question queries that matter most commercially, with location and language set to the market being served, and record whether the People Also Ask expansion carries an AI Overview label.
• In the same sample, record whether the site appears as a cited link inside the generated answer, and which page is cited.
• Repeat the sample monthly and keep the screenshots. A single snapshot cannot separate a rollout from a test, and this format has moved fast enough that last month's screenshot is already history.

Doing this before changing anything has one practical benefit: it turns an argument about a global percentage into a measurement of one site. If a Thai-language question set still shows extracted passages, that is a fact worth knowing before rewriting a content plan around an English-language sample.

What this means for Thai marketers

The honest starting position is that nobody has published the Thai number. The measurements above cover English queries, so a Thai business cannot read 97% as a description of its own results pages. What can be carried across is the direction of travel and the speed, both of which have been visible in AI Overviews reporting for more than a year.

The planning consequence is about where question-led content earns its keep. If a question page is written only to win a snippet slot, its value is tied to a format that is being replaced. If the same page also answers the question well enough to be worth reading after the click, supports a commercial page, and can be cited by an answer engine, it keeps working under either format. That is the practical case for treating generative engine optimisation as part of the same programme as classic rankings rather than a separate experiment.

There is also a measurement job. Most Thai reporting still treats impressions and average position as the health check for a content programme. When the answer is generated above the results, those two numbers can hold steady while clicks fall, which is why AI SEO work starts with tracking citation and appearance rather than rank alone. Teams that add a citation check to their monthly reporting will see the shift in their own market before a global dataset describes it.

For sites where question content is the main acquisition channel, the useful response is editorial rather than technical. Pages that carry original data, first-hand product detail, pricing, local specifics or anything that a generated summary cannot reproduce give a reader a reason to click that a summary cannot satisfy. Building that into an ongoing content marketing plan is slower than chasing a snippet, and it does not depend on which format Google renders next.

Questions marketers are asking about this

Is this live for Thai-language searches?

The source did not say. Search Engine Roundtable described two English-language samples on 9 September 2026 and reported no figures for Thai or any other language, so the only way to know what Thai results pages look like today is to sample them directly.

Did clicks to websites drop because of this?

The report contains no click or traffic data at all. It measured what fills the People Also Ask box, not what happens afterwards, and treating the two as the same measurement would be an assumption rather than a finding.

Has Google confirmed these percentages?

No Google confirmation is mentioned in the report. The 97% figure comes from AlsoAsked and the 100% figure from Allintitle, both published by the companies that produce the tools, and Search Engine Roundtable reported no comment from Google on either number.

Do I have to change anything on my site today?

Nothing in the report forces an immediate change. The reasonable first move is measurement: record how your own question queries render now, keep the record, and let a change in your own data drive the decision rather than a global average from an English-language sample.

Does this mean People Also Ask is being removed?

The report did not say that. It reported that the answers inside the box are now generated by AI Overviews and labelled as such, which is a change in what fills the section rather than a statement about the section disappearing.

Where this leaves a content plan

Two independent samples, published within days of each other, describe a format that has gone from an occasional experiment to the default answer inside People Also Ask in a bit over a year. What is not established is how that behaves outside English, what it does to clicks, and how Google would characterise it. Those gaps are worth stating plainly, because plenty of coverage will round 97% up to a conclusion the data does not carry. The full write-up is at Search Engine Roundtable.

If question-led pages are doing the heavy lifting for your organic traffic in Thailand, the useful next step is a baseline of how your own question queries render right now. Relevant Audience can help set that measurement up and read it month over month.