ENTRY 024 · ANSWER-ENGINES · By Answer Engineered Research
12.6%97%
One Tracker Says 100%. The Other Says 86%. Same Feature, Same August.
Search Engine Roundtable reported People Also Ask is nearly 100% AI-answered. The two datasets behind that one headline disagree by 14 points on August.
What the article actually printed
Everything in this section is quoted from the live page, re-read on 10 September 2026. It is a short post and the relevant text is nearly all of it.
Schwartz’s starting point was Allintitle:
I first spotted this based on a post from Saeed Khosravi from Allintitle who posted on LinkedIn saying, “Allintitle’s Also Ask Miner data shows that starting August, 2026 Google People Also Ask Answers Are 100% AI Generated.”
He then went and asked the other tracker:
So I asked Mark Williams-Cook about AlsoAsked data, the one that saw it at 12% 14 months ago, and he looked at its almost 20 million sized dataset and said it was almost at 100% but not there yet, it is at 97%.
And Williams-Cook’s own figures, as quoted:
AlsoAsked took a sample of 19.2M English language made in 2026, and found a dramatic monthly increase in AI-generated answers. 97% of PAA answers in the first week of September are AIOs, up from 86% in August.
Two things in that passage are worth slowing down on. First, the 97% was produced on request, for this article. It is not a published study with a methodology section attached; it is a founder checking his own dataset because a reporter asked. That is a perfectly normal way for a trade-press number to come into existence, and it is worth knowing when the number starts getting cited elsewhere as though it were research.
Second, “19.2M English language made in 2026” is missing a word in the published quote, almost certainly queries. What is not missing is the scope: English, 2026, AlsoAsked’s own tracked set. Not a random sample of Google.
The 14-point gap
Set the two claims against each other for the one month both of them cover.
| Allintitle | AlsoAsked | |
|---|---|---|
| Claim about August 2026 | 100% | 86% |
| Most recent figure given | 100%, from August onward | 97%, first week of September |
| Stated sample | Not stated in the article | 19.2 million English-language queries, 2026 |
| Stated method | Not stated in the article | Not stated in the article |
| How the figure reached print | Public LinkedIn post | Supplied on request to the reporter |
The right-hand column is thinner than it looks, and the middle column is thinner still. What is genuinely on the page is a number from each company and a sample size from one of them. Everything else a reader might want, what counts as an AI-generated answer, what geography, what device, what query mix, is absent from both.
Fourteen points is not a rounding difference. On a metric expressed as a share of a feature, in the same month, it is the difference between a feature that is finished and a feature that is not. The article does not reconcile it, and both figures appear under a headline that quietly resolves it in favour of the higher one.
The one number with a series behind it
There is a real trend here, and it belongs to AlsoAsked, because AlsoAsked is the only one of the two with a public reading on the same metric from more than a year ago.
Search Engine Roundtable covered that earlier reading on 3 July 2025:
Data from Mark Williams-Cook of AlsoAsked said that 12.6% of the people also ask results are AI Overviews, the rest are featured snippets.
12.6% in July 2025. 86% in August 2026. 97% in the first week of September 2026. That is the same source, the same company, the same metric name, and it is the strongest evidence in this story. The direction is not in dispute by anyone. Google’s People Also Ask box has gone from mostly quoted web snippets to mostly generated text in about fourteen months.
Notice what that series does to the argument, though. It does not need the 100% claim at all. The interesting fact is the slope, and the slope is fully carried by one tracker’s own numbers. The second tracker adds a rounder headline and a discrepancy, and not much else that a reader can check.
What we could not verify, and why the post stops here
Both underlying LinkedIn posts are linked from the Search Engine Roundtable article. Neither could be retrieved during this run in a form worth quoting from. Automated scraping of those URLs was refused outright, and the fallback retrieval path returns a model-generated summary of a page rather than the page, which is not a source.
That matters because the most interesting question in this story is a methodology question, and the answer to it is sitting in those two posts. Specifically: does either company classify the content of a People Also Ask answer, or does one of them infer the result from a change in how Google serves the feature? Those are different measurements and they would explain a 14-point gap cleanly.
We are not going to guess at it. Several plausible-sounding methodology details circulated during research for this post, and none of them survived the requirement that a number be read on a page that actually loaded. So this post claims only what Search Engine Roundtable printed, and the methodology question stays open. If either company publishes its method, the gap probably closes in a sentence.
Nobody in this chain measured a click
This is the caveat that keeps getting lost when the number travels.
Every figure above describes what is present in a box. 12.6%, 86%, 97%, 100%: all of them are about whether the answer in a People Also Ask accordion was generated or quoted. None of them is about whether anyone clicked, whether traffic moved, or whether a publisher lost a visit. No click, session, or referral figure appears anywhere in either article.
That is not a criticism of the trackers, who measured what they said they measured. It is a warning about the sentence this number will turn into. A feature going from 12.6% to 97% generated is a real, large change in what Google shows. It is not, by itself, evidence of a traffic outcome, and there is nothing in this reporting that can be converted into one.
If you track People Also Ask, three things follow
Stop reporting the box as a ranking win. If your dashboard counts People Also Ask appearances as visibility, that count now mostly measures the presence of generated text near your topic, not the presence of your text in it. The metric changed underneath the label.
Ask any vendor which of the two things they count. Presence of an AI answer in the box, and attribution of that answer to a source, are different products. After fourteen months of this shift, a tool that has not disclosed which one it reports is not usable for a before-and-after comparison.
Treat the 100% as a claim, not a reading, until a method appears. The 97% has a sample size next to it. The 100% does not, in the reporting available. That is the whole difference in confidence, and it is enough of one.
Sources
- Search Engine Roundtable, “Almost 100% Of People Also Ask Results Are Answered By Google AI Overviews”, 9 September 2026. Every quote in the first three sections is from this page.
- Search Engine Roundtable, “People Also Ask Results Are Google AI Overviews”, 3 July 2025. The 12.6% baseline.
Both pages were fetched and read on 10 September 2026. The LinkedIn posts they cite were not retrievable and are therefore not quoted here.