People Also Ask questions come from real searches. Google’s own patent describes a store of questions people have actually typed, matched to your query through what searchers clicked. This page sits under how accurate People Also Ask information is because Google put it there. When we searched “How accurate is the ‘People also ask’ information?”, one of the child questions Google showed underneath was “Where do the people also ask questions come from?”.
Google has never published the full recipe. Here’s what it has said, what its patents describe, and what we saw in 149 real questions harvested on 27 September 2026.
What Google says in its documentation
Not much. Google’s Visual Elements gallery, its official guide to the parts of a results page, defines a related questions group as “a cluster of questions that are related to what the user initially searched for (also known as ‘People Also Ask’)”. It adds that site owners can’t control what shows up there (Google Search Central). Google’s featured snippet documentation explains how the answers get picked. On where the questions themselves come from, the documentation is quiet.
What Google’s related-questions patent describes
The clearest public description is a Google patent, “Generating related questions for search queries” (US9213748B1), filed in March 2013 and granted in December 2015 (Google Patents). It describes a question database of “search queries that have previously been submitted by users” that “have been determined to be in question form”, for example because they contain a question word. Optionally, the database can also hold questions found in published web content and on question-and-answer sites.
To choose which questions to show, the patent starts from the pages ranking for your search. For each page it builds topic sets from “previously submitted search queries that have resulted in users selecting search results identifying the search result resource”. In plain terms: the searches that led people to click that page. Then it pulls questions matching those topics. So the questions are tied to real behavior twice. People typed them, and they’re matched through what people clicked.
One caution. A patent shows what Google invented, not what runs today. It’s the best public evidence, not a spec.
Where query refinements fit in
A second patent looks at the follow-up search. “Generating query variants using a trained generative model” (US11663201B2, granted May 2023) learns from pairs of queries where a person searched, then searched again. One of its examples: “did leonardo da vinci paint mona lisa” followed by “who commissioned leonardo da vinci to paint the mona lisa” (Google Patents). Its background section cites “people also search for” as one use of similar-query collections. It doesn’t name People Also Ask, and as Glenn Gabe wrote when he covered it, “we don’t know if Google has implemented this yet, or if they will” (GSQi).
Still, it fits how the box behaves. Open a question and new ones appear under it, much like a person refining a search. In our harvest the box started with 4 visible questions and grew to 50 after we opened about 25 of them.
Where do the People Also Ask come from?
That exact wording is a PAA question too. It appeared at position 8 in our “people also ask” harvest, and Google’s answer said the questions “come from real user search behavior and machine learning analysis of related queries”. It cited AlsoAsked, Zenbrief, SEOZoom, Advanced Web Ranking and Semrush, among others.
Look at who it cited, though. The claim that Google “studies billions of actual searches” was credited to Zenbrief’s blog, not to Google. The same answer said each answer box “includes a link to the original website where the text was found”, which describes the older one-snippet format (see where People Also Ask answers come from for how that changed). So the headline holds up (real searches), but the detail in that answer is the SEO industry’s explanation, repeated back by Google.
What we saw in 149 questions
Our harvest backs up the “real searches” idea in a few visible ways.
- The wording is human. “What the heck is SEO?” and “Do SEO need coding?” are not questions an editor would write. They read like things people type.
- Near-duplicates pile up. “Can I see who googled me?”, “Can I see if anyone has googled me?” and “Can I tell who has googled my name?” all showed up in the same box, along with three phrasings of the question about Google’s biggest enemy. Different people word the same worry differently.
- The questions follow searchers, not the topic. For “people also ask”, only 6 of 50 questions were about PAA itself. The first off-topic one, “Can I tell who has googled my name?”, came at position 5, and by position 13 Google was asking “Who is the #1 Googled person?”. For “people also ask seo”, 0 of 50 were about PAA.
That last pattern is what you’d expect if questions come from what people search next, not from a list of facts about the topic. Our page on why People Also Ask drifts off topic goes deeper.
Why this matters for accuracy
A question in the box means people ask it. It doesn’t mean the premise is true. “What is Google Dark Secret?” is a real PAA question with a full answer under it, and that says nothing about whether a secret exists. Read the questions as a demand signal: what your readers wonder about, in their words.
To see which question spawned which, PAA Box unpacks the box from D0 “Lid off” to D3 “All the way”, and its CSV export keeps the parent question next to each child. Parent-to-child edges like that are how this site decides where each page lives.