13.4 Billion Web Questions Show FAQ Frequency Is Not Search Demand

The top 1,000 questions became overwhelmingly manufactured

The paper You Can Tell Who’s Asking examines 110 FineWeb snapshots spanning 2013 through 2025. The authors distinguish questions that appear in discussion-like human contexts from questions that appear to have been manufactured for navigation, conversion or page templates.

Among the 1,000 most frequent questions, the classified manufactured share rose from 71.7% in 2015 to 86.8% in 2020 and 93.2% in 2024.

Those percentages do not describe Google searches, support tickets or spoken questions. They describe a classified subset of question occurrences found on crawled web pages.

Why a repeated FAQ can look like demand

A product template can place the same question on thousands of URLs. A marketplace can generate location and category combinations. A publisher can copy a standard FAQ block across an entire section. Each appearance increases web frequency without adding another person who wants the answer.

This creates a supply-side echo. Keyword tools, scraped question databases and competitor audits may repeatedly encounter the phrase because publishers reproduce it, not because independent demand keeps rising.

The paper’s classifier reached an area under the curve of 0.725 for genuine versus manufactured questions, but only 0.554 when separating genuine questions from commerce FAQ questions. That weaker distinction is important. Some useful customer questions live inside commerce pages, and some human-sounding questions are produced by templates. A single automated label should not decide a content roadmap.

Use four signals before turning a question into a page

A strong candidate does not need all four signals, but it needs more than repetition on competitor pages. Record which signal supports the decision so an editor can revisit it later.

Choose the smallest useful publishing unit

When the question is a step inside a larger task, answer it in the relevant guide. When it needs a distinct method, tool, dataset or decision, a dedicated page may be justified. When the wording is merely a variation of an existing answer, improve the existing page and add an internal anchor.

• Dedicated article: distinct intent, substantial answer and a useful next action.
• Section in an existing guide: supporting question within the same reader journey.
• Product or support copy: answer belongs at the decision point, not in a search article.
• No new content: copied wording, no independent demand signal and no meaningful answer gap.

This choice also reduces thin FAQ archives. It gives one strong page more complete coverage instead of scattering small answers across many indexable URLs.

The long-term trend is real, but its size is uncertain

The raw data suggests a large decline in the share of genuine questions. After the authors controlled for observable crawl and content features, their estimates were smaller, roughly in the 42% to 56% range depending on specification. They describe about 45% as a useful best estimate while emphasizing uncertainty.

A large residual share, roughly 61% to 73% in reported classifications, remains unknown. The corpus is English-language FineWeb data, and crawl composition or filtering can change what is visible. The analysis therefore supports a direction, not a precise census of every question on the web.

That limitation strengthens the editorial lesson. If even a very large crawl cannot cleanly identify human demand from page text alone, a publisher should not let an FAQ scraper make the final content decision.

A five-minute editorial test

1. Write the decision or task the answer will help a reader complete.
2. Attach one behavioral signal from search, customers, community discussion or product use.
3. Search the site for an existing page that should own the answer.
4. State the information gain: original measurement, worked example, tool, comparison or judgment.
5. Do not publish if the only evidence is that many other pages use the same question.

For related planning work, our evidence-led publishing guide shows how to connect a claim to inspectable support. The same discipline belongs at the topic-selection stage.