Blog · Content that gets cited

FAQ content that is not filler

Published 28 September 2026 · Updated 12 October 2026 · 9 min read

FAQ content that is not filler
Photograph: Kelsy Gagnebin / unsplash

The short answer: An FAQ earns its place when every answer carries a fact that appears nowhere else on the page and reads correctly with the rest of the page deleted. Filler is the opposite: five questions nobody asked, answered in sentences lifted from the paragraph above. Google's own rating guidelines now have a section for this, and retrieval raises the stakes for a second reason, because what an engine quotes back is one answer block rather than your whole page.

Two forces arrived at the same conclusion from opposite directions, which is why this is worth a thousand words rather than a bullet.

Google wrote the definition down

On 11 September 2025 Google published the current General Guidelines, the document its quality raters work from. Section 5.2.2 is titled "Filler as a Poor User Experience" and defines filler as low-effort content that adds little value and does not directly support the purpose of the page. The sentence that should worry anyone with a bolted-on FAQ is the next one: filler can artificially inflate content, creating a page that appears rich but lacks content website visitors find valuable.

The guidance then gives raters two conditions for a Low rating. A page gets one if it holds a large amount of low quality, unhelpful filler, or if the filler is visually prominent enough that the useful material becomes hard to find. The same document is explicit that no single rating can move a page up or down in Google Search, and says ratings are used instead to measure how well the systems are working and to improve them. So this is Google's written vocabulary for quality, not a ranking factor anyone outside Google can observe. It is also Google's document alone. We have found no published rater guidance of this kind from the other engines, so nothing here should be read as a cross-engine rule.

A generated FAQ sitting under every product page fits that description precisely. It is long, it looks like substance, and it answers questions that were invented to fill the template.

The retrieval reason, which is sharper

Set the rating guidelines aside. An answer engine does not read your page the way a rater does. The retrieval-augmented pattern these systems are built on cuts documents into passages, keeps a handful, and writes from what it kept. A question heading followed by a short self-contained answer is close to an ideal passage: it carries its own question, its own subject and its own fact, and it survives being pulled out of context. That is the whole mechanical argument for writing an FAQ in 2026, and it has nothing to do with rich results. See how chunking and passage retrieval work for the shape of that cut.

Now look at what the average FAQ sends into that process. Answer one says "our platform is designed to help teams of all sizes". Pulled out of the page, that passage names no company, states no fact, and answers no question. It competes against a passage from somebody else's site that says "the starter plan is 12 dollars per seat per month, billed annually, with no minimum". Only one of those is quotable.

Be plain about the limits of that picture. Passage boundaries, ranking and deduplication happen inside indexes that are not published, so the paragraphs above describe retrieval as the engines and the research literature describe it, not something we have inspected on any particular engine.

There is a second, less obvious loss, and this one is inference rather than observation. Most content systems paste the same FAQ block onto hundreds of URLs. A retriever that collapses near-identical passages would be holding one copy, attributed to whichever URL it happened to keep, with the remaining pages contributing nothing. We cannot confirm that from outside, because the engines do not expose which passage they kept or which URL they credited. The practical response is to treat a duplicated block as unearned rather than to assume it is working.

The four tests

Run these on an existing FAQ before you write a new one.

The deletion test. Delete the FAQ. If nothing factual left the page, the FAQ was restating the body copy and should go. An FAQ is a place for the facts that did not fit the narrative: the edge cases, the limits, the awkward comparisons, the thing support gets asked every Tuesday.

The cover test. Cover the question with your thumb and read the answer alone. If it opens with "It does" or "Yes, we do", it will not survive retrieval. Each answer should restate its subject by name in the first clause. The habit feels repetitive on the page and reads correctly everywhere else, which is the trade you want.

The provenance test. For each question, name where it came from: a support ticket, a sales call recording, a site-search log, a query that returned your page. A question you cannot source is a question you invented. "Why choose us?" is the signature of this failure: it is a sentence a brochure writes for itself, and in the audits we have run it has not once arrived from a support ticket or a site-search log.

The expiry test. Scan each answer for a number, a price, a version or a date, then ask when it will become false. An FAQ answer that goes stale does not just sit there harmlessly. It becomes a wrong fact in a clean retrievable block, which is the most quotable shape a wrong fact can take. Mark the volatile answers and review them on a fixed date.

The self-check has the broader version of this pass, ten things you can test in an afternoon without tooling.

What the markup is still for

The rich result is gone. Google's structured data gallery, last updated 15 June 2026, no longer lists FAQ or How-to among its supported features, and the company's guidance on AI features, last updated 10 December 2025, says plainly that no special schema.org structured data is needed to appear in AI Overviews or AI Mode.

So why keep FAQPage? Because it is still a current, maintained type at schema.org, carried through version 30.1 dated 16 September 2026, and because it is readable by any parser that chooses to consume it. Which parsers do, and what they make of it, is not visible from outside, so treat that as a reason to keep the markup correct rather than a benefit you can measure. Structured data that honestly describes visible page content is a cheap, machine-readable statement of which text is a question and which text is its answer. What it will not do is rescue a weak answer. Markup describes content, it does not upgrade it.

One rule survives from the rich-result era and is worth keeping. Google's structured data general guidelines, last updated 10 July 2026, tell you not to mark up content that is not visible to readers of the page, and illustrate it with markup describing a performer the HTML body has to describe too. An FAQPage node holding answers a human visitor cannot see is a correctness problem, not a tactic. That is Google's policy and we have found no published equivalent from the other engines, so read it as the strictest rule in play rather than the only one.

Where FAQ content actually wins

Three patterns recur in our audit work. That is a qualitative reading of the audits we have run, not a counted frequency: we have not measured how often each appears.

The first is the objection FAQ on a pricing or plan page. Buyers ask engines about contract terms, minimums, refunds and what happens on cancellation, and most vendors answer none of it in text. Pair it with a pricing page that states real numbers.

The second is the disambiguation FAQ. If your name collides with another company, a common noun or a product in a different category, a short "what this is not" answer gives an engine the sentence it needs to stop confusing you. Entity work and what makes a page citable by an AI engine both lean on this.

The third is the honest limits FAQ. Stating who the product is wrong for, in plain text, produces a passage nobody else has written. It is also the kind of answer a person will quote back in a thread, which matters because forums, communities and user content turn up often among the sources cited in the prompt sets we run, though we have published no count.

A fourth pattern is the one to avoid: answering in a graphic. An FAQ rendered as an image, or buried in a downloadable PDF, risks being content engines cannot use, however good the writing is. Text extraction varies by engine and by file, so the safe assumption is that the words have to exist as words. The same discipline applies to comparison pages that get cited, where the table has to exist as text.

Honest limits

We have no clean number for how much an FAQ section moves citation rates on its own, because it is rare for a live site to change one thing at a time and then wait. Three of the claims above rest on inference rather than measurement: that retrieval systems collapse near-identical FAQ blocks, that parsers other than Google's consume FAQPage, and that the four tests improve the odds of being quoted. We cannot see inside an index, so none of those is measured here. What we see in audit work is narrower: when a brand's facts come back wrong in an answer, the correct version is often missing from the site entirely, or present only in a sentence that cannot stand alone. That is a reading of the sites that hire us, which is a sample and not a census, and anyone quoting a precise lift from an FAQ rewrite is estimating.

The sample audit report shows how we write up a finding like this, with the evidence attached so you can argue with it.

FAQ

How many questions should an FAQ have? As many as you can source and no more. Four questions drawn from real support tickets beat twelve written to fill a template. If the fifth question needs inventing, stop at four.

Should the same FAQ block appear on every page? No. A block repeated across many URLs is the kind of near-identical passage a retrieval system is likely to collapse to a single copy, though that is an inference and not something we can observe from outside. Write page-specific questions, or put the shared ones on a single page and link to it.

Does FAQPage markup still do anything? It no longer produces a rich result in Google Search, and Google states that no special structured data is required for its AI features. It remains a valid schema.org type that other parsers read, so keep it where it describes visible content accurately.

How long should each answer be? Long enough to be self-contained and short enough to be lifted whole. Roughly forty to eighty words, opening with the subject named in full rather than a pronoun.

Is an FAQ a substitute for a proper page? No. A question that deserves five hundred words deserves its own URL, with the FAQ answer acting as the short version that links to it.

Sources

Related reading

What makes a page citable by an AI engineComparison pages that get citedForums, communities and user content in AI answersContent locked in images and PDFsDefinition pages, and why engines love themPrompts are not keywords: planning content for questions