Heading vs FAQ: Which Content Format Do AI Engines Prefer to Cite? (With a Decision Framework)

Every GEO rewrite hits the same fork in the road: should the answer live in a heading-plus-paragraph block, or in an explicit question-and-answer pair? Both formats get cited. Neither wins universally. And the wrong choice doesn't just underperform — it wastes the rewrite entirely, because you publish your best answer in the format the engine skips.

This guide settles the fork with a decision framework, the extraction logic behind each format, and the scoring method to settle it per topic in minutes.

Why this is a real fork (not a style preference)

Answer engines don't read pages — they select passages, and the passage candidate is structurally defined: a heading + what follows it. Your two formats produce genuinely different candidates:

• Heading version: H2: "The best hair oils for a dry scalp" → paragraph starting with the direct answer
• FAQ version: H2/H3: "What is the best hair oil for a dry scalp?" → answer paragraph

The heading version presents a statement-shaped passage. The FAQ version presents a question-shaped passage with the question verbatim. For a user query shaped like a question, the FAQ candidate's question-text aligns almost perfectly; for a query shaped like a fragment or a topic phrase, the statement version aligns better. Same content, different collision geometry.

The extraction logic behind each format

Heading + paragraph: the "definition block"

How it wins: the heading acts as a semantic anchor, and a direct-answer-first paragraph underneath passes the rip-out test (see the full extraction logic). It's also how encyclopedic content — the most-cited corpus on the web — is structured. Engines have seen billions of heading-then-answer blocks.

Where it excels:

• Topic-defining queries: "what is GEO", "how do solar inverters work"
• Comparison and "vs" queries: the passage can carry both sides
• Queries where the user's phrasing is a fragment ("best hair oil dry scalp") rather than a full question

Where it weakens:

• Long-tail, question-shaped queries — the heading rarely mirrors the exact phrasing
• Pages with many intents on one topic: the block covers the "main" question and leaves the variants unaddressed

FAQ pair: the "mirror block"

How it wins: the question text is the user's question — near-zero semantic distance — and explicit Q&A is the most historically machine-readable structure on the web (FAQ schema trained generations of parsers). The answer that follows is conventionally tight and self-contained.

Where it excels:

• Full-interrogative long-tails: "why is my scalp flaky after oiling?"
• Troubleshooting and objection queries: "does argan oil clog pores?"
• Non-native or dialect phrasing: the question absorbs the user's register verbatim while the answer stays clean

Where it weakens:

• Broad definitional queries — a one-question block feels narrow for a "what is X" query
• Pages that over-FAQ: twenty thin Q&As dilute every passage's depth

The decision framework: three questions, one winner

Ask these per topic — in this order:

Q1 — What is the dominant question shape in your real query data?

Pull the page's top queries from Search Console, verbatim. Count full questions vs fragments vs topic phrases. Dominant shape decides the starting format:

• Full questions dominant → start FAQ
• Fragments/topic-phrases dominant → start heading block

(This is step one of the query-language method — and the reason "which format" is a data question, not a taste question.)

Q2 — Is the intent singular or split?

One intent ("best oil for dry scalp") → one strong block of the matching format covers it. Multiple intents (comparison + troubleshooting + how-to on the same page) → each intent gets its own section in whichever format fits it — the page is a portfolio of passages, not one passage.

Q3 — Which candidate scores higher on extraction quality?

Write both. Score both 0–10 on the five properties: direct-answer-first, self-containedness, question-shape match, specificity, language match. Publish the winner. The loser usually isn't waste — it's a section for a different query cluster.

Worked example: one topic, both formats, one winner

Topic cluster (real query shapes): "best hair oil for a dry scalp", "hair oil itchy scalp", "which oil reduce flaking".

Heading version:

The best hair oils for a dry, itchy scalp Argan and jojoba oil are the best choices for a dry, itchy scalp. Both mimic the scalp's natural sebum, absorb without clogging follicles, and reduce flaking within 2–3 weeks of twice-weekly use.

FAQ version:

What is the best hair oil for a dry scalp? Argan or jojoba oil, applied 2–3 times weekly directly to the scalp, are the best options for a dry scalp — they mimic natural sebum and typically reduce flaking within two to three weeks.

Scoring on the cluster's dominant shape (full questions, 2 of 3 queries): the FAQ version's question text matches three of three query phrasings near-verbatim; the heading version matches the topic but not the interrogative form. FAQ wins 9 vs 7 — and the heading block still ships as the page's main section, because the fragment-shaped query ("best hair oil dry scalp") hits it perfectly. Both got used; each for the query shape it wins.

Where schema fits (and where it doesn't)

FAQPage schema and the visible FAQ format are related but not the same lever:

• The visible Q&A passage is what extraction reads. This is the GEO lever — the passage selector consumes text, not markup.
• FAQPage markup describes the structure to search systems. It can help eligibility for rich treatments and reinforces the page's shape; it cannot rescue a weak passage.

Practical rule: mark up the FAQ sections you genuinely wrote, keep them visible and extractable, and never treat schema as a substitute for the passage-quality work. A scored FAQ passage with clean markup is the full stack; schema on a buried answer is a label on an empty box.

The scoring rubric, made concrete

Score both candidate formats per topic on five properties (0–2 each, 10 max). The sheet:

Property	Heading version	FAQ version	What earns the 2
Direct-answer-first	answer = first sentence after H2	answer = first sentence after question	conclusion in sentence one, zero runway
Self-containedness	survives the rip-out test	question+answer survive as a pair	no pronouns pointing outside the passage
Question-shape match	heading mirrors fragment phrasing	question mirrors interrogative phrasing	near-verbatim match to real query strings
Specificity	mechanism/number/named entity present	same	"argan oil, 2–3 weeks, sebum-mimicking" not "many benefits"
Language match	register of the dominant query cluster	question absorbs the register verbatim	the audience's dialect, not the brand's

Two rules for reading the scores: never let the total hide a zero — a format scoring 8 with a 0 on question-shape match loses to a 7 with no zeros, because zeros are where extraction actually breaks. And re-score after rewrites: the winner flips when the underlying query mix shifts (seasonal, campaigns, new products).

The rubric takes three minutes per topic by hand — and it's precisely the computation the platform automates across every topic on every page, in one run.

The scale problem — and the answer to it

One topic, two formats, a five-property score: three minutes. A real audit is 20 pages × 5 clusters × 2 formats = 200 scored items — every one needing the query data attached, in the audience's language, written properly in both registers. That's a specialist-week per site, per cycle.

This is the exact computation the Scrabio GEO platform automates: dual-format content per topic generated from your real Search Console queries, bot-preference scores per item, winner named with reasoning. You review and publish.

If you want to run the decision framework by hand first, audit your top pages' AI visibility to find which pages have money riding on the fork — start where the impressions are.