Pages with FAQ schema whose answers were not present in the visible body were cited by LLMs at roughly half the rate of FAQ-schema pages with on-page matching answers, in our 2026 probe set.
We classified 50 pages from our audit set into two cohorts: those with FAQPage JSON-LD where every answer appeared verbatim or near-verbatim in the visible body (n=30), and those with FAQPage JSON-LD where 30%+ of the answers were schema-only (n=20). The visible-match cohort was cited on 36% of probed queries; the schema-only cohort on 17%. Removing the unjustified schema brought several schema-only pages back up to citation parity within 4–6 weeks.
Methodology
Cohort classification of 50 audit pages, 40-query LLM probe basket, Q1 2026.
Limits of this finding
Schema-only FAQ pages also tend to be on lower-quality sites overall; the suppression effect is real but the magnitude attributable purely to schema mismatch is bounded.
More AEO findings
Removing pronoun-led sentence openings lifted LLM citation probability on our test set by a measurable margin.
We rewrote 14 high-traffic agency pages to replace 'It usually takes…' style openings with explicit subjects ('Local SEO results usually appear…'). Citation rate across a 30-query basket on ChatGPT and Perplexity rose by roughly 18 percentage points over six weeks.
A 4-bullet TL;DR placed immediately after H1 was extracted into LLM answers far more often than identical content placed mid-page.
We tested two layouts of the same content across 8 article pages: TL;DR at top vs. TL;DR after the first body section. The top-placed TL;DR was extracted in 71% of probes; the buried version in 23%.
Adding a year or version number inside a claim ('as of Q1 2026', 'on Core Update of March 2026') correlated with materially higher selection during LLM answer generation.
We added in-sentence date stamps to 22 evergreen claims across 9 articles. Of those 22 claims, 14 were quoted verbatim in subsequent ChatGPT and Claude probes. The unstamped control set (also 22 claims) had 6 verbatim quotes.
Run experiments like this on your site.
Senior-led measurement plans, hand-counted, no laundered numbers.
Book a strategy call