Do FAQ blocks and schema earn AI citations?

KEY TAKEAWAYS

Across sites, FAQ blocks and schema look like a citation lift. Within one site the lift disappears.
At equal ChatGPT retrieval rank, FAQ blocks, schema and author boxes went with slightly lower odds of being cited.
A title and paragraph that answer the buyer's question in its own terms is the on-page change with evidence.
Most of the gap between sites sits off the page, in retrieval, model memory and third-party coverage.

On 10 October 2026, Mahe Chen and Nick Koudas posted a paper that tests the most common on-page GEO advice against 158,099 AI answers and 468,037 citations. They tracked ChatGPT, Gemini and Perplexity daily for 90 days in two markets: US home internet and a small B2B software niche.

The finding that matters for anyone with a content budget is about method. FAQ blocks, schema markup, summary boxes, author boxes and page speed all looked like citation wins when pages were compared across websites. When pages were compared with other pages on the same website, every one of those gains disappeared. This piece covers why, what still works, and how to test it on your own site before you fund another round of FAQ sections.

Why do FAQ blocks and schema look like they work?

Because most of the evidence compares different websites. Three examples you will see quoted:

Ahrefs scanned about 6 million URLs and found AI-cited pages nearly three times as likely to carry JSON-LD as pages that were not cited.
The GEO-16 paper from Kumar and Palkhouski audited 1,100 URLs cited 1,702 times by Brave, Google AI Overviews and Perplexity. Structured data correlated with citation at r = 0.63.
AirOps reported that ChatGPT cited pages with JSON-LD 38.5% of the time, against 32.0% without, across 16,851 queries.

All three are cross-sectional. GEO-16's authors say brand reputation and backlinks may drive both their quality scores and citations. Ahrefs called its own scan a correlation.

Chen and Koudas reproduced the same pattern, then removed it. Their model compared each page only with other pages on its own website, which strips out everything that makes one site stronger than another.

Practice	Effect across websites	Effect within the same website
FAQ block	+18.1%	-3.4% (not significant)
Summary box	+20.9%	+6.9% (not significant)
Author byline	+14.9%	-4.6% (not significant)
Schema.org markup	+7.5%	-5.3% (not significant)
Page speed, one SD faster	+11.7%	-1.1% (not significant, 466 pages)

The reason is concentration. FAQ blocks appeared on 43% of pages from the most-cited quarter of websites and on 23% from the least-cited quarter. The authors call it an instance of Simpson's paradox: the comparison across sites credits the FAQ block with the strength of the site.

What does the trap look like in numbers?

A hypothetical example, with made-up figures, shows how easily it happens.

Site A is a well-known publisher. It has 100 pages, 60 of them with FAQ blocks, and every page averages 20 AI citations a month whether or not it has one. Site B is a smaller brand. It has 100 pages, 10 with FAQ blocks, and every page averages 4 citations.

Pool the two sites and compare:

Pages with FAQ blocks: (60 × 20 + 10 × 4) / 70 = 17.7 citations a month.
Pages without: (40 × 20 + 90 × 4) / 130 = 8.9 citations a month.

The FAQ pages appear to earn twice as many citations. Inside each site, the FAQ block changed nothing. AirOps says its schema result holds regardless of domain authority, but it does not describe how that was controlled. Adjusting for an authority score is not the same as comparing pages on the same site, because the score captures only part of what makes a site get cited.

What happens when you add schema to a page?

Ahrefs tested that directly. It followed 1,885 pages that added JSON-LD between August 2025 and March 2026 and matched each with three control pages from other domains that never did. Over 30 days either side of the change, AI Mode rose 2.4% and ChatGPT 2.2% relative to controls, both indistinguishable from zero. AI Overviews fell 4.6%, which was statistically significant, though both groups were already declining and Ahrefs drew no causal conclusion.

The limits are real. Every page already had 100 or more AI Overview citations, so the study says nothing about pages with no visibility. The window was 30 days, and all schema types were pooled.

Chen and Koudas add one more detail. Holding ChatGPT's own retrieval rank equal, pages with FAQ blocks had 0.93 times the odds of being cited, schema 0.92 times and author boxes 0.77 times. These are associations, but they point the wrong way for the standard advice.

What does survive inside a site?

Question match. Two page properties passed all five of the paper's robustness tests within websites:

Best paragraph match. The paragraph closest in meaning to the questions buyers ask. About 15.5% more citations per standard deviation.
Title match. About 14.1% more per standard deviation.

Match here means similarity of meaning, measured with sentence embeddings against the 651 questions the team tracked. It is not about question marks. That matters, because Google's AI optimisation guide pushes back on rewriting headers as questions and chopping content into chunks. The two positions fit: say what the page answers in the buyer's own terms, without reformatting for its own sake. AirOps found something similar for ChatGPT, with closely matching headings cited 41% of the time against 29%.

The effect varies by engine. Paragraph match held for all three. Title match was strong for Gemini (+16.8%), small for Perplexity (+4.7%) and absent for ChatGPT (-0.2%). It was also stronger on third-party pages than on brand pages, where no feature passed all five tests in the main market.

Controlled tests agree. Vishwakarma and colleagues ran 252,000 paired trials across six models and found topical relevance and list position mattered most, while formatting-only edits had little impact. The original GEO paper from Aggarwal and colleagues found adding quotations, statistics or citations lifted visibility by up to about 40% on its benchmark. Both changed what the text said, not how it was wrapped, and both started from sources the engine had already retrieved.

Where does the gap between sites come from?

Mostly off the page. In the consumer market, 13% of the variation in citations sat between websites and 87% between pages on the same site. Within sites, ChatGPT's own retrieval explained about 32% of the variation. All page features together explained 2.8%. Every page ChatGPT cited had been retrieved by its own search, and 97% of those retrieved pages were absent from Google's top 50.

Model memory plays a part too. When a brand was already in ChatGPT's prior for a question, ChatGPT wrote it into its searches 67% of the time, against 11% otherwise. In the B2B niche, a cited third-party page naming a vendor raised the odds that ChatGPT named that vendor 39.7 times.

A separate Fudan University paper, posted on 8 October 2026, points the same way. Eight of ten AI search tools cited a made-up concept within a week of it being posted on sites they already cite often. Forty posts on a self-hosted WordPress site drew no citations in seven days. Where the text lives counted for more than how it was formatted. That was a test of invented topics, not brands, and Google has said since 15 May 2026 that its spam policies cover attempts to manipulate AI responses. The lesson is to earn coverage in cited sources, not to plant it.

In our terms, this is digital authority. The between-site gap is press coverage, brand entity and AI visibility. Page formatting is a small share of it.

Which on-page practices should you keep?

Practice	What the evidence shows	Decision
Title written to the buyer question	Passes all five tests within sites	Do first, on every page you want cited
A paragraph that answers the question directly	Passes all five tests, all three engines	Do first
FAQ block	No gain within sites. The format adds nothing; real buyer questions answered well can	Keep only where it answers questions the page does not already cover
FAQPage or Article schema	No gain within sites or after adding. FAQ rich results left Google Search on 7 May 2026, though Google still reads the markup	Keep if accurate and matched to visible text. Do not fund it for citations
Organization schema with sameAs	Not tested by these studies. It serves entity consistency, not page citation	Keep
Summary or TL;DR box	+6.9% within sites, not significant	Optional, for readers
Author byline	Lower odds at equal rank, not significant	Keep for readers. Do not expect citations
Page speed	No within-site effect, small sample	Keep for users, not for AI
llms.txt	Google says Search ignores it	Skip for Google

One open question is Copilot. Microsoft's Fabrice Canel confirmed in March 2025 that schema helps Microsoft's models understand content. Neither the Ahrefs study nor the Chen and Koudas paper measured Copilot, so understanding has not been shown to turn into citations there.

How do you test this on your own site?

The fix for the trap is to run the within-site comparison yourself.

Pick 20 to 40 similar pages on one site. Same template, same section. Mixing page types reintroduces the problem.
Write the buyer question each page should answer. Use the wording buyers use, not your product names.
Get a per-page citation baseline. Bing Webmaster Tools' AI Performance report lists citations per URL for Copilot and Bing's AI summaries. For other engines, use repeated prompt sampling, as in how to measure AI visibility when answers change.
Split pages into matched pairs by baseline and assign one of each pair to change. The untouched half is your control.
Change one thing. Rewrite the title and the key paragraph to answer the question. Leave format and markup alone.
Measure at least four weeks and compare the change in each group. Answers drift. In the October study, only about 40% of the variation in ChatGPT citations between pages on a site carried over after two weeks. Bing's data is also sampled and can be refined later, so ignore small moves.
Test format separately. Only after the content test, add FAQ blocks or markup to a fresh treatment group, again against a control.

If a page is never retrieved, no rewrite will get it cited. Check retrieval first, as in our three gates breakdown for Claude.

The short version

FAQ blocks and schema sit on pages that get cited because they sit on sites that get cited. Compared like for like, they do not move citations. Spend first on titles and paragraphs that answer the buyer's question, then on the coverage that makes engines retrieve you at all. Ethos Engine is building per-page and per-engine tracking into its digital authority reports. Join the waitlist to hear when it opens.

Questions

Does FAQ schema help pages get cited by ChatGPT, Gemini or Perplexity?

The best current evidence says no. Chen and Koudas compared pages on the same website across ChatGPT, Gemini and Perplexity and found FAQ blocks, which they counted by FAQ headings or FAQPage markup, went with 3.4% fewer citations, a result not distinguishable from zero. Ahrefs found no clear gain when 1,885 already-cited pages added JSON-LD.

Why do so many studies say schema pages are cited more?

Because they compare pages across different websites. Strong, often-cited sites use FAQ blocks and schema more, so the practice gets credit for the site. In the October 2026 study, FAQ blocks sat on 43% of pages from the most-cited quarter of websites and 23% from the least-cited quarter.

Should I remove FAQPage or Organization schema?

No. Keep markup accurate and matched to visible text. Google still uses FAQ structured data to understand pages, and Organization schema with sameAs links does a different job, keeping your identity records consistent. Just do not expect page-level markup to win citations.

What on-page change does have evidence behind it?

Matching the page to the question. A title and a paragraph that are close in meaning to the questions buyers ask were the only page properties that passed all five robustness tests within websites, at about 14% and 16% more citations per standard deviation.

Sources

arXiv, Chen and Koudas, What drives AI citations? Content, retrieval and model knowledge in generative engine optimization arxiv.org
Ahrefs, We tracked 1,885 pages adding schema. AI citations barely moved ahrefs.com
Google Search Central, Optimizing for generative AI search developers.google.com
arXiv, Kumar and Palkhouski, AI answer engine citation behavior. Bringing the GEO-16 framework in B2B SaaS arxiv.org
AirOps, The fan-out effect report airops.com
arXiv, Vishwakarma, Kumar and Jamidar, What gets cited. Competitive GEO in AI answer engines arxiv.org
arXiv, Aggarwal et al., GEO. Generative engine optimization arxiv.org
alphaXiv, Liu et al., From public posts to AI-search citations. Measuring the fragility of AI search alphaxiv.org
Search Engine Journal, AI search tools cited a made-up concept within a week, paper finds searchenginejournal.com
TechWyse, Google ends FAQ rich results for all sites techwyse.com
Search Engine Land, Microsoft Bing and Copilot use schema for its LLMs searchengineland.com
Search Engine Roundtable, Google says spam policies apply to AI responses seroundtable.com
Practical Ecommerce, Bing's improved AI Performance section practicalecommerce.com
Momentic, Initial thoughts on AI Performance reports in Bing Webmaster Tools momenticmarketing.com

Esh

Founder, Ethos Engine. Builds the Digital Authority Score and runs the placement marketplace. esh@ethosengine.io

NEXT STEP

See where your brand stands.

Your Digital Authority Score, who AI recommends for your target questions, and the placements that close each gap. Your first report is free.

Get your free report →

Do FAQ blocks and schema earn AI citations? Not once you compare pages on the same site