A team adds FAQPage schema to a page, waits a few weeks, and sees no change in ChatGPT or Perplexity citations. The instinct is to add more schema. The real problem is usually simpler: FAQ schema and the FAQ text a visitor reads are two separate things, and AI engines do not weigh them the same way. AskEO checks both sides of this gap during an audit, because a site can have a technically valid FAQPage block and still be invisible to an AI answer.
FAQ schema is a JSON-LD block that marks a set of questions and answers so a machine can parse them without loading the rendered page. Visible FAQ text is the same questions and answers written into the page's HTML, the way a person reading the page would see them. A page can have one without the other, both, or a version of each that no longer matches.
What Is FAQ Schema, and What Is Visible FAQ Text?
FAQ schema lives inside a <script type="application/ld+json"> tag, usually in the page's head or just before the closing body tag. It carries a mainEntity array of question and answer pairs in a format search engines can parse directly, with no HTML rendering required.
Visible FAQ text is the opposite: plain HTML elements, often a heading and a paragraph, or a list, that a browser renders and a person can read and expand on the page itself. It has no special markup unless the page also carries schema pointing at it.
The two are meant to describe the same content twice, once for machines and once for people. Problems start when a team writes one and forgets the other, or edits one without updating its twin.
Do AI Crawlers Actually Read the Schema Block?
Yes, but not as their primary source. GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and Applebot-Extended each fetch a page's full HTML, and JSON-LD is part of that HTML, so it is technically available to all of them. What differs is how much weight each answer engine gives the schema versus the rendered text a reader sees.
Google's own structured data guidelines require that a FAQPage schema block match the visible page content exactly. If the JSON-LD answer says one thing and the on-page text says another, Google can drop the rich result entirely, and in some cases treat the mismatch as a violation of its structured data policies. That single requirement tells you which version Google treats as the source of truth: the visible page.
Which One Gets Cited in AI Answers?
Generative engines like Perplexity and ChatGPT build their answers from the text they can read on the rendered page, not from a hidden data block a person never sees. A generative engine quoting your FAQ is quoting your paragraph, your wording, your phrasing choices. Schema does not get quoted; it gets parsed as a confirmation that the page's structure matches what the crawler already extracted from the visible content.
This means the schema's real job shifts depending on the engine. For classic Google search, FAQ schema is what earns the expandable rich result under your blue link. For AI Overviews, chat answers, and voice assistants, the visible paragraph is what gets lifted and cited. Skip the visible version and write schema only, and there is nothing for a generative engine to quote even if the crawler can technically parse the JSON-LD.
JSON-LD Only vs Visible Text Only vs Both
| Approach | Google rich result | AI Overview / chat citation | Maintenance risk |
|---|---|---|---|
| JSON-LD only, no visible FAQ | Possible, if valid | Rare — nothing to quote | Low effort, easy to go stale unnoticed |
| Visible FAQ only, no schema | Not eligible | Full coverage | Low, since there is only one copy to maintain |
| Both, kept in sync | Eligible | Full coverage | Higher — every edit needs two updates |
The middle row is the one most teams underestimate. A visible-only FAQ section, written well, already covers the AEO side of the job. Schema is the layer you add on top once the visible version is doing the actual work.
How to Structure FAQ Content So Both Sides Read It
- Write each question as a real long-tail query, phrased the way a person types it, not as a keyword phrase.
- Write the answer's first sentence as the complete answer, then add one or two sentences of context after it.
- Keep every answer between 40 and 80 words so it stands alone with no surrounding page context.
- Publish the question and answer as visible HTML on the page first.
- Generate the FAQPage JSON-LD from that same HTML, copying the text verbatim rather than rewriting it.
- Re-check both copies after any content edit, since a schema block that silently drifts from the visible text is the most common cause of a dropped rich result.
Doing this by hand across a growing blog is where teams fall behind — checking whether last month's ten posts still have matching schema and visible text is tedious work nobody schedules. This is exactly the kind of drift AskEO's AEO check flags in a single audit: it reads the rendered page, reads the schema, and tells you where the two disagree.
You can see this on a full site crawl rather than checking pages one at a time.
What About the Objection That This Is Duplicate Content?
Writing the same question and answer twice, once as schema and once as visible text, can feel like it risks a duplicate content or keyword stuffing flag. It does not, for one specific reason: JSON-LD is not rendered to a visitor and is not treated by Google as separate visible content competing with itself. It is metadata describing content that already exists once, on the page. The risk that matters is not duplication, it is drift — the schema and the visible text quietly saying two different things after one gets edited and the other does not.
What to Do With This Today
Pick one page that already ranks or gets traffic and has an FAQ section. Open its page source, find the FAQPage JSON-LD if one exists, and compare each answer word-for-word against the visible text below it. If they match, you are set. If they do not, or there is no schema at all, that is a one-page fix, not a site rebuild. Create a free AskEO account and run that page through a scan — it checks the schema-to-text match automatically and lists exactly which question pairs are out of sync, in about 30 seconds.
FAQ
Does FAQ schema improve my Google ranking directly?
No. Google's structured data guidelines describe schema as eligibility for a rich result format, not a ranking signal on its own. It can improve click-through rate by making your result larger in the search results, which indirectly helps traffic, but it does not raise your position.
Can I use FAQ schema without any visible FAQ text on the page?
You can technically publish the JSON-LD alone, but Google's guidelines require the schema to reflect content that is actually visible on the page, and generative engines have nothing to quote without visible text. A schema-only FAQ section rarely earns either a rich result or an AI citation in practice.
Do GPTBot and PerplexityBot read JSON-LD schema at all?
Both crawlers fetch the full rendered HTML of a page, which includes any JSON-LD script tags, so the schema is technically accessible to them. In practice, the visible paragraph text is what gets extracted and quoted in a generated answer, since schema is structured metadata rather than prose an engine would repeat verbatim.
How often should I re-check that my schema matches my visible text?
Re-check any time you edit an existing FAQ answer, since that is when drift starts. For a site with dozens of FAQ-bearing pages, a monthly audit catches mismatches introduced by small copy edits that a person would not think to re-check against the schema.
What happens if Google finds a mismatch between my schema and page text?
At minimum, Google can drop the FAQ rich result from that page's search listing. Google's structured data guidelines also list mismatched or hidden structured data as a spam policy violation in more serious cases, so a small wording drift is a low-stakes fix worth making before it compounds across many pages.