The conventional AEO advice sounds reasonable: add schema markup everywhere, structure your JSON-LD correctly, tag your FAQs and how-tos, and the AI answer engines will reward you. A growing number of practitioners are now pushing back on that assumption. Hard.

The argument is not subtle. AI answer engines, including Perplexity, ChatGPT with web browsing, and Google's AI Overviews, are not parsing your @type: FAQPage declaration when they decide what to cite. They are reading your content the same way a well-read researcher would, and they are forming judgments about authority and answer quality that no schema tag can fake.

The Core Claim: Schema Does Not Drive AI Citations

The argument holds that language models generating answers do not rely on structured data the way a traditional search crawler does. When a model retrieves and reads a page, it processes the visible text. It evaluates whether that text actually answers the question clearly, whether the source appears trustworthy, and whether the content is topically coherent. Schema markup is largely invisible to that process.

This tracks with what AEO teams monitoring citation patterns have observed. Pages without any structured data get cited regularly. Pages with meticulous schema implementation get ignored if the underlying content is thin or vague. The signal is in the substance, not the wrapper.

Tools like Promptwatch and Peec that track which pages actually earn citations confirm this pattern across industries. High-citation pages tend to share content characteristics: specific answers, named authors or organizations with verifiable expertise, and consistent topical focus. Their schema implementations vary widely.

What AI Engines Actually Appear to Evaluate

Four signals come up repeatedly when practitioners audit their citation-earning pages:

Topical clarity. The page commits to answering one question or covering one narrow topic with depth. It does not bury the answer under three introductory paragraphs.

Authority signals in the content itself. Named experts, cited studies with dates, and institutional affiliations mentioned in the body text all appear to improve citation rates. These are signals a language model can read directly; a schema tag claiming author: John Smith carries far less weight than a paragraph that quotes John Smith with his title and affiliation.

Direct answer quality. The first 200 words matter disproportionately. If a model can extract a clean, factual answer from that window, it is more likely to surface the page. RankScale and Vaylis AI both offer testing environments where you can measure how well a specific passage performs as a citation source before publishing.

Trust and consistency. Pages on domains with a long track record of accurate, stable content outperform newer pages even when the newer content is technically superior. This is harder to manufacture quickly.

Where to Reallocate Budget

If schema markup is consuming significant dev or agency hours, the reallocation question becomes concrete. Most practitioners running AEO audits with tools like Athena HQ or Profound will find faster citation gains by redirecting that effort toward:

• Rewriting thin pages to lead with direct answers in the first paragraph
• Adding author bylines with verifiable credentials to content that currently runs anonymous
• Building topical clusters around specific question sets rather than broad keyword categories
• Commissioning original data or research that gives AI engines something citable that does not exist elsewhere

The last point is particularly effective. Scrunch and OmniSEO both track when AI engines cite original statistics or studies. Pages with proprietary data get cited at rates that far exceed what any schema optimization could produce.

Where Schema Still Has Indirect Value

This is not an argument for ripping out all your structured data. Two cases where schema retains real value:

Google AI Overviews. Google's system has a closer relationship with structured data than third-party AI tools because it controls both the crawl and the answer layer. FAQ schema and HowTo schema still appear to influence which passages Google surfaces in AI Overview panels. If Google organic is a meaningful channel for you, maintaining schema there is defensible.

Rich snippets feeding training data. There is a longer-horizon argument that well-structured pages, including those with schema, are more likely to end up in training datasets for future model versions. This is speculative and hard to measure, but it is not implausible.

Outside those two cases, schema is not where the citation gains are.

A Quick Audit Checklist for This Week

Run through your top 20 pages by organic traffic and ask:

• Does the page answer its primary question in the first 150 words?
• Is there a named, credentialed author visible in the body text?
• Does the page contain any data, statistic, or example that cannot be found on five other domains?
• Is the topic narrow enough that a model could describe the page's subject in one sentence?
• Has this page been stable and accurate for at least six months?

Pages that fail three or more of these checks are better candidates for rewriting than for schema work. The content gap is the citation gap.

Platforms like Gauge and xFunnel can help you track citation changes as you make these content adjustments, so you are measuring actual AI visibility shifts rather than proxy metrics. Start there before scheduling another schema audit.