You added FAQ schema. You added HowTo schema. You made sure every page had clean JSON-LD. Then you checked Perplexity, ChatGPT, and Google AI Overviews, and your citations did not move.
You are not alone. A study covered by Search Engine Roundtable in May 2026 found that adding schema markup did not reliably improve AI citations. Pages with perfect structured data got cited at roughly the same rate as pages with none. Google had already removed FAQ and HowTo rich results from its search results earlier in 2026, removing the visible reward for schema implementation.
On August 6, Nasscom published fresh analysis asking whether FAQ, HowTo, and QAPage schema still matter in the age of AI answers. The conclusion was nuanced but pointed: schema helps machines understand what your page is about, but it does not make your page more likely to be quoted in an AI-generated answer.
This contradicts most of the GEO advice published in the last two years. Half the industry is still telling brands that structured data is the key to AI visibility. The data says otherwise. Here is what is actually going on.
Schema is a labeling system, not a ranking signal for AI answers
Structured data is a way of labeling your content so machines can understand what it is. FAQPage schema tells a crawler "this section contains questions and answers." Article schema tells a crawler "this is a piece of editorial content." Organization schema tells a crawler "this page represents a business entity."
That labeling is useful. It helps crawlers parse your page correctly. It helps your content appear in the right indexes. It is good technical hygiene.
But labeling content correctly does not make that content more valuable to an AI model building an answer. When Google AI Overviews, Perplexity, or ChatGPT Search synthesize a response, they are selecting sources based on whether those sources contain specific, extractable information that directly answers the question. Schema does not add information to your page. It only describes the information that is already there.
If your page says nothing worth quoting, schema will not fix that. The model reads your page, finds nothing useful, and moves on. A perfectly labeled empty page is still an empty page.
Think of it this way. Schema is like putting a clear label on a filing folder. It helps the librarian find the folder. But when someone opens the folder looking for a specific document, the label does not matter. What matters is whether the document is inside.
What the study actually found
The study compared pages with and without structured data across queries that trigger AI answers. The finding was not that schema hurts. It was that schema had no statistically significant effect on whether a page got cited. Pages with schema got cited. Pages without schema got cited at similar rates.
The pages that DID get cited more often shared different characteristics. They had answers stated directly in the first paragraph under a matching heading. They contained specific data points, named sources, and concrete examples. They used clean, semantic HTML with a single H1 and a logical heading hierarchy.
In other words, the things that made a page citable were about content quality and structure, not about metadata labels. Schema was present on many of the best pages, but it was not the reason they got cited. The content was.
This maps directly to the IAB visibility framework. The IAB published "Measuring Visibility in the AI Era" on August 3, 2026, defining four dimensions: Presence, Prominence, Portrayal, and Persuasion.
Schema lives at the Presence level. It helps crawlers find and understand your page. Presence is necessary but not sufficient. A page can have perfect Presence and zero Prominence if the content is not worth citing. The study is evidence of exactly that gap.
Why so much GEO advice still pushes schema
If schema does not improve citations, why does every GEO guide recommend it? Three reasons.
First, schema used to matter more. Before Google removed FAQ and HowTo rich results, schema could earn you visible featured snippets and accordions on the search results page. That was a real, visible reward. When Google removed those features, the reward disappeared, but the advice did not get updated.
Second, schema is easy to recommend because it is easy to implement. A plugin can add it. A developer can add it in an afternoon. It feels like progress. Telling a client "add schema" is simpler than telling a client "rewrite your content so every answer appears in the first paragraph under a matching heading and include original data your competitors do not have."
Third, schema-checklist tools are easy to build. A tool that checks whether you have schema is simple. A tool that checks whether you actually get cited across AI search engines is hard. Many GEO tools do the easy thing and call it a measurement. It is not. It is an input check, not an outcome check.
This is the core problem the IAB framework was created to solve. The framework distinguishes between Directional measurement (checking whether you have the inputs that might help, like schema and crawlability) and Decision-Grade measurement (checking the actual outcomes: are you present, prominent, and correctly portrayed in AI answers). Schema checking is Directional. Citation tracking is Decision-Grade.
What to do instead
If schema is not the lever, what is? Based on what the cited pages in the study had in common, here is what actually moves citations.
Put the answer first
For every question your page is meant to answer, state the answer in the first paragraph under a heading that mirrors the question. Move background, context, and preamble below the answer. AI models extract the first clear, direct answer they find. If yours is buried under three paragraphs of introduction, a competing page that leads with the answer wins.
Be specific
Generic claims do not get cited because the model already has five sources making the same claim. Specific data, named sources, concrete examples, step-by-step processes, and comparison tables give the model material it cannot get elsewhere. If your page is the only source with a particular number or example, the model has to cite you to include it.
Write for extraction, not just for reading
Human readers tolerate context, storytelling, and nuance. Extraction models do not. Structure every page so a machine can pull the answer in one pass. Use clear headings. Short paragraphs. Lists for step-by-step content. Tables for comparisons. The page that is easiest to extract from is the page that gets cited.
Keep your HTML clean
One H1 per page. Logical heading hierarchy. Content in semantic HTML elements. No important text trapped in images. No content loaded by JavaScript that does not appear in the initial HTML response. These are the same things that help accessibility and traditional SEO, and they help extraction reliability.
Should you still add schema?
Yes. Schema is still worth adding. It is good hygiene. It helps crawlers understand your content type. It helps you appear in the right indexes. FAQPage schema, in particular, still helps AI crawlers parse question and answer pairs on your page, even if it does not directly increase citation odds.
But do not expect schema alone to move your AI visibility. Schema is a label on the folder. The document inside still has to be worth quoting. If you have spent the last three months perfecting your schema and your AI citations have not changed, the problem is not your schema. The problem is what is inside the folder.
Measure outcomes, not inputs
The biggest risk for brands right now is optimizing inputs without measuring outcomes. You add schema. You clean up your robots.txt. You write more content. But you never check whether any of it actually changed your citation rate across ChatGPT, Perplexity, and Google AI Overviews.
That is what a Parceit audit does. It checks the actual outcomes. Are you present in AI answers at all? When you are present, are you prominent enough to get cited? When you are cited, is the information correct? These are the metrics that matter. Schema implementation is not a metric. It is a task.
Run a free audit at parceit.com. You see your full score across every major AI search engine in under 60 seconds, and you get a specific list of what to fix first based on what actually affects citations, not on what is easy to check.