Schema AI engines use, and schema nobody reads

Every AI SEO proposal now has a schema section. It usually lists FAQ, HowTo, speakable and a few types I have to look up, presented as the technical layer that gets a site into AI answers. Most of it marks up features Google has already removed, and none of it is mentioned in any AI engine’s documentation. Three types do real work. They are the boring ones, and they were doing that work before anyone said “generative”.

The three that matter: Organization, Person and Article. The rest marks up features Google has already removed.

What schema does, according to Google

Google’s own introduction says structured data helps it understand a page and can enable richer results. It does not say structured data is a ranking factor, and the AI features documentation says there are no additional technical requirements to appear in AI Overviews or AI Mode. So the honest frame is: schema helps an engine understand who and what a page is about. It does not make a page rank, and it does not make a page quotable.

That is still worth a lot, because “who is speaking” is the question answer engines ask before they quote anyone.

The three that matter

Organization. Google’s Organization documentation describes it as the markup that helps Google understand an organization’s details and “disambiguate” it in results. The properties that carry that: name, alternateName, sameAs, logo, foundingDate. This is the entity. A company with a Hebrew name, an English transliteration and three different descriptions across the web is three entities until something says they are one. Organization schema, with the alternate name and the links to the profiles, is the one place on the site that says it. It is the same mechanism I described in AI citations with no link building: engines quote sources they can identify.

Person. For a consultant, a founder or an author, the Person entity is what the Organization is for a company. Name, the same job title everywhere, sameAs to the profiles that confirm it, and a link from every Article back to it. An engine that sees twenty articles by a Person it can identify treats them differently from twenty articles by a string.

Article. Google’s Article documentation recommends author with a name and a URL, datePublished, dateModified, headline and image. The author link is the part most sites skip and the part that matters: it is what attaches the page to the Person above. The dates tell an engine whether the claim is current.

Three types, one job: tell every engine that this page, this site and these profiles are the same entity, saying the same thing, on a known date.

The ones nobody reads

FAQPage. Google’s FAQ documentation now says the feature “is no longer shown in Google Search results”, with the deprecation dated May 2026. It had already been limited to government and health sites since 2023. The questions and answers on the page are still useful to a reader and to an engine assembling a response. The JSON-LD around them marks up a rich result that no longer exists.

HowTo. Removed from results in 2023. Still shipped in 2026 by plugins that never updated their defaults.

speakable. Google’s speakable documentation marks it as beta, built for the Google Assistant answering news queries on smart speakers, for U.S. users in English. It is sold as “voice and AI optimization” to B2B companies in markets it does not cover, for a device category the proposal never mentions.

The retired list. Google’s changelog carries deprecation notices for practice problems, course info, estimated salary, learning video, special announcement and vehicle listing. If a proposal lists a type, check the changelog before paying for it.

None of this is harmful. It is maintenance without return, and it is usually the whole schema section of the proposal, which means the three types that carry the entity are not in it.

Does schema help AI search at all

Directly, there is no documentation that says so. Google’s AI features need nothing beyond indexing. OpenAI’s and Perplexity’s crawler documentation does not mention structured data. What the engines need is to know who is speaking, and schema is one of the three places that tell them, next to the page text itself and the third-party pages that describe you. It is the most machine-readable of the three, which is why I keep it, and the least important of the three, which is why it is never the first line of the plan. The first line is usually which AI crawlers to allow, because schema on a page the agents cannot fetch is schema nobody reads.

What I ship

1. Organization on the home page, with alternateName for every spelling the company uses, sameAs to the profiles that are actually maintained, and founder where the founder is public.
2. Person for the people who publish, linked from Organization and from every Article, with the same title and description used on LinkedIn.
3. Article on every post, with the author pointing at that Person and real dates.
4. Nothing else unless there is a live rich result for it: a product with prices, an event with a date, a recipe. If the type is on the retired list, it comes out, including the FAQ markup a previous plugin left behind.
5. Validation after every change, with the Rich Results Test and a read of the rendered JSON, because plugins change their output silently.

Then the measurement is the same as for everything else in AI search: a fixed set of buyer questions, per engine, monthly. The method is in how to measure AI search visibility. Schema shows up there as a cause, never as a metric.

On this site the FAQPage markup came out on 31 October 2026. The questions stayed on the page. Google stopped showing FAQ rich results on 7 May 2026, so the code had been describing a feature that no longer existed for five months.

Where this sits

Entity schema is the second week of AI search optimization work, after crawler access and before any page is rewritten. If your proposal has a schema section, tell me what your schema says and I will tell you which lines are the entity and which lines are a feature from 2021.