Structured data that ticks every schema box can still fail AI search, because large language models look for reassurance of entities and relationships rather than correctly labeled page types. That is the core argument from Search Engine Journal columnist Helen Pollitt, answering the latest reader question in its Ask An SEO series: "What are the most common structured data mistakes that hurt AI visibility?"

Pollitt identifies five recurring errors. Each one reflects the same gap: SEOs have optimized schema for search bots, but LLMs demand more context, consistency and factual alignment.

1. Treating markup as a checklist, not an entity strategy

Many marketers mark up the main schema types for a page as a matter of rote. That approach made sense when the goal was simply helping search engines understand content clearly. It falls short for AI search.

An ecommerce blog article with article schema is technically correct implementation — the bots understand the page is informational, not commercial. But it does nothing to connect the article to other entities on the site.

"Does our structured data make it easier for machines to understand the context and relationship of information on our page?" is the question organizations should ask, according to Pollitt. A better implementation links the article to its author and to the organization the author works for, adding author and organization schema alongside the article markup. The goal, she writes, is to reinforce brand identity, content authorship, product entities and the relationships between them, giving context across the entire website.

2. Relying solely on structured data to clarify entities

Clarifying what type of information sits on a page is not enough. The marked-up information must align with information the LLMs might ingest elsewhere — and it is not a given that your marked-up content will be the information the LLMs trust as the most authoritative.

Part of the strategy, Pollitt argues, is correcting misinformation online. Websites using your old brand name or carrying outdated pricing can undermine the schema on your own site. Marking up the correct brand name and product prices might not convince the LLMs that your data is the one to trust.

3. Not using consistent entity identifiers

Entity identifiers let sites reuse the same schema markup across pages without repeating code. Defining an entity once with an @id means subsequent pages can refer back to that identifier instead of creating new schema each time.

Pollitt illustrates this with a fictional retailer, helensecommercestore.example. The organization entity is defined once:

{
"@context": "https://schema.org",
"@type": "Organization",
"@id": "https://www.helensecommercestore.example/#organization",
"name": "Helen's Ecommerce Store",
"url": "https://www.helensecommercestore.example/"
}

The @id is not a real URL. It acts as a unique identifier, so any later reference — for example, a publisher property inside article schema — carries through all the same information without redefining the entity.

This is not just a coding shortcut. It provides consistency and lessens the risk of conflicting information across the site. Pollitt gives a concrete failure case: product-page schema added in 2022 carries the old company name "Helen's Shop," while homepage schema updated in 2026 says "Helen's Ecommerce Store." The agents face a question: which is the correct name? With an @id, the homepage update would have been inherited across all pages using that identifier — no conflicts, no multiple updates.

4. Using valid schema that misrepresents the visible content

Markup that describes content absent from the page risks a manual penalty from Google and confuses LLMs, because the signals in schema need reinforcement from on-page content.

Pollitt's example is a product page for "Helen's Premium Coffee Machine" at £79.99, in stock, with no customer reviews shown on the page. The schema, however, declares an aggregateRating with a ratingValue of 4.9 and a reviewCount of 127.

"This is ripe for a Google manual penalty, and a very confusing experience for AI bots," she writes.

5. Allowing schema to go stale or conflict with other sources

The same coffee machine page demonstrates the final mistake. The visible price is £79.99, but the schema's Offer lists the price at 59.99 GBP — £20 lower than what customers see.

AI systems are trying to establish fact, Pollitt explains. When two "statements" of fact directly contradict each other, the reliability of the webpage as a source of information about the product drops. At best, the mismatch confuses the bots. At worst, it damages customer satisfaction and could leave the site open to a manual penalty.

The so-what for marketers

The pattern across all five mistakes is the same: schema written for crawlers no longer suffices when LLMs weigh entity relationships, external corroboration and factual consistency before citing a source. Brands that audit their structured data for entity linkage, identifier consistency and on-page alignment — and that correct stale third-party information about their names and prices — stand a better chance of becoming the trusted source AI systems cite.