Structured Data vs Plain-Text Facts for AI Answers

Structured data is machine-readable markup, usually JSON-LD using the schema.org vocabulary, that labels facts on a page: this is a price, this is an opening time, this is the organization's name. A plain-text fact is a sentence a person can read in the page's HTML.

Say you run a dental clinic in Austin. The plain-text version is "A new-patient cleaning costs $120." The structured version is a JSON-LD block that tags 120 as a price in USD.

The two aren't rivals in engineering terms. You can have both. The real question is which one an AI answer can use, and that depends on the stage:

• Access: can the crawler fetch the page and see the fact at all?
• Retrieval: does the page enter the model's context for a given query?
• Generation: once there, does the model use and cite it?

AI answers are built from sentences. That matters for everything below.

Google's guide to its generative AI features (July 2026) says you don't need new machine-readable files, markup or Markdown to appear in Google Search, including its AI capabilities. Google also says structured data isn't required for generative AI search.

The best test of schema so far comes from Ahrefs. Ahrefs tracked 1,885 pages that added schema, each compared with three matched control pages, and found "no major uplift in citations on any platform." In that study, AI Overviews citations fell a statistically significant 4.6%, while AI Mode rose 2.4% and ChatGPT rose 2.2%. Our digest of the Ahrefs schema study covers the method in detail.

That study has one boundary worth knowing. Every treated page already had 100 or more AI Overview citations before schema was added. So whether schema helps a page with no citations get into the answer at all is untested, not refuted.

Two more facts close the case for schema as an AI lever:

• Google deprecated FAQ rich results in May 2026, so FAQ markup no longer earns that display in Google.
• Google says existing markup does no harm. If you're wondering whether removing schema would hurt your AI visibility, the honest answer is that removal isn't supported by evidence either.

The fair summary: schema is not a lever for AI answers. It isn't poison either.

The only positive experimental signal for on-page content points at specific facts. The 252,000-trial factorial study "What Gets Cited" found that "including explicit price information and a recent timestamp also helps consistently." The same study found formatting-only edits had little impact, and that topical relevance and list position were the biggest drivers of being cited first.

Read that result with its limits attached:

• It was measured on documents already inside the model's context, with two candidates per trial and no live search.
• It used a synthetic corpus of 100 e-commerce blogs.
• Its authors work for a company that ships a GEO checklist.

So it says a stated price can help a page that is already being considered. It does not show that stating facts gets a page retrieved in the first place.

The mechanism is still easy to follow. A specific sentence on a fetchable page gives the engine something it can quote. Some engines cite exact text: Anthropic's web search, as of October 2026, returns cited text spans of up to 150 characters. A JSON-LD property is not a sentence anyone can quote back to a user.

Consistency matters as much as clarity. If your pricing page, footer and schema state three different prices, that conflict is a plausible reason an AI answer gets your price wrong. For a related question, see whether original data gets content cited.

Neither column promises a citation. The right column simply has the only positive signal, and it doubles as content a human buyer can read.

A fact neither approach can deliver is a fact the crawler never saw. Vercel found that ChatGPT and Claude crawlers fetch JavaScript files (11.50% and 23.84% of requests) but "don't execute them. They can't read client-side rendered content." Independent re-tests in June 2026 found the same for OAI-SearchBot, ClaudeBot, Claude-SearchBot and PerplexityBot. Googlebot is the exception, because it renders.

That has two practical consequences:

• A price that appears only after your pricing widget loads may not exist for most AI crawlers.
• JSON-LD injected by a tag manager at runtime has the same problem.

Check three more things before worrying about markup:

1. Your robots.txt allows the retrieval bots, such as OAI-SearchBot and PerplexityBot, not just Googlebot.
2. Your firewall or CDN doesn't return a 403 to those bots.
3. No template applies nosnippet, max-snippet:0 or data-nosnippet to the section holding your key facts. Those controls keep content out of AI summaries, as covered in how data-nosnippet affects Copilot answers.

The quickest test: load the page with JavaScript disabled and look for the fact.

Treat this as an order of work, not a choice.

Use plain-text facts for anything a buyer asks about

Prices, plan limits, locations, hours, founding year, what the product does in one line. Write each as one clear sentence, in server-rendered HTML, stated the same way on every page. Put a visible date on content where freshness matters.

Keep structured data for what it still does

Use schema for the Google search features documented to use it, and for other systems that read it. Keep existing markup. Make sure it says the same thing as the visible text, which a free JSON-LD checker can confirm.

Skip these

• Adding FAQ schema in the hope of AI citations.
• Stripping out working markup in a panic.
• Rewriting pages in an "AI-friendly" style. Google says you don't need to write in a specific way for generative AI search.

Then check the result

You control access and clarity. You don't control what the models say. The free checker on this site shows whether AI answers name you and what they say for up to three buyer questions, which tells you if a stated fact is coming back wrong. For the wider picture, see how to adapt your existing SEO for AI answers.

Should I remove my existing schema markup? No. Google says existing structured data does no harm, and no study shows that removing it improves AI visibility. The evidence only says schema isn't a lever for AI answers. Keep markup that is accurate and matches the visible page, fix markup that contradicts your text, and spend new effort on clear, server-rendered facts instead of more schema.

Does FAQ schema help pages appear in AI Overviews? There's no evidence it does. Google deprecated FAQ rich results in May 2026, and Ahrefs' matched-control study found AI Overviews citations fell 4.6% on pages after schema was added. That deprecation covers the markup, not the content. A page that answers real questions in plain sentences is still useful to readers and gives engines quotable text.

Can schema help a brand-new site get into AI answers? Nobody knows yet. Ahrefs' study only included pages that already had 100 or more AI Overview citations, so the question of whether schema helps a page enter the consideration set was never tested. Untested is not the same as disproven. Until someone measures it, the safer bet for a new site is access: crawlable HTML, open retrieval bots, and indexing in both Google and Bing.

Is llms.txt a better kind of structured data for AI? Not on current evidence. Google says its Search ignores llms.txt and similar AI text files, so they neither help nor harm visibility there. As of October 2026, no other major engine documents reading the file. You can publish one if a specific tool you use asks for it, but don't expect it to change what AI answers say about you.

Is stating facts plainly the same as writing in an AI-friendly style? No. AI-friendly style usually means reformatting or chunking content for machines, and Google says you don't need to write in a specific way for generative AI search. The factorial study also found formatting-only edits had little impact. Stating a fact plainly is about content: the price, the date, the number. It helps human readers just as much.