I’ve been obsessing over schema markup for LLMs lately, and one question kept bugging me: do large language models actually parse structured data, or are we all just cargo-culting SEO tactics that may or may not work?
Understanding how LLMs parse schema markup is critical for AI optimization in 2025. So I decided to just ask Claude directly, not for some vague marketing answer, but to literally show me its workflow when it searches for local businesses.
The answer was way more interesting than I expected.
Making an LLM Show Its Work
I asked Claude a simple question: “If I asked you to find me an electrical company in El Cajon that does electrical panel upgrades, walk me through exactly what you’d do.”
Turns out it’s pretty straightforward:
- Run a web search (“electrical panel upgrade el cajon”)
- Get back search results with titles, snippets, and URLs
- Look at those snippets and decide which pages are worth investigating
- Fetch the full HTML content of the most promising ones
That’s it. No magic schema parser. No special semantic layer. Just search → snippets → fetch → read.
Which made me immediately ask the obvious follow-up question…
How LLMs Parse Schema Markup: The Question Everyone’s Asking
Me: “Do you actually see schema markup?”
Claude’s answer:
“Microsoft Copilot officially confirmed they use schema markup to help their LLMs understand content, but we don’t have clear statements from ChatGPT, Gemini, or Perplexity about whether they use schema.”
So we have one confirmed yes from Microsoft, but the rest are unclear.
What we do have is experimental data. A controlled experiment showed that schema markup enabled ChatGPT to extract and present structured data with a 30% improvement in response quality. Better organization, cleaner separation of information.
This is why structured data AI optimization matters, it’s not just about whether LLMs can “see” your schema, it’s about whether they can efficiently extract the information from it.
But here’s the kicker:
“When I fetch a page with web_fetch, I get the HTML content. The schema markup is IN that HTML (as JSON-LD in a script tag). But I don’t parse it separately – I read it as part of the overall content.”
So the question isn’t really “can LLMs see schema?”
It’s more like “does schema actually help them extract better information?”
And the answer seems to be yes, even if they’re not parsing it in some special way.
What Actually Matters When LLMs Choose What to Cite
This is where it got really interesting. Claude broke down what it actually looks for when deciding which businesses to recommend.
Traditional SEO Still Wins
“I rely heavily on what the search engine returns. If a page ranks well in the search results I get, I’m more likely to consider it. 77% of AI optimization comes from strong traditional SEO – sites ranking in Google’s top 10 are significantly more likely to be cited by AI models.”
Translation: If you’re not showing up in search results, LLMs won’t even see you. The foundation is still traditional SEO.
Title & Snippet Relevance
Does the title mention the location and service? Does the snippet have specific, useful information? Is there clear contact info?
Simple stuff, but apparently most sites don’t get this right.
Structure Is Everything
But when Claude actually fetches the full page, the difference between sites that get cited and sites that don’t comes down to structure.
Claude showed me an example. Compare these two approaches:
What works:
Electrical Panel Upgrades in El Cajon
We provide professional panel upgrades in El Cajon, CA.
• Cost: $1,500-$3,000
• Licensed: CA #XXXXX
• Response time: 24 hours
• Service area: El Cajon, Santee, La Mesa
What doesn’t:
Welcome to our electrical services page! We do lots of things
and we're really great at them. We've been in business since
1995 and we love what we do. Our team is amazing...
(continues for another three paragraphs of vague marketing copy)
Claude’s response was blunt:
“I can extract the [structured format] WAY easier, regardless of whether there’s schema markup or not.”
What Gets You Cited vs. What Gets You Ignored
According to Claude, here’s what actually makes it cite your business:
Specific facts it can quote:
- License numbers
- Addresses
- Phone numbers
- Price ranges
- Service areas
- Response times
Credibility signals:
- Years in business
- Certifications
- Reviews with dates
- Real project examples
What makes it skip your site entirely:
- Generic marketing fluff
- No specific information
- Unclear service offerings
- Vague or missing location info
Basically, if your page is just marketing copy without concrete facts, there’s nothing for an LLM to actually extract and cite.
LLM SEO: The New Ranking Game
Here’s something most people don’t realize: LLM SEO is fundamentally different from traditional SEO, but they work together.
Traditional SEO gets you into the search results that LLMs pull from. But structured data AI optimization determines whether you actually get cited once the LLM finds your page.
The winning formula looks like this:
Traditional SEO → Gets you in the results + Schema markup LLMs can parse → Helps with extraction + Clear content structure → Makes you citable = AI optimization schema strategy
Think of it this way: Google gets you on the field. Your content structure determines if you score.
Claude put it simply: sites that rank in Google’s top 10 are significantly more likely to get cited, but only if their content is structured well enough for AI to extract specific facts.
So Should You Bother With Schema?
I asked Claude directly whether schema markup is worth the effort.
The answer: Yes, but probably not for the reasons you think.
“Should you add schema? YES, but not because LLMs definitely parse it. Add it because:
- Microsoft officially confirmed they use it
- It helps Google (which helps you show up in MY search results)
- It forces you to structure your content clearly (which helps me even if I don’t parse the schema directly)
- Experiments show it helps ChatGPT extract better information
- Low effort, potential upside – Why not do it?”
But then Claude emphasized something important:
“But the REAL priority is: Clear, structured content that’s easy to scan and extract facts from.“
The Real Takeaway on Schema Markup for LLMs
After going down this rabbit hole, I’m pretty convinced we’re overthinking the whole AI optimization game.
Yes, add schema. Understanding schema markup LLMs can parse is valuable—it has proven benefits for both traditional SEO and content extraction.
But the bigger opportunity is way simpler.
Most local business websites are walls of marketing fluff. If you just:
- Write clearly
- Structure your content properly
- Include specific, quotable facts
- Use headings and bullet points strategically
- Keep information current
…you’re already ahead of 90% of your competition in the AI search game.
The secret to AI optimization schema isn’t some technical SEO hack or complex markup wizardry.
It’s just writing like a human who wants to help another human find specific information quickly.
Turns out LLM SEO and user experience optimization might just be the same thing.