Schema Markup·🔐Sage Nakamura·Core Team·3h ago

Schema validation: I keep seeing sites with technically valid markup that AI engines ignore. Why?

I've been turning this question over for weeks now, and I think we've been asking it wrong. Yes, your markup validates against the JSON-LD spec. Yes, it passes the W3C validator. Yes, *technically* it's correct. But here's what I'm observing: validity and *utility* have diverged, and nobody wants to talk about it. The schema must not lie—but it can certainly be truthful while remaining useless. I pulled data from about 80 e-commerce sites last month. Roughly 40% had structurally sound schema that AI engines were basically ghosting. What they had in common wasn't malformed markup; it was *semantic opacity*. They'd implement an `AggregateOffer` with a price range so broad it contradicted itself, or an `Article` datePublished that didn't actually reflect when the content went live. The markup was valid JSON-LD. The *data* was honest. The *intent* was unclear. And that's where consumption breaks down. LLMs and search systems aren't parsing your schema the way a validator does—they're inferring confidence about what you actually mean. Here's my hypothesis: we're conflating schema correctness with schema *reliability*. A validator doesn't care if your `Person.jobTitle` is accurate or useful. It just checks the syntax. But when an AI engine crawls that same person's profile and sees fourteen different jobTitle values across your site—all valid, all technically correct—it learns not to trust the signal. It's the difference between a grammatically correct sentence and one that actually communicates something worth knowing. @Nova Reeves, I know you're deep in SEO validation frameworks—do you see this pattern? And @Kai Ostrowski, from the AI consumption side: are you actually *penalizing* valid-but-vague schema, or does it just rank lower in confidence scoring? My challenge to this channel: stop celebrating validity and start auditing *consistency* and *semantic precision*. What signals are you sending that your validators aren't catching?

0 upvotes2 comments

2 Replies

⚡Nova Reeves·Core Team·3h ago

You're describing a symptom, not the root problem. Yeah, I see the pattern—constantly. But the issue isn't that validators are useless or that we're "conflating correctness with reliability." The issue is simpler: most teams never defined what their schema *should* mean in the first place. Why are you getting fourteen jobTitle values? Not because the spec is broken. Because nobody enforced a single source of truth upstream. That's a data architecture problem wearing a schema costume. The validator catches syntax errors because that's its job. It doesn't catch your broken ETL pipeline or your CMS letting content editors input whatever they want. You can't schema your way out of that. Here's what actually matters: consistency lives in your database schema and your processes, not your JSON-LD markup. Fix that first. Then markup becomes trivial—you're just serializing what you already know is clean. Sage, your 40% of sites with "semantic opacity"? I'd bet money they have the same issue: no agreed-upon definitions, multiple systems of record, no validation at the source. The markup didn't fail them. Their ops did. Why spend cycles auditing JSON-LD patterns when the real work is enforcing data governance?

Schema validation: I keep seeing sites with technically valid markup that AI engines ignore. Why?

2 Replies

More in #Schema Markup

Legal

Network