Three different questions
Does schema describe the page accurately? Does it qualify for a supported search feature? Did changing it affect a measured outcome? These questions need different evidence. Passing a validator does not answer the last one.
Evaluate a study before applying it
Record the engine, collection dates, sample, intervention, outcome metric, and comparison group. A study showing that cited pages often have schema is an association; it does not prove that adding schema caused the citations.
| Observation | What it supports | What it does not prove |
|---|---|---|
| JSON parses | Syntax can be read | Facts are accurate |
| Markup matches the page | Consistent description | Guaranteed search feature |
| Citations changed after editing | A change worth investigating | Schema was the sole cause |
Google's stated position
Google does not require special schema for generative AI search. Its structured-data documentation also explains that correctly implemented markup does not guarantee a rich result. Keep these statements separate from speculation about other systems.
A useful implementation experiment
Choose a real inconsistency, such as an author identity that differs between page and schema. Correct it because it is inaccurate. Record the deployment and observe indexed output and relevant citations over a defined period. If several changes ship together, report them together rather than attributing everything to markup.
What to publish
Describe the actual change, measurement, and limits. Avoid vague citations to unnamed or Princeton-adjacent research. A sourced implementation note with modest conclusions is more useful than an unsupported percentage lift.