If you spend much time in search or SEO circles, you have heard the verdict by now: AI search is just SEO with a new coat of paint. Structured data is dead. There is no separate game to play for getting cited by AI. The whole "structure" argument is junk, so do not waste time on it.
The proof is usually an Ahrefs study. The study is real. The conclusion people keep stapling to it is not.
What the study actually tested
Ahrefs ran a matched difference-in-differences test. It looked at 1,885 pages that added JSON-LD schema between August 2025 and March 2026, then compared them with 4,000 similar pages that did not. After that, Ahrefs measured how AI citation rates changed across three surfaces:
- ChatGPT: +2.2%, statistically indistinguishable from zero.
- Google AI Mode: +2.4%, also indistinguishable from zero.
- Google AI Overviews: -4.6%, a small but statistically real decline.
The narrow conclusion is solid: adding schema produced no meaningful citation uplift on any platform. The problem starts when people turn that finding into a broader claim than the study can support.
The decline started before schema
The same data includes the detail that changes the read: both the schema pages and the control pages were already in a steep citation decline before any schema was added. The -4.6% result is not proof that schema sabotaged those pages. It points to a broader downward trend, such as Google pulling back AI Overviews for some content, pages going stale, or crawls slowing. Schema simply did not reverse it.
So the study did not measure "schema hurts." It measured "schema, added late to pages that were already sliding, does not rescue them." Those are different claims. Only one of them is getting repeated.
Correlation was never the lie
The same research found that pages cited by AI are nearly three times more likely to carry JSON-LD. That correlation is real. It just does not run in the direction the hype assumed.
Pages with clean schema tend to be pages that are already well built, well maintained, and authoritative. Schema rides along with a strong source. It does not manufacture one. Adding schema to a weak page does not make the page citeable, and the correlation does not vanish just because the shortcut failed.
The "it is all just Google" tell
Notice what the "structure is dead" case leans on: Google's developer docs and a study weighted toward Google's surfaces. That is the tell. Treating AI search as one thing, mostly Google's thing, is the mistake.
Google Search Central documents traditional indexing, page classification, and rich results. It was never written to explain how a generative answer engine grounds a response. The people quoting it are assuming AI search behaves like Google search because some of it runs on Google's platform.
Most of it does not. A growing share of these queries happen on ChatGPT, Perplexity, and other engines, each with its own retrieval habits, source trust hierarchy, and grounding behavior. A finding that schema gave "no uplift on Google AI Overviews" tells you about Google AI Overviews. It does not settle how every other engine decides whom to cite.
Optimize for one platform's quirks, call it "AI visibility," and you have recreated the single-platform blind spot that gets businesses misread everywhere else.
What actually decides whether a machine can use you
This is where both extremes miss. Structure matters. It just does not work as a single magic tag.
Start with static, raw HTML. Your core facts, entities, and evidence need to be present before any JavaScript runs. If a crawler cannot see the content without executing your site, it may not see the content at all.
Rendered HTML comes next. If JavaScript controls important page content, the rendered page has to be clean enough for crawlers to read.
JSON-LD sits on top of that. It classifies and disambiguates what is already there. It still helps with rich results and click-through in traditional search, and it gives engines clearer context. It is not the thing that buys you a citation.
The honest read
One study should not move you from "schema is everything" to "schema is nothing." Both positions are wrong. The useful takeaway is narrower: adding schema to a weak page will not earn AI citations, but readable structure, starting with the raw HTML, is still how a machine resolves, trusts, and cites you.
Structure did not die. The shortcut did. Take a breath before you tear anything out.
Citeable proof: In a matched test of 1,885 schema-added pages against 4,000 controls, Ahrefs measured +2.2% for ChatGPT, +2.4% for Google AI Mode, and -4.6% for AI Overviews. Both groups were already declining before schema was applied. The data debunks "schema causes citations," not "structure matters."
