Short answer: no. Ahrefs tracked 1,885 pages that added JSON-LD schema markup between August 2025 and March 2026, matched them against 4,000 control pages already cited by AI, and found the schema change produced no meaningful citation gain on any platform: Google AI Overviews citations fell 4.6% relative to controls, AI Mode moved 2.4%, and ChatGPT moved 2.2%, neither change distinguishable from zero. If a vendor is selling a treatment center on schema markup as the fix for being invisible on ChatGPT, that claim does not survive the largest test run on it to date.
This matters for addiction treatment and behavioral health marketers specifically because schema has been sold as the easy GEO win for two years: add FAQ schema, add Organization schema, get cited. A October 2026 preprint by Chen and Koudas, analyzing 158,099 AI answers and 468,037 citations across ChatGPT, Gemini and Perplexity, found the same result from a different angle. Comparing pages on the same website, an FAQ block went with 3.4% fewer citations, not more, and none of FAQ blocks, structured data, summary boxes, author boxes or page speed had a significant effect within a website. What did matter: 100% of the pages ChatGPT cited were pages its own search had already retrieved, and pages whose title and a paragraph matched the buyer’s actual question were cited more often, regardless of markup.
“Every client who has come to us after a GEO vendor sold them a schema rollout had the same result: nothing moved, because the schema was never the gap,” said Ryan Rivard, founder of BSPKN. “The gap is almost always retrieval and being named somewhere besides your own site.”
What the data actually shows
Three separate outcomes get treated as one thing in most GEO advice, and the 2026 evidence says to split them apart:
| Outcome | What it means | What moved it (2026 data) |
|---|---|---|
| Retrieved | The engine’s own search returns your page at all | Crawlability and indexing; ChatGPT’s retrieval rarely matches what ranks in Google |
| Cited | The answer links your page | Title and paragraph wording matched to the buyer’s question; schema and FAQ blocks showed no significant lift |
| Named | The answer mentions your program by name | Credible third-party pages (directories, reviews, press) that already name the brand, especially for lesser-known programs |
A treatment center in the Twin Cities or anywhere else chasing “AI SEO” by bolting schema onto existing pages is optimizing the wrong layer. The 2026 tests point to three things that actually moved the needle: being indexed and eligible to appear in Google Search at all (Google states no special schema is required for AI Overviews eligibility), writing the page’s title and one paragraph in the literal words a family searches, and getting named on outside pages, such as directories and reviews, that an engine can retrieve independently of your own site.
What still has a real, narrower case
The evidence is not “schema is worthless.” Kurt Fischman’s single-author preprint, analyzing 730 citations across 75 commercial queries, found one exception worth testing: Product or Review schema with populated pricing or rating fields was cited significantly more than generic Organization or Article markup (p = 0.012). That is a narrow, unreplicated finding, not a blanket case for every schema type. Keep accurate, valid schema for the rich-result and knowledge-graph role it still plays. Do not fund a schema rollout as the centerpiece of an AI-citation plan and call the job done.
What to do instead this quarter
- Fix retrieval first. If a program page is not indexed and eligible to appear in Google Search with a snippet, no engine will cite it, schema or not.
- Write to the question, not the keyword. Put the exact phrasing a family would type (what are the levels of care at a 30-day program, does insurance cover detox) into the page title and the opening paragraph.
- Get named somewhere besides your own site. Directory listings, licensing records, and accurate third-party coverage are the strongest documented path to a brand being named by an AI answer when the model does not already know it.
- Measure retrieval, citation and naming separately. Run the actual questions your admissions team hears, in the AI products your buyers use, and track whether your page shows up, gets cited, and whether your program gets named, over time, not as a one-time snapshot.
This is the same gap a free Peek AI visibility scan is built to show: run the real questions a prospective patient or family asks, see whether your program is retrieved, cited or named today, and get the report to compare against a competitor down the street. No pricing, no obligation, just the data most behavioral health marketing teams are currently guessing at.
Frequently Asked Questions
Does adding schema markup help a treatment center get cited by ChatGPT?
No controlled or large-sample test published in 2026 shows a clear citation gain from adding schema to pages already being cited. Ahrefs’ 1,885-page test found no meaningful uplift on Google AI Overviews, AI Mode or ChatGPT.
Do I need schema markup to appear in AI Overviews at all?
No. Google states no special schema.org markup is required to be eligible as a supporting link in AI Overviews or AI Mode. The requirement is that the page is indexed and eligible to appear in regular Search with a snippet.
Does FAQ schema help get a page cited?
No test found evidence that it does. A 2026 preprint comparing pages within the same website found FAQ blocks went with fewer citations, not more, once the website’s overall reputation was controlled for.
What actually gets a behavioral health program named by an AI assistant?
Being retrievable by the engine’s own search, having page copy worded in the buyer’s actual question language, and being named accurately on outside pages the engine can independently retrieve, such as licensing directories and reviews.
Should treatment centers stop using schema markup entirely?
No. Valid, accurate schema still supports rich results and knowledge-graph eligibility. The finding is narrower: do not buy a promised AI-citation increase on the strength of adding it alone.
Why this is BSPKN’s lane, not a vendor’s
We run GEO programs for treatment centers, dock builders and financial advisors, and the same pattern shows up in every vertical: an agency promises a citation lift from a technical checklist item and cannot show a single test where the item moved the number. The three verified studies above (Ahrefs, Chen and Koudas, and Fischman) are the first large-sample, methodologically documented look at what actually correlates with being cited in 2026, and they say the checklist items are not the lever. Retrieval, question-matched copy and outside, third-party naming are. That is also the exact framework behind our healthcare marketing program and the companion piece on which directories actually get a treatment center cited by AI, which is the “named” layer of this same framework in practice.
For a Twin Cities outpatient program or a national residential program competing for the same handful of AI answers, the fix is not a developer ticket to add JSON-LD across the site. It is confirming the program’s core pages are indexed and readable, rewriting the opening paragraph of each service page to answer the real question a family types, and auditing which outside directories and review sites already name the program accurately. Book a 15 minute strategy call after running the scan if the gap is bigger than expected.
Our own Propel OS reporting runs this exact retrieved/cited/named breakdown for every client, and the same questions are the basis of the client results we track before recommending any technical change.
