Marketers have been told for years that structured data is the secret sauce of search. Add the right schema markup, the story went, and you would unlock rich results, featured snippets, and now AI citations. The third part of that promise is starting to fall apart. Recent data shows that schema for AI citations is not the lever brands assumed it would be, and the real drivers of ChatGPT and Perplexity mentions live somewhere else entirely.
This post breaks down what the latest evidence says, why schema still matters for traditional search, and what your team should actually invest in if you want to be cited by AI search engines in 2026.
Photo by Matheus Bertelli on Pexels
What the Ahrefs Study Actually Found
In early 2026, Ahrefs published a study of 1,885 pages that examined which on-page signals correlated with citations from large language models. The results surprised a lot of SEO teams. Pages with detailed Schema.org markup were not meaningfully more likely to be cited than pages without it. Article schema, Product schema, FAQPage schema, and HowTo schema all showed essentially no lift in citation frequency once other variables were controlled for.
The pages that did get cited shared a different set of traits. They had clear, scannable answers near the top. They covered specific questions in depth rather than skimming many topics. They were referenced by other authoritative sites. And the brands behind them showed up in Reddit threads, YouTube comments, and forum discussions where people swap real opinions.
That last point matters more than most teams realize. Large language models are trained on the open web, which includes a heavy dose of community content. When your brand is talked about in those spaces, it becomes part of the model's reference set in a way that no schema field can replicate.
So if schema for AI citations is not the answer, what should you focus on? The next sections walk through it.
Why Schema Still Earns Its Spot, Just Not for AI
Schema markup is not dead. It still helps Google understand entities, products, and reviews. It still feeds the Knowledge Graph. It still powers some structured result types in traditional search. Local businesses, ecommerce sites, and recipe publishers all see real benefits.
The thing is, Google announced in May 2026 that FAQ rich results are being deprecated, which removed one of the biggest visible payoffs of FAQPage schema. We covered the full impact in our piece on what to do after the FAQ schema deprecation, and the same logic applies here. Schema is plumbing. It does not generate demand, build authority, or make your content more quotable to an LLM.
Treat schema as a hygiene task. Add it where it makes sense, validate it with Google's testing tool, and move on. Do not expect it to be the thing that gets you into ChatGPT answers.
Photo by César Gaviria on Pexels
The Real Drivers of AI Citations
If schema is not the lever, what is? Based on the Ahrefs data and a growing body of agency case studies, four factors do most of the heavy lifting. Here is how they compare.
| Factor | What It Looks Like | Why It Matters for AI Citations |
|---|---|---|
| Direct answer formatting | First 60 to 100 words answer the question clearly | LLMs pull short, self-contained passages they can quote |
| Topical depth | A single page covering one question from multiple angles | Models reward pages that match the user's full intent |
| Off-site brand mentions | Reddit threads, YouTube reviews, podcast mentions | Training data and live retrieval both index conversational content |
| Original data and quotes | First-party stats, named experts, primary research | Models prefer sources that introduce new information |
Notice what is missing from that list. There is no row for keyword density, no row for schema, no row for meta description tweaks. The new game rewards content that reads like a knowledgeable human wrote it for another human.
A practical example. If you want to be cited for a query like "best way to vet influencers for a SaaS launch," you need a page that opens with a crisp answer, supports it with steps and data, and is supported by other sites and threads that mention your brand in the same context. Schema will not save you if those pieces are missing.
How to Audit Your Pages for AI Visibility
You can run a simple audit on any page that you want to win citations for. Start with these five checks.
First, read the first paragraph. Does it answer the page's main question in plain language? If a stranger had three seconds, would they get what the page is about? If not, rewrite the lede.
Second, count your sources. Does the page link to or quote at least two authoritative sources? Original studies, named experts, or primary data sources all count.
Third, search for your brand on Reddit, Quora, and YouTube. Type your brand plus the topic into each. If you do not see organic discussion, that is a gap. Building presence in those spaces is a separate project, but you cannot fix it from your own site.
Fourth, check whether the page introduces something new. Is there a stat, a quote, a framework, or a hands-on example that the reader could not find elsewhere? If the page is a remix of what is already ranking, models will likely pull from the originals instead.
Fifth, test the page in ChatGPT and Perplexity. Ask the question your page is supposed to answer. See which sources get cited. If yours is not there, look at what is. The cited pages usually share two or three of the factors above.
This audit takes about fifteen minutes per page. It will tell you more about your AI visibility than any schema validator.
One more tip. Save the results of your ChatGPT and Perplexity tests in a simple tracker. Note which queries you tested, which sources got cited, and where your page placed. Run the same checks every two weeks. AI search results shift more often than Google rankings do, and patterns emerge quickly once you have a few data points. Some teams build this into a weekly content review so writers and SEO leads can see what is changing in real time.
Photo by Leeloo The First on Pexels
Where to Invest Instead
If you are reallocating budget away from heavy schema projects, here is where it tends to pay off in 2026. None of these are quick wins, but they compound.
Voice-of-customer research. Spend time reading the actual language your customers use in Reddit, in support tickets, and in sales calls. Write pages that use those exact phrases. LLMs are trained on the same language and will surface pages that mirror it.
Original data. A small survey of 200 users, a benchmark of 50 tools, or a teardown of a real campaign produces content that other people quote. That secondary citation is what feeds AI models over time.
Brand mentions in community channels. Sponsor a thoughtful Reddit AMA. Get a customer to do a YouTube review. Show up on podcasts in your niche. The Ahrefs study and several follow-up reports point to off-site mentions as one of the strongest predictors of LLM citations.
Tools that help you operationalize this work. At Bizkol, we have written extensively about how AI is changing influencer and creator workflows, including our overview of MCP for marketing research. The same shift applies to SEO. Workflow tools that surface community conversations, track brand mentions, and connect creator partnerships to content production are now table stakes, not nice-to-haves.
If you want to see how a modern AI-driven workflow can fit into your stack, take a look at our getting started guide for Bizkol MCP.
What to Stop Doing
A short list of things teams should stop doing in 2026.
Stop adding FAQPage schema in the hope of winning rich results. Google is sunsetting those. Keep the FAQ section on the page if it helps readers, but do not invest engineering hours in the schema itself.
Stop assuming that more schema equals more visibility. It does not. Use the schema types that fit your content, validate them, and ship.
Stop writing thin pages that target many keywords lightly. Models reward depth. One strong page on a specific question will beat ten shallow pages every time.
Stop treating AI search as a separate strategy from your community presence. The two are tightly linked. Off-site mentions feed AI citations. If you are not investing in where your audience actually talks, your AI visibility will plateau.
The Bottom Line
Schema for AI citations is one of the most overrated levers in SEO right now. The data shows that structured data does not meaningfully move the needle on ChatGPT or Perplexity citations. What works is clear answers, original data, deep topical coverage, and a brand presence in the communities where real conversations happen.
Spend less time on markup and more time on content that humans actually quote. The models will follow.
Start your free trial at Bizkol
Photos provided by Pexels