Schema Markup for AI Optimization Didn’t Improve AI Rankings or LLM Citations
For the past couple of years, schema markup has been positioned as one of the more actionable levers SEOs can pull to improve their visibility in AI-generated answers. The logic seemed reasonable: if you give AI systems cleaner, more structured data, they should be able to understand and cite your content more confidently. It made intuitive sense – until a large-scale controlled experiment put that assumption directly to the test.
Ahrefs ran that test. They analyzed over 6 million URLs, ran a matched difference-in-differences analysis over 30 days, and examined citation behavior across AI Overviews and AI agents. What they found doesn’t completely invalidate schema as a practice, but it does dismantle a very specific and widely held belief about why schema supposedly helps AI visibility.
We’ve been tracking developments in AI search optimization closely at Marketing 1on1, and this study is one of the more rigorous attempts to separate signal from noise in a space that desperately needs it. Let’s break down what the study found, what it missed, and what it actually means for your SEO and AI optimization strategy going forward.
What the Ahrefs Study Actually Tested
Ahrefs tested whether adding JSON-LD schema markup to pages that were already receiving AI Overview citations would increase those citations over a 30-day observation window. Using a matched difference-in-differences methodology, they compared treated pages (with schema added) against control pages (without schema added) that shared similar citation baselines. No statistically significant positive effect was found across four separate tests.
This wasn’t a correlational study or an anecdotal observation. Ahrefs used their Brand Radar tool and an AI agent to execute a controlled comparison. Pages with schema markup were matched to three equivalent control pages without JSON-LD, and citation changes were tracked over 30 days before and after the addition of schema. That’s a meaningful attempt at causal inference in a domain where most published “insights” are just pattern-matching.
The initial observation that sparked the experiment was interesting in its own right: AI-cited pages were three times more likely to include JSON-LD than non-cited pages. That kind of correlation would lead almost any SEO to conclude that schema is driving AI citations. But Ahrefs didn’t stop at the correlation – they tested the causal direction, and that’s where things got complicated.
The Four Tests and What They Showed
- AI Overview citations: A 4.6% decline was observed in the treated group, slightly steeper than the control group. The difference amounted to roughly 12 fewer daily citations. Whether the schema caused this or whether it was a coincidental trend noise is genuinely unclear.
- Agent A analysis: No clear positive or negative effect detected.
- Brand Radar comparisons: Results were similarly inconclusive.
- Direct-fetch tests: AI systems were shown to not actually use schema markup when fetching pages in real time.
That last point is particularly worth sitting with. The direct-fetch test result tells us something important about how current AI retrieval architectures actually work – and it’s not what most schema advocates have been assuming.
The Correlation vs. Causation Problem That’s Been Confusing Everyone
The 3x correlation between JSON-LD presence and AI citations is real. But Ahrefs offered the most coherent explanation for why it exists: sites that implement schema markup tend to be better-maintained, technically healthier, and higher-quality overall. The schema isn’t driving the AI citations – the underlying site quality is driving both the schema adoption and the AI citations independently.
This is a classic confounding variable problem. High-quality editorial sites, well-structured e-commerce platforms, and established authority domains are more likely to have technical SEO infrastructure in place – including schema – because they have the resources and expertise to implement it. Those same characteristics make them more likely to get cited by AI systems regardless of their structured data implementation.
“The correlation between schema and AI citations is real. The causation isn’t. What you’re actually seeing is that quality sites tend to do both – implement schema and earn AI citations – for completely independent reasons.”
This distinction is not semantic. If you believe schema causes AI citations, you’ll invest time and budget into structured data as an AI optimization lever. If you understand the confounding variable, you’ll recognize that schema serves specific, legitimate purposes – but AI citation frequency isn’t one of them for pages that are already in AI systems’ awareness.
What Schema Markup Actually Does (and Doesn’t Do) for AI Systems
What Schema Genuinely Supports
- Rich results in traditional Google Search – star ratings, event listings, product prices. These are still schema’s strongest use case.
- Knowledge Graph population – structured entity data helps Google’s Knowledge Graph understand relationships between entities, which has downstream effects on brand recognition in AI systems trained on Google’s data.
- Crawl signal enhancement – for pages not yet indexed or recognized, schema may help signal content type and relevance to crawlers.
- Semantic clarity for less-established content – on pages that lack strong inbound signals, structured data can supplement contextual clarity.
What Schema Does Not Do
- It does not cause AI systems to cite your page more frequently if that page is already recognized.
- It is not used by AI systems during real-time page fetching, based on Ahrefs’ direct-fetch testing.
- It does not substitute for content quality, topical authority, or E-E-A-T signals.
- It does not improve AI Overview placement or citation frequency for pages already in the citation pool.
“Schema markup is a communication protocol between your site and search infrastructure – not a citation trigger for AI systems. Treating it as an AI ranking signal is a category error.”
The Important Limitations the Study Acknowledged
Credit to Ahrefs for being transparent about what their study doesn’t resolve. Understanding these limitations is critical for applying the findings correctly rather than overcorrecting.
Pages Already in the Citation Pool
The entire dataset consisted of pages that already had 100 or more AI Overview citations before schema was added. This is a meaningful constraint. The study answers the question: “Does adding schema help pages that AI already knows about?” It does not answer: “Does schema help uncited pages get discovered or cited for the first time?”
Those are fundamentally different questions. It’s entirely plausible – and Ahrefs acknowledged this directly – that schema could play a role in helping AI crawlers discover and index pages that haven’t yet entered AI systems’ awareness. That remains an open research question.
The 30-Day Window
Structural changes to how AI systems process and weight pages may not manifest in citation behavior within 30 days. Longer-term effects, if they exist, wouldn’t be captured here.
Schema Types Were Pooled
All schema types were grouped together in the analysis. It’s possible that specific schema implementations – say, FAQ schema, HowTo schema, or Speakable schema – behave differently than generic Article or Organization markup. Pooling them obscures type-specific effects.
Concurrent Page Changes
Isolating schema as the sole variable is difficult in practice. Other page-level changes occurring during the test period could contaminate results in either direction.
How AI Systems Actually Retrieve and Cite Content
To understand why schema markup for AI search optimization may be less impactful than assumed, it helps to understand what AI retrieval systems are actually doing when they surface citations.
Large language models powering AI Overviews, Perplexity, ChatGPT, Gemini, and similar systems don’t parse JSON-LD in the same way a search crawler does. Their citation behavior is driven by a combination of:
- Pre-training data quality and frequency – content that appears across many authoritative sources and contexts gets stronger model representation.
- Retrieval-augmented generation (RAG) signals – when AI systems do real-time retrieval, they’re typically indexing and ranking based on semantic embeddings, domain authority, and content relevance, not structured data annotations.
- Traditional index signals – Google’s AI Overviews heavily leverage the existing organic search index, meaning your traditional SEO performance is a major input into your AI citation probability.
- Content clarity and extractability – AI systems favor content where the answer to a query is clearly stated, logically structured, and not buried in preamble.
Notice that JSON-LD schema is not on that list as a direct driver. The Ahrefs direct-fetch finding aligns with this: AI retrieval agents weren’t looking at or using schema annotations when they accessed pages in real time.
What This Means for Your Schema Markup Strategy
None of this means you should stop implementing schema. It means you should implement it for the right reasons and have accurate expectations about what it will and won’t accomplish.
Keep Implementing Schema For:
- Rich result eligibility (Product, Review, Event, Recipe, etc.)
- Local SEO signals (LocalBusiness schema remains valuable)
- Knowledge Graph entity disambiguation (Organization, Person schema)
- Breadcrumb and SiteLinks enhancement
- Video and Article schema for discovery in specialized verticals
Stop Expecting Schema to Do:
- Increase the frequency with which AI systems cite your established content
- Substitute for topical authority and content depth
- Serve as an AI ranking signal in the way a traditional meta signal might work
- Compensate for thin content or weak E-E-A-T signals
What Actually Moves AI Citations
Based on our direct experience working across SEO and AI optimization, the factors that consistently correlate with higher AI citation frequency are:
- Topical authority – being a recognized source across a specific domain, not just on individual pages
- Content that directly answers questions – AI systems extract direct answers; content structured around clear question-answer formats gets cited more
- Strong traditional organic performance – pages ranking well in organic search are more likely to be included in AI Overview citation pools
- Entity strength – brands and authors with robust Knowledge Graph presence and cross-platform mentions get cited more reliably
- Content uniqueness and information gain – AI systems don’t benefit from citing content that simply repeats what’s already everywhere
Myths vs. Facts: Schema Markup and AI Optimization
| Myth | Fact |
|---|---|
| Adding JSON-LD schema will increase your AI Overview citations. | Ahrefs’ controlled study found no statistically significant positive effect on AI citation frequency for already-cited pages. |
| The 3x correlation between schema and AI citations proves schema drives them. | The correlation is explained by site quality as a confounding variable – quality sites implement schema AND earn citations independently. |
| AI systems read and use JSON-LD when fetching pages in real time. | Direct-fetch testing showed AI retrieval agents did not utilize schema markup during live page fetches. |
| Schema markup is a core AI ranking signal. | AI citation behavior is driven by content quality, topical authority, and traditional index signals – not structured data annotations. |
| Schema has no value for AI search at all. | Schema may still help uncited pages get discovered and indexed – this remains an open research question not addressed by the Ahrefs study. |
The Broader Lesson About AI Search Optimization Advice
The schema-for-AI narrative is a useful case study in how optimization myths propagate. Here’s the pattern we see repeatedly in this industry:
- Someone observes a correlation (pages with X tend to rank better or get cited more).
- The correlation gets published as a recommendation (“implement X to improve AI visibility”).
- The recommendation spreads through the SEO community without the underlying causal question being tested.
- A controlled study eventually reveals the causal link doesn’t exist, or is much weaker than assumed.
- The community slowly updates – but the original advice continues circulating for years.
We saw this with exact-match anchor text, with meta keywords, with exact-match domains, and now we’re seeing it with schema as an AI citation lever. The responsible approach is to distinguish between what’s correlated and what’s causally effective before building strategy around it.
“In AI search optimization, the gap between ‘this correlates with better performance’ and ‘this causes better performance’ is where a lot of wasted budget lives. The Ahrefs schema study is valuable precisely because it tried to close that gap experimentally.”
What Needs Further Research
The Ahrefs study opens as many questions as it closes. Here’s where we believe the most important research gaps remain:
Schema’s Effect on Previously Uncited Pages
This is the most critical open question. If schema helps AI crawlers discover, index, or understand pages that haven’t yet entered the citation pool, that’s a meaningful finding – it would mean schema is an entry point signal rather than a frequency signal. Testing this would require a dataset of pages with zero or near-zero AI citations, applying schema, and observing whether they enter the citation pool at higher rates than equivalent control pages.
Schema Type-Specific Effects
Pooling all schema types masks potential differences between implementations. Speakable schema, for instance, was explicitly designed to support voice and AI assistants. FAQ and HowTo schemas create structured answer formats that may interact differently with AI extraction logic. These deserve isolated testing.
Longer Observation Windows
AI system behavior changes over time as models are updated and index signals shift. A 30-day window may not capture delayed effects of structural page changes. Six-month or annual longitudinal studies would provide more confidence in null findings.
Interaction Effects with Content Quality
Schema on high-quality, original, information-rich content may behave differently than schema on average-quality content. Understanding whether schema functions as an amplifier for strong content – even if it doesn’t independently move the needle – would be strategically valuable.
Our Perspective: Where to Focus Your AI Optimization Energy
We’ve worked with enough sites across enough verticals to have a grounded perspective on what actually moves AI citation performance. Schema is not the lever most people think it is for this specific goal. Here’s where the real leverage is:
1. Build Genuine Topical Authority
AI systems favor sources that are recognized authorities on a topic – not just individual pages that rank for a query. Comprehensive coverage of a subject domain, consistent publishing standards, and strong inbound link signals from topically related sources all contribute to the topical authority that AI systems use to identify citation-worthy sources.
2. Write for Direct Answer Extraction
Content that contains clearly stated, factual, concise answers near the beginning of relevant sections is structurally optimized for AI extraction. This isn’t about keyword placement – it’s about making the answer so obvious and well-articulated that any extraction system can identify and use it.
3. Strengthen Your Entity Presence
AI systems that operate on entity-based understanding – which includes Google’s systems – cite sources with strong entity signals more reliably. This means consistent NAP data, Knowledge Panel presence, Wikipedia mentions, authoritative author profiles, and cross-platform brand mentions all contribute to citation probability in ways that JSON-LD alone cannot replicate.
4. Prioritize Traditional SEO Performance
For Google AI Overviews specifically, there is strong evidence that organic ranking performance is a major input into citation selection. A page ranking in positions 1-5 for a query is far more likely to appear in the AI Overview than a page ranking on page two. This means core SEO – technical health, content quality, link authority – remains the foundation of AI search visibility.
5. Apply Schema for Its Actual Benefits
Implement structured data for rich results, local SEO, Knowledge Graph signals, and content-type clarity. These are genuine, measurable benefits. Just don’t expect schema to directly lift your AI citation counts.
Frequently Asked Questions: Schema Markup for AI
Does schema markup help with AI Overviews and AI-generated search results?
Based on Ahrefs’ controlled study, adding schema markup did not significantly increase AI Overview citations for pages that were already being cited. The study found no clear positive or negative effect across four separate tests. However, the study was limited to already-cited pages – whether schema helps uncited pages get discovered by AI systems remains an open question that requires separate research.
Why do AI-cited pages have more schema markup if schema doesn’t cause citations?
The correlation exists because of a confounding variable: site quality. High-quality, well-maintained websites are more likely to implement schema markup as part of their technical SEO infrastructure, and those same quality signals – domain authority, content depth, technical health – independently drive AI citation probability. Schema and AI citations are both outcomes of site quality, not causally linked to each other.
Should I stop implementing schema markup for my website?
No. Schema markup remains valuable for traditional rich results (star ratings, FAQs, events, products), local SEO (LocalBusiness schema), entity disambiguation in Google’s Knowledge Graph, and structural content clarity. The finding that schema doesn’t directly drive AI citation frequency doesn’t diminish these use cases. Implement schema for the right reasons – just don’t expect it to function as an AI ranking signal.
What actually influences AI citation frequency and AI search visibility?
The strongest drivers of AI citation frequency include: topical authority (being a recognized comprehensive source in a domain), traditional organic search ranking performance, entity strength and Knowledge Graph presence, content that contains clear and directly extractable answers, and information originality that gives AI systems a reason to cite your source specifically rather than a generic competitor. These signals collectively matter far more than structured data annotations.
Could schema markup still help pages that aren’t yet appearing in AI results?
Potentially yes – and this is the most important unanswered question from the Ahrefs study. The research only tested pages with 100 or more existing AI Overview citations. It’s plausible that schema could help AI crawlers discover, understand, or initially index pages that haven’t yet entered the citation pool. This represents a meaningful research gap, and the findings from the current study should not be extrapolated to uncited pages without dedicated testing.
The Bottom Line
Schema markup for AI optimization was, for a period, one of the most frequently repeated recommendations in the AI search visibility space. The Ahrefs controlled experiment doesn’t erase schema’s value – but it does dismantle the specific claim that adding JSON-LD will meaningfully increase your AI citation frequency for pages already in AI systems’ awareness.
The 3x correlation between schema and AI citations is real. The causal link, when tested rigorously, isn’t there. That’s a meaningful distinction for anyone building an AI search optimization strategy.
What does matter: content depth, topical authority, traditional search performance, entity strength, and the structural clarity of your answers. These aren’t new principles – they’re the fundamentals of good SEO applied to a new context. Schema remains part of a complete technical SEO foundation, but it belongs in that category: foundational hygiene, not an AI citation accelerator.
We’ll continue watching for research that addresses the remaining open questions – particularly around schema’s potential role for uncited pages. But for now, the evidence points clearly toward investing AI optimization budget in content quality and topical authority over structured data implementation.
Work With a Team That Understands What Actually Drives AI Visibility
At Marketing 1on1, we don’t chase tactics that look good in theory but fail under scrutiny. We build SEO and AI search strategies grounded in what the evidence actually supports – including knowing when a widely-recommended approach doesn’t hold up to controlled testing.
If you want a clear-eyed assessment of your AI search visibility, your schema implementation, and where your optimization budget will actually generate measurable returns, we’re worth talking to. The difference between correlation-based advice and causally-grounded strategy is the difference between activity and results.
Get in touch with Marketing 1on1 to build an AI search strategy based on what the data actually shows.