How Schema Markup Boosts AI Search Citations

The Site and the Setup

B2B SaaS client. Marketing analytics space. Domain authority: 51. The site had been around for 6 years, had solid backlink profile, ranked well for most of its target keywords. Schema had never been prioritized. Classic case of "the site works fine without it."

We selected 14 pages based on one criterion: each was tracking in our AI citation monitoring system, and each was being cited by at least one AI platform for at least one query before the test. This gave us a measurable baseline.

``` Pages in test: 14

Domain authority: 51

Niche: B2B SaaS / Marketing Analytics Prior schema: None

Test duration: 21 days

Baseline period: 30 days prior (same 14 pages, no schema) AI platforms tracked: Perplexity, ChatGPT (GPT-4), Claude, Google AI Overviews Queries per page: 3-7 target queries tracked ```

---

The Schema Stack We Added

Not all schema was equal in this test. We added the following types to all 14 pages:

Tier 1 — Added to every page:

• Article (with author, datePublished, dateModified, publisher)
• BreadcrumbList
• WebPage
• SameAs (entity linking to company's Wikidata entry, LinkedIn, Crunchbase)

Tier 2 — Added where applicable:

• HowTo (8 of 14 pages had process-oriented content)
• FAQPage (11 of 14 pages had questions sections)
• SpeakableSpecification (markup identifying sections suitable for voice/AI reading)

Tier 3 — Added to 3 pages with specific content:

• Dataset (pages citing proprietary research data)
• ClaimReview (pages making verifiable industry claims)

All schema was added as JSON-LD. No microdata. Validated with Google Rich Results Test and Schema.org validator before going live.

---

Results: Day 21

Perplexity — The Biggest Mover

``` Metric | Before | After (Day 21) | Change

• -----------------------------|--------|----------------|--------

Pages cited at least once | 5/14 | 12/14 | +140% Avg citation rate per query | 12% | 61% | +408% Avg citations per page/month | 2.1 | 8.7 | +314% ```

Perplexity moved fastest and furthest. Within 9 days of schema going live, we saw citation rate increases on 7 pages. By day 21, 12 of 14 pages were being cited at a meaningfully higher rate.

The two pages that didn't move: both were listicle-format pages with no clear HowTo or FAQ structure. We added FAQPage schema but the underlying content didn't have the right structure for it to be semantically coherent. We're restructuring those pages now.

ChatGPT (GPT-4) — Slower but Consistent

``` Metric | Before | After (Day 21) | Change

• -----------------------------|--------|----------------|--------

Pages cited at least once | 7/14 | 11/14 | +57% Avg citation rate per query | 18% | 34% | +89% ```

ChatGPT responded more slowly. The citation improvements were visible by day 12–14, not day 9. The increase was real but less dramatic than Perplexity's. We believe this reflects different re-indexing cycles and different weighting of structured data signals between the platforms.

Google AI Overviews — Moderate

``` Metric | Before | After (Day 21) | Change

• -----------------------------|--------|----------------|--------

Pages cited at least once | 9/14 | 11/14 | +22% Avg citation rate per query | 29% | 38% | +31% ```

Google AI Overviews moved less than Perplexity or ChatGPT in percentage terms. Google already had the highest baseline citation rate, so the ceiling was lower. The improvement is real but not the headline finding.

Claude — Minimal Change

``` Metric | Before | After (Day 21) | Change

• -----------------------------|--------|----------------|--------

Pages cited at least once | 4/14 | 5/14 | +25% Avg citation rate per query | 9% | 12% | +33% ```

Claude's citation behavior changed the least. Based on our broader 90-day dataset, Claude cites the fewest sources per query and shows the highest citation stability — it changes slowly in both directions. 21 days isn't enough to see the full effect on Claude. We'll check at 60 days.

---

Which Schema Types Did the Work

This is the part most schema guides don't tell you, because most guides are written based on theory, not controlled experiments.

HowTo schema was the highest-leverage markup for AI citations. Of the 8 pages where we added HowTo markup, 7 showed citation improvements in Perplexity within 10 days. HowTo markup makes the content structure explicit to the model — it labels each step, the tools required, and the expected output. That's precisely the information an AI model needs to cite a source confidently.

SpeakableSpecification moved the needle on AI Overviews specifically. This schema type, which Google introduced for voice search, marks sections of content as suitable for text-to-speech. Google's AI Overview appears to use similar signals when selecting content to include. 4 of the 5 pages where we added SpeakableSpecification saw Google AI Overview citation improvements.

FAQPage had mixed results. It improved citation rates on pages where the FAQ content was genuinely structured as Q&A. It had no effect on pages where we added FAQ markup to content that wasn't actually FAQ-formatted. Schema that doesn't match the content structure doesn't appear to help.

SameAs entity linking may be underrated. We added SameAs links connecting each page's publisher to the company's Wikidata entry and LinkedIn profile. This didn't directly affect citation rates in 21 days — but it likely contributes to entity trust over a longer horizon. We're watching this.

For a full breakdown of every schema type and how it performs across platforms, we ran a separate analysis: Every Schema Type for 2026: How Google and AI Platforms Use Them Differently.

---

What This Changes

The takeaway isn't "add schema and get more citations." The takeaway is more specific:

Schema markup for AI search requires content alignment, not just markup addition. The pages that didn't respond to schema addition were the pages where the markup didn't match the actual content structure. FAQPage schema on a listicle doesn't help. HowTo schema on a page with no process steps doesn't help.

The right order is: structure your content first, then mark it up. If you add schema to content that doesn't have the right underlying structure, you're wasting implementation time.

Perplexity is the platform most responsive to schema changes right now. If you're tracking AI citations and Perplexity is in your dataset, schema additions will show the fastest measurable signal there. Use it as your leading indicator.

The 21-day window is real. Citation changes from schema additions started appearing within 9–14 days on Perplexity and ChatGPT. You don't need to wait 6 months to see if it worked. Run the experiment on 10 pages, check at 3 weeks.

---

How to Run This Experiment on Your Site

1. Pick 10 pages that are already getting some AI citations (you need a baseline). If you don't have AI citation tracking set up, this post covers how to build a manual tracking system.

1. Audit current schema — use Google's Rich Results Test + Schema Markup Validator. Most sites have minimal schema even if they think they've implemented it.

1. Identify schema opportunities per page type:

- Process pages → HowTo

• FAQ sections → FAQPage
• Blog posts with key quote sections → SpeakableSpecification
• Any page → Article + SameAs entity linking

1. Implement as JSON-LD only — avoid microdata for new implementations. JSON-LD is cleaner, easier to maintain, and easier to validate.

1. Set a 21-day checkpoint and compare citation rates manually across Perplexity, ChatGPT, and Google AI Overviews for your tracked queries.

---

Related Experiments

AI Citation Instability: 90-Day Tracking Data — the dataset that first surfaced the correlation between schema markup and citation stability. Schema-rich pages rotated out of citations 3.1x less often.

From Page 1 to Position Zero in 11 Days — AEO Framework Case Study — the position zero experiment used a similar schema stack. The content structure work came first.

AI Search Optimization: Strategic Moves for 2026 — the strategic framework this experiment sits inside.

---

Key Takeaways

• Schema markup produced a 408% increase in Perplexity citation rate across 14 pages in 21 days. The effect was real, fast, and measurable.
• HowTo and SpeakableSpecification are the highest-leverage schema types for AI citations. FAQPage only works when the underlying content is genuinely structured as Q&A.
• Perplexity responds to schema changes fastest. Claude responds slowest. Use Perplexity as your leading indicator in any schema experiment.
• Schema that doesn't match content structure doesn't help. Structure the content first, then mark it up.
• 21 days is enough to see the early signal. You don't need a 6-month timeline to decide if schema is working.

---

We're now 30 days in on this client. The next check is at 60 days — we want to see whether the gains hold or whether Perplexity's citation rotation starts to pull some pages back. We also added SameAs entity links 10 days ago and want to isolate that effect separately.

The 60-day update will be published in late May.