Why We Ran This Test
Search Console shows impressions and clicks. It doesn't show you whether your content is being cited in AI-generated answers. For sites with significant TOFU content, that blind spot is a real problem.
We started this because a client noticed something strange: their traffic from branded searches was flat in GSC, but their site kept appearing and disappearing in ChatGPT answers when we manually tested branded queries. We couldn't explain it. So we built a manual tracking system and started logging it.
The setup: 3,900 queries across 8 topic clusters, tested weekly on ChatGPT (GPT-4), Perplexity, Claude 3.5 Sonnet, and Google AI Overviews. Same queries, same time of day, same phrasing. We logged which sources each platform cited for each query, then tracked changes week over week.
This is not a dataset that comes out of GSC or Ahrefs. We built it manually.
The Setup
```
Queries tracked: 3,900
Topic clusters: 8 (B2B SaaS, local services, e-commerce, agency services, healthcare informational, legal informational, finance,marketing/SEO)
Platforms tested: ChatGPT (GPT-4), Perplexity, Claude, Google AI OverviewsDuration: 90 days
Cadence: Weekly snapshots, same day and time Tracking method: Manual logging into structured spreadsheet Sites in dataset: 47 domains (mix of client sites and competitors) ```We were not measuring click-through rate or traffic. We were measuring one thing: which source does each AI platform cite for each query, and does that source stay consistent over time?
What We Found
Finding 1 — Citation Sources Rotated Every 11 Days on Average
Across all 3,900 queries, the citation source changed at least once every 11 days on average. For informational queries with multiple credible sources, rotation happened as frequently as every 5–6 days.
This is not a margin-of-error artifact. The rotation is real, consistent, and appears across all four platforms — though the rotation frequency differed.
``` Platform | Avg citation rotation (days)
- ----------------------|-----------------------------
Perplexity updates its citation sources fastest. Claude is the most stable. If you're optimizing for consistent AI citations, the platform matters more than any SEO metric.
Finding 2 — Domain Authority Had No Significant Correlation With Citation Frequency
We pulled domain authority scores for all 47 domains in our dataset. We then compared DA against how often each domain appeared as a citation source across the 90-day period.
The correlation coefficient was 0.12. Effectively zero.
The sites cited most often were not the high-DA generalist sites. They were smaller, more focused sites with tight entity coverage on specific topics. One domain with a DA of 34 was cited more often than a DA 78 competitor across the same query cluster — because the lower-DA site had more specific, more consistently structured content on that exact topic.
This is not an argument against building domain authority. It's an argument that domain authority alone tells you nothing about your AI search exposure.
Finding 3 — Schema Markup Created a Measurable Stability Effect
The finding we didn't expect: pages with comprehensive schema markup (beyond just article/breadcrumb — including HowTo, FAQ, SpeakableSpecification, and SameAs entity linking) appeared as citation sources with significantly higher consistency.
Non-schema pages rotated out of citations at a 3.1x higher rate than schema-rich pages in the same query cluster.
We can't prove causation from this data alone. What we can say: when a page dropped out of citation rotation, it was 3x more likely to be a non-schema page. When a page held its citation across all 12 weekly snapshots, 78% of those pages had comprehensive schema.
This aligns with a plausible mechanism: AI platforms are using structured data signals to establish entity trust, not just content relevance. A page that explicitly tells the machine what it is and who published it gets treated with more reliability.
For more on which schema types are doing the work, we ran a separate 30-day test: Every Schema Type for 2026: How Google and AI Platforms Use Them Differently.
Finding 4 — The "Citation Window" Closes in 90 Days for New Content
New content — pages less than 30 days old at the start of our tracking — appeared in AI citations in the first two weeks on Perplexity and ChatGPT. But they also dropped out most frequently by week 6–8.
The pattern: new content gets tested in the citation pool early. If it generates engagement signals (we're hypothesizing here — we can't see platform-level engagement data), it stabilizes. If it doesn't, it rotates out and older, more established pages take the slot.
This suggests a publishing strategy implication: new content that you want cited consistently needs to be promoted actively in the first 30 days. Not just published and indexed — actively distributed so it accumulates real engagement signals before the citation algorithm deprioritizes it.
Finding 5 — AI Platforms Don't Agree With Each Other 61% of the Time
For the same query, across all four platforms, all four platforms cited the same source only 39% of the time.
``` Query: "how to reduce customer churn in SaaS"
ChatGPT cites: Site A (DA 71) Perplexity cites: Site B (DA 44) Claude cites: Site C (DA 58) Google AI Overview: Site A (DA 71)
Agreement: 2/4 platforms (50%) ```
This was roughly consistent across topic clusters. The four platforms agreed on citation sources about 39% of the time. They agreed on two or more sources about 61% of the time.
The implication: "AI SEO" is not one thing. Optimizing for Perplexity citations requires a different approach than optimizing for Claude or Google AI Overviews. The platforms weight different signals differently. Anyone selling you a one-size-fits-all "AI SEO" strategy hasn't looked at this data.
What This Changes
If AI citations rotate every 11 days on average, the "win and hold" model of SEO doesn't apply here. You're not trying to capture a position — you're trying to stay in the rotation pool.
That requires a different operational model:
- Schema is now maintenance, not setup. It's not something you do once during a site audit. It needs to be on every new page, reviewed quarterly, and updated when schema standards change.
- Entity clarity beats backlink volume for AI. The question is no longer "who links to this page" but "does an AI model know unambiguously what this page is about, who published it, and why they have authority on this topic."
- New content needs a 30-day promotion sprint. Publish and pray doesn't build citation stability. Publish, distribute, build engagement signals in the first 30 days.
- Platform-specific monitoring matters. If 61% of your AI citation opportunity is platform-specific, you need to know which platform your audience uses. For B2B, that's probably Perplexity and ChatGPT. For general consumer queries, Google AI Overviews. Track them separately.
How to Test This on Your Site
You don't need to track 3,900 queries. Start with 50.
- Pick your 50 highest-impression informational queries from GSC. These are the queries where you're most likely to have AI Overview or AI platform competition.
- Manually test each query on ChatGPT, Perplexity, and Google. Log whether your site is cited. Do this on a Tuesday morning to reduce day-of-week variance.
- Repeat the same test 14 days later. Compare which queries now cite different sources. Calculate your citation retention rate.
- For every query where you rotated out: check whether the page that replaced you has schema markup yours doesn't. Check entity clarity — does the page explicitly define what it is, who it's by, and what makes it authoritative?
- Prioritize schema additions for pages where you lost citations between week 1 and week 3.
Related Experiments
This data connects directly to work we've done on schema markup and position zero:
AI Search Optimization: Strategic Moves for 2026 covers the broader strategic shift from traditional SEO to AI-mediated discovery. If you haven't read it, the citation instability data makes more sense in that context.
Every Schema Type for 2026 is the companion post — it covers which markup types actually move the needle for AI citations vs which ones are Google-specific.
We also ran a separate experiment using the AEO (Answer Engine Optimization) framework that got one of our pages to position zero in 11 days. That case study is here. The schema patterns in that experiment are consistent with what we're seeing in citation stability.
Key Takeaways
- AI citation sources are unstable. The same query cites different sources every 11 days on average. This is not a bug — it's how these systems currently work.
- Domain authority is not predictive of citation frequency. Entity clarity and schema density are better predictors.
- Comprehensive schema markup creates citation stability. Pages with it rotated out of citations 3.1x less often than pages without.
- New content has a 30-day window to establish citation stability. What happens in that window matters.
- The four major AI platforms agree on citation sources only 39% of the time. "AI SEO" is not one optimization target.
We're now 30 days into a follow-up experiment: we're testing whether adding SameAs entity links (connecting pages to their Wikidata and schema.org entity equivalents) changes citation stability. Preliminary data suggests it does. That post will be up in May.
In the meantime: run the 50-query test. The results from your own site will tell you more than any industry report.
Continue Reading
Ready to Grow Your Business?
Let's discuss how Digital Ingenuity can help you achieve your marketing goals with AI-powered strategies.
Get Started TodayNeed help with this?
If implementing these strategies feels overwhelming, we're here to help. Fill out the form below and we'll schedule a free consultation to discuss your specific situation.