ObsurfableIf you optimize for "AI search" as one channel, you are optimizing for a surface where 97.3% of domains are invisible to four out of five engines.
SurfacedBy ran a controlled experiment between March and June 2026: real buyer and category questions, each posed to five AI engines, every cited source logged. The result was 127,198 citations pointing at 11,647 different websites across ChatGPT, Claude, Gemini, Perplexity, and Google AI Mode.
The first finding: the five engines barely read the same web.
- Only 2.7% of cited domains (309 out of 11,647) appeared in all five engines
- 69.6% of domains were cited by exactly one engine
- Adding domains cited by four engines, you still reach only ~7% that most engines agree on
Seven in ten sources were cited by only one of them. Winning on one barely carries to the others.
How the experiment worked
SurfacedBy took a sample of real buyer and category questions — the queries actual purchasers ask when researching products, comparing vendors, and evaluating categories. Each question was posed to all five engines. Every source cited in every answer was recorded and counted at the domain level.
The study looked only at cited sources — public websites the engines linked to — not at brand mentions in answer text. Domain-level counting prevents a single site with many pages from skewing results.
No vendor sponsorship. No single-industry focus. A straightforward cross-engine citation census.
The overlap numbers
| Overlap level | Domains | Share |
|---|---|---|
| Cited by all 5 engines | 309 | 2.7% |
| Cited by 4 engines | ~500 | ~4.3% |
| Cited by 3 engines | ~800 | ~6.9% |
| Cited by 2 engines | ~1,900 | ~16.3% |
| Cited by exactly 1 engine | 8,108 | 69.6% |
Read the bottom row again: more than two-thirds of all domains that any AI engine cites are invisible to the other four. A domain cited by ChatGPT has roughly a 30% chance of also being cited by any other single engine — and a 2.7% chance of appearing in all five.
This is consistent with — and more granular than — earlier overlap studies:
| Study | Sample | Overlap finding |
|---|---|---|
| SurfacedBy | 127K citations, 5 engines | 2.7% all-five overlap |
| CiteMetrix | 680M citations | 11% ChatGPT-Perplexity overlap |
| BrightEdge | Cross-engine | 16% minimum two-engine overlap |
| Foglift | 1,373 answers, 5 engines | ChatGPT ↔ Claude Jaccard: 0.027 |
| Yext | 6.8M citations | AI Overviews ↔ AI Mode: 13.7% URL overlap |
The SurfacedBy study is the most direct test: same prompts, same time period, all five engines, full domain mapping.
The engines cite different amounts — and different types
Citation volume per answer varies sharply:
| Engine | Avg sources per answer |
|---|---|
| Gemini | 11.0 |
| Perplexity | 8.6 |
| Google AI Mode | 7.8 |
| Claude | 6.8 |
| ChatGPT | 3.7 |
Gemini lists 3× more sources per answer than ChatGPT. Perplexity's high count reflects its strategy of citing multiple sources per claim. ChatGPT is the most selective — fewer sources, but from a marginally broader domain spectrum.
The source-type split is even more dramatic:
| Source type | Google AI Mode | Perplexity | ChatGPT | Claude |
|---|---|---|---|---|
| YouTube | 11.2% | 8.8% | Moderate | 0.02% |
| 4.0% | High | Moderate | 0.01% | |
| Documentation/vendor pages | Moderate | Moderate | High | High |
| Major publishers | High | Moderate | High | Moderate |
Claude almost never cites YouTube or Reddit. Google AI Mode and Perplexity lean heavily on both. ChatGPT favors major publishers and Wikipedia. A YouTube strategy that dominates Google AI Mode citations is nearly irrelevant for Claude.
Match the source type to the engine. This is not optional nuance — it is the primary driver of the 2.7% overlap.
Why overlap is so low: four retrieval backends
The 2.7% figure is not random. It tracks four distinct retrieval architectures:
- ChatGPT — Bing index + OAI-SearchBot + entity signals
- Claude — Brave Search + Claude-SearchBot
- Gemini / Google AI Mode — Google Search index + fan-out retrieval
- Perplexity — Proprietary index + freshness bias
Each backend indexes different pages, applies different ranking signals, and reranks for different source-type preferences. A page ranking #1 on Google may be absent from Brave's index. A Reddit thread dominating Perplexity may be invisible to Claude. Wikipedia powering ChatGPT may not appear in Gemini's citation stack.
CiteLens's July 2026 study quantified the organic-rank divergence: Google AI Mode cites 93% from Google's top 10; ChatGPT cites only 30%. See Three Citation Rulebooks.
Foglift's June 2026 study added a critical nuance: engines often agree on whether a brand belongs in the answer (85–98% brand-mention agreement) while disagreeing sharply on which sources to cite (0.027–0.643 Jaccard overlap). Perplexity agrees with Gemini on the brand decision 96.8% of the time but shares only 0.197 citation-domain overlap. Same answer, different sources.
What the 309 "universal" domains have in common
What do the 309 domains cited by all five engines share?
Cross-referencing with Ahrefs' most-cited domain data and the earned media studies, the universal cohort skews toward:
- Major publishers — Forbes, Business Insider, Wikipedia, In Plain English, established editorial brands
- High domain authority with presence across multiple indexes (Google, Bing, Brave)
- Broad topical coverage — not niche pages, but category-defining resources
- Earned media status — editorial, not brand-owned or paid
- Server-side rendered, crawlable content accessible to all retrieval bots
These are not pages you optimize into universal citation with formatting tricks. They are institutions — publications, encyclopedias, platforms — that every retrieval backend indexes and trusts.
For mid-market brands, the realistic goal is not universal citation. It is engine-specific citation on the backends your buyers use.
Strategic implications
Stop reporting "AI visibility" as one number
A brand cited on 3 of 5 engines does not have "60% AI visibility." It has three wins and two gaps — each requiring a different strategy on a different retrieval backend.
Prioritize engines by buyer behavior
Users under 44 average five search platforms (EMARKETER). ChatGPT processes 72 billion messages per month. Identify which engines your buyers actually use, then optimize and monitor those specifically — not all five equally.
Build engine-specific source portfolios
| If your buyers use... | Prioritize these surfaces |
|---|---|
| Google AI Mode / Gemini | YouTube, Google rankings, structured brand pages, topic clusters |
| Perplexity | Reddit, fresh content (<30 days), G2, niche reviews |
| ChatGPT | Wikipedia, Bing rankings, major publishers, entity mentions |
| Claude | Documentation, vendor pages, verifiable data, Brave-indexed content |
Accept that 97.3% of the web is engine-specific
The 2.7% universal citation rate is not a failure metric. It is the structural reality of four retrieval backends. Your job is to be in the citation set for the engines your buyers use — not to chase universal presence across all five.
Measure persistence, not just presence
A domain cited once by one engine is a transient event, not a position. With 73.5% of cited URLs appearing exactly once (Trakkr), run repeated checks across weeks to distinguish stable citation from noise. See AI Citation Volatility.
What to do now
1. Run your top 15 buyer prompts through all five engines.
Log which domains are cited per engine. Map your domain's presence: all five? Three? One? None?
2. Identify your strongest and weakest engines.
Double down on the engines where you are cited. Diagnose gaps on engines where competitors appear and you do not.
3. Match content distribution to engine source preferences.
YouTube for Google AI Mode. Reddit for Perplexity. Wikipedia and publishers for ChatGPT. Documentation for Claude. One content type does not serve all five.
4. Build earned media across 4+ platforms.
Brands on 4+ third-party platforms are 2.8× more likely to be cited in ChatGPT (5WPR). Earned media drives 84% of citations (Muck Rack). Third-party footprint is the closest path to multi-engine presence.
5. Monitor weekly with engine-separated dashboards.
Same prompts, five engines, five columns. Report gaps, not averages.
How Obsurfable fits
Obsurfable is built for the 2.7% problem — or more precisely, for the 97.3% where engines disagree. It runs your defined prompts across ChatGPT, Claude, Gemini, Perplexity, and Google AI Mode in parallel, tracking citation presence, brand mentions, and competitor positioning per engine.
When you are cited on Gemini but invisible on ChatGPT for the same buyer question, Obsurfable shows both outcomes — not a blended score that hides the gap.
The Visibility Director identifies engine-specific gaps and drafts content tuned to the retrieval backend where you are missing.
For the infrastructure behind the divergence, see Why Each AI Engine Reads a Different Search Index.
FAQ
Is 2.7% overlap increasing or decreasing?
No published trend data yet, but the figure has been stable across studies from June through July 2026. The divergence appears structural — tied to retrieval architecture — not a temporary artifact.
Should I try to get cited by all five engines?
For major publishers and encyclopedic resources, multi-engine presence is achievable. For most brands, prioritize the 2–3 engines your buyers actually use and accept engine-specific citation as the realistic outcome.
Does paying for AI visibility tools guarantee multi-engine presence?
Tools measure presence. They do not create it. Multi-engine citation requires indexing across four backends, earned media on engine-appropriate surfaces, and content matched to each engine's source preferences.
How does this relate to the 11% ChatGPT-Perplexity overlap?
The 11% figure (CiteMetrix) measures overlap between two engines. The 2.7% figure (SurfacedBy) measures overlap across all five. Both confirm the same pattern at different granularity: cross-engine citation agreement is structurally low.
Bottom line
Only 2.7% of domains are cited by all five AI engines. 69.6% are cited by exactly one. The five engines read different indexes, prefer different source types, and cite different amounts per answer. "AI search optimization" is five separate strategies — or it is a strategy that leaves 97.3% of the citation surface unaddressed. Monitor all five in parallel, or accept that four of them are blind spots.