AI Summary
Desktop accounts for 94% of AI search referral traffic and mobile for just 6%, according to BrightEdge’s April 2025 analysis of thousands of brand referral sessions. The same primary data shows ChatGPT at 94% desktop, Perplexity at 96%, Gemini at 94%, and Bing at 95%. This is the opposite of traditional Google search, where mobile accounts for the majority of query volume globally.
Before citing this finding: we have verified each figure against BrightEdge’s published press release and the SEJ article authored by BrightEdge’s co-founder. The “90% of citations come from desktop” framing in the original post title is accurate as a summary; the per-platform numbers tell the more precise story. Where we cannot confirm a specific percentage against a primary source, we say so and build on what is verifiable instead.
What the data actually shows, platform by platform
BrightEdge’s June 2025 press release reported April 2025 referral data across leading AI platforms. The desktop splits:
| Platform | Desktop referrals | Mobile referrals |
|---|---|---|
| ChatGPT (chatgpt.com) | 94% | 6% |
| Perplexity.ai | 96% | 3% |
| Microsoft Bing | 95% | 4% |
| Google Gemini | 94% | 5% |
| Google Search (for comparison) | 44% | 53% |
Source: BrightEdge Generative Parser, April 2025, based on thousands of actual website referrals for medium to large brands. The lone exception is Google Search itself, which retains a mobile majority due to its entrenched position as the default browser search engine, particularly Safari on iPhones.
SparkToro’s March 2026 analysis of 5,000 most-visited sites cited the same BrightEdge figure, noting “AI usage is heavily desktop-skewed (about 94% desktop vs. 6% mobile).” One caveat: SparkToro re-cites the same BrightEdge data rather than measuring device split separately, so treat it as a single primary source, not independent corroboration.
Where AI search actually happens by device
The behavior split follows the interface, not just the platform. On desktop, AI search tools (ChatGPT, Perplexity, Gemini, Claude) are primarily accessed through web browsers. On mobile, the same tools are primarily accessed through native apps. Those two environments behave differently in ways that matter for measurement and optimization.
Browser-based access (desktop-dominant) sends referrer headers when users click through to sources. App-based access (mobile-dominant) frequently strips those headers. The ChatGPT iOS and Android apps open outbound links in environments that drop referrer data before the request reaches your server. The result: a real user interaction from mobile AI search arrives in your analytics as a Direct session, not a referral. We cover this attribution problem in detail in the GA4 AI search attribution guide and the AI traffic dashboard template.
Why the referrer-stripping problem is a mobile-specific issue
Three mechanics drive mobile referrer loss:
- In-app preview sandboxing. Mobile AI apps show citation content inside the app before routing users outward. That sandboxed intermediate step breaks the referrer chain. On desktop, first clicks on citations typically go directly to the source URL.
- strict-origin-when-cross-origin policy. Browser security policies on mobile apps send only the origin (not full URL) as referrer on cross-origin navigation, making the session appear direct once it lands on your domain.
- Voice entry points. Mobile AI search increasingly starts via voice on Siri, Google Assistant, or in-app voice modes. Voice-initiated sessions have never carried standard referrer headers. There is no verifiable volume data we can cite for voice-specific AI search share; we are being explicit about that gap rather than fabricating a percentage.
The practical consequence: your GA4 “Direct” bucket almost certainly contains AI-originated mobile sessions you cannot identify by source. The dark funnel problem in B2B is partly an AI measurement problem now.
Desktop AI search: the structural reasons for dominance
The 94% desktop skew is not random. Three structural factors sustain it:
- AI search is a work activity. ChatGPT, Claude, and Perplexity usage concentrates in business hours on work devices. Users are researching vendors, drafting documents, or analyzing data, not casually browsing.
- Interface design favors larger screens. Desktop AI Overviews command 80% more screen real estate than mobile (1,110 pixels versus 617 pixels) and appear for 39% more keywords than their mobile counterparts, according to BrightEdge’s data cited in the SEJ analysis. Longer prompts, multi-paragraph responses, and citation panels are genuinely harder to use on a 6-inch screen.
- Apple controls the mobile bottleneck. BrightEdge found Apple phones account for 58% of Google’s mobile traffic to US and European brand websites. Until Apple embeds AI-native search into Safari’s default behavior, the mobile AI referral market remains structurally constrained by that gatekeeping role.
What the device split means for content strategy
If 94% of AI search referral traffic is desktop-based, content design should follow that audience. This does not mean ignoring mobile; it means understanding the two surfaces are optimized differently.
For desktop AI search, which currently drives most verifiable referrals: long-form, structured content works. Our May 2026 study of 153,425 citations found 74.9% of cited sentences sit in the first half of the document. Tables extract cleanly on desktop. Desktop AI interfaces display citation panels prominently; the title tag and meta description of your cited URL influence whether users click through. See the BLUF writing format guide for the sentence-level approach and content chunking for RAG for the structural approach.
For mobile AI search, which is mostly invisible in referral data today but represents future surface area: optimize for discovery intent rather than research intent. Mobile AI queries are shorter, more local, and more transactional. The content spec is different from desktop research queries. Google’s AI Overviews data shows ecommerce queries are three times more likely to trigger mobile AI Overviews (13.5% versus 4.5% on desktop), with shopping-oriented discovery intent dominant.
Measurement: segmenting device in AI traffic
Most GA4 setups mix AI referral traffic into a general Referral channel without device breakdowns. To act on the desktop skew, segment explicitly:
- Build a custom GA4 channel group filtering for known AI referrers: chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, copilot.microsoft.com. The GA4 attribution guide has the full referrer list and setup steps.
- Layer the Device Category dimension on top to confirm the desktop-to-mobile ratio in your own data. Your ratio may differ from the BrightEdge average depending on industry and audience.
- Track the Direct channel in parallel. Mobile AI sessions that lost their referrer still arrived somewhere; unusually high-quality Direct sessions (low bounce, high pages-per-session, conversions) may be AI-originated mobile traffic you cannot see as referrals.
- Use the GEO/AEO Tracker to monitor which pages earn citations across platforms. Cross-reference those pages against their device breakdown in GA4 to find where desktop citation traffic is converting.
What we do not know (and why it matters)
Several things are frequently cited in this space that we are not repeating because we cannot verify them against primaries:
- Voice-specific AI search volume splits. Many sources cite percentages for voice-initiated AI search on mobile. We have not found a primary study with a verifiable methodology that isolates voice-to-AI from traditional voice search. We are not inventing a number here.
- Conversion rate multiples (desktop AI vs mobile organic). Various reports claim desktop AI converts at 1.5x to 3x mobile organic. We have not been able to confirm this range against a primary source with a disclosed methodology. Use your own GA4 data to measure this ratio in your context.
- Session duration comparisons. Claims about 2x to 3x longer desktop AI sessions circulate without disclosed primary data. Your own Engagement Time metric in GA4, segmented by the custom AI channel group, is the only number worth trusting here.
Honest uncertainty is more useful than confident fabrication. The desktop dominance finding is solid (BrightEdge primary data). Build your strategy on that and measure the rest yourself.
Is the desktop skew permanent?
No. The BrightEdge analysis from June 2025 framed the mobile frontier as “open” precisely because it has not been captured yet. Two shifts could change the device split rapidly:
- Apple’s Safari default. Apple controls the default search behavior on roughly a billion mobile devices. A shift of Safari’s default to an AI-native search experience, or deep integration of Apple Intelligence into search, would redistribute AI referral traffic toward mobile overnight. BrightEdge’s Jim Yu called this “the most valuable real estate in the mobile search landscape.”
- Referrer improvements in mobile apps. If ChatGPT’s iOS app, Perplexity’s Android app, or others change how they handle outbound link clicks to preserve referrer headers, the measured desktop dominance could shrink substantially without underlying behavior changing at all. What looks like a device preference may partly be an instrumentation artifact.
The implication for your AI search measurement framework: treat the current 94%/6% split as a baseline to track over time, not a permanent truth to optimize around. Monitor it quarterly against your own GA4 data. For a systematic view of where AI search is heading platform by platform, the Perplexity vs Google market share post and our 42,971-citation study give the citation-side context.
If you want to understand how your brand is currently cited across platforms and devices, the AI citation tracking service tracks this across 6 platforms with device-aware reporting.