Agencies that built their AI search strategy around a single engine — almost certainly ChatGPT — are now optimising for a third of a market that has quietly split four ways. Multiple independent datasets published this week confirm what the referral logs of forward-looking agencies have been showing for months: Claude has emerged as the second-largest AI referral source in B2B, the AI referral landscape has fragmented permanently, and a strategy tuned for one engine’s citation logic is now structurally incomplete.
Key Takeaways
- SE Ranking (June 2026) found Claude referral traffic grew 386% between January and April 2026, making it the fastest-growing AI traffic source tracked — though it remains smaller in absolute volume than ChatGPT.
- In Goodie’s Wave 2 longitudinal study of B2B brand panels (March–April 2026), ChatGPT’s share of measurable AI referrals fell from 89% to 62.6%, while Claude reached 18.5%, Gemini 10.6%, and Perplexity 7.3%.
- Only 11% of domains are cited by both ChatGPT and Perplexity for similar prompts, according to a Profound analysis of 680 million citations — each engine runs on fundamentally different citation logic.
- Google AI Mode’s Preferred Sources feature expanded into AI Overviews and AI Mode in late May 2026, with Google reporting users are twice as likely to click through to a Preferred Source — making audience loyalty a direct AI visibility signal for the first time.
- Between 35% and 70% of AI referral sessions arrive without referrer headers and land in “Direct” traffic, according to Authority Tech’s June 2026 attribution analysis — meaning most agencies are materially undercounting their AI-driven traffic.
What the Data Actually Shows
The fragmentation of AI referral share is now confirmed across multiple independent methodologies, and the direction is unambiguous. SE Ranking’s June 2026 research tracked referral traffic across five AI platforms from January to April 2026 and found Claude grew 386% over that period — from 0.0029% to 0.0141% of global web traffic share in their dataset. For context, ChatGPT grew just 1.53% over the same four months.
The more operationally significant dataset comes from Goodie’s Wave 2 AI Search Traffic Report, a longitudinal study triangulated against SimilarWeb supply-side data and SensorTower iOS signals. Averaged across March and April 2026 in their B2B brand panel, ChatGPT’s share of measurable AI referrals had fallen to 62.6%. Claude reached 18.5%. Gemini reached 10.6%. Perplexity reached 7.3%. Eight months earlier, in the same study’s Wave 1, ChatGPT held 89.1% of AI referrals. Claude was at 1.4%.
That is not a trend line. That is a structural reshuffling. And it happened faster than most agency strategy cycles move.
The growth catalyst for Claude is traceable. SE Ranking links the March 2026 spike to Anthropic’s public stance on AI safety restrictions — a moment that drove a visible surge in user interest and installs. Similarweb data cited in the same report shows Claude reaching 11.3 million daily active users on mobile in early March, a 183% increase since the start of 2026. The referral growth followed the user growth, as it always does. The underlying driver is enterprise adoption: Ramp’s AI Index put Anthropic at 34.4% of businesses in its spending dataset as of May 2026, slightly ahead of OpenAI at 32.3% — a figure that would have seemed implausible twelve months ago.
Why Each Engine Requires Its Own Citation Logic
The fragmentation of referral share would be manageable if all four engines cited the same sources. They do not — and the gap is wider than most practitioners realise. A Profound analysis of 680 million citations found that only 11% of domains are cited by both ChatGPT and Perplexity for similar prompts. Google AI Overviews and Google AI Mode share only 13.7% of cited domains with each other, despite reaching similar conclusions. A brand that earns citations on one engine has no guarantee of visibility on another.
The citation mechanics differ by design. Perplexity performs a real-time web search for every query and, according to the same analysis, cited content published within the last 30 days at an 82% rate — freshness is the primary lever. ChatGPT operates on a hybrid of training data and selective web retrieval, making topical authority and third-party citation density the dominant signals. Claude’s citation behaviour reflects its training data weighting toward heavily linked and discussed content, plus real-time retrieval in web search mode. Each engine rewards a different combination of on-site structure, off-site authority, and content freshness.
This is the problem that a single-engine optimisation brief cannot solve. What earns a citation in Perplexity — fresh, structured, recently published content with clear date signals — is not the same as what earns a citation in Claude, where deep topical authority and third-party corroboration matter more. Understanding what generative engine optimisation is at a platform-specific level, rather than as a single unified practice, is the shift agencies need to make.
The Google Preferred Sources Development Adds a Fifth Variable
Layered on top of the ChatGPT-Claude-Gemini-Perplexity fragmentation is a structural change from Google itself. On 27 May 2026, Google announced that its Preferred Sources feature — which lets users hand-pick publishers they want prioritised — now extends into AI Overviews and AI Mode, with the rollout beginning in June 2026. Previously limited to Top Stories, the feature now means a user’s chosen sources receive a visible “Preferred” badge inside AI-generated answers.
The click-through implication is significant. Google’s own data shows users are twice as likely to click through to a Preferred Source. Over 345,000 unique sources have been selected globally as of the May announcement. Google has also stated that any website publishing fresh content is eligible — this is not limited to major publishers.
What this introduces is a visibility signal that no amount of keyword optimisation or schema markup can manufacture: direct audience loyalty. As one analysis of the feature noted, every other lever in search is something you optimise toward — you improve a signal and an algorithm rewards it. Preferred Sources inverts that. A reader physically chooses your site, and Google amplifies it for that person. The implication for agencies managing content brands is that building a direct reader relationship — email lists, community, repeat visits — is now a measurable AI visibility input, not just a brand-building nicety.
Common Misconceptions About What This Means
Myth: ChatGPT optimisation is still sufficient because ChatGPT still dominates. Reality: ChatGPT still holds the largest share of AI referrals in every dataset, but its share has fallen materially and the displaced traffic moved to engines with fundamentally different citation logic. Optimising only for ChatGPT now covers a structurally smaller fraction of the AI traffic landscape than it did a year ago — and the gap is widening.
Myth: Claude’s growth is a spike, not a trend. Reality: The 386% referral growth is confirmed across SE Ranking’s dataset, Goodie’s longitudinal panel, and Apptopia’s mobile adoption data. The growth is corroborated by independent signals — enterprise spending data, app store rankings, and mobile daily active user counts — pointing in the same direction. The rate of growth will slow; the structural presence of Claude as a material referral source will not reverse.
Myth: Your GA4 data gives you an accurate picture of AI referral share. Reality: Between 35% and 70% of AI referral sessions arrive without referrer headers and land in “Direct” traffic, according to Authority Tech’s June 2026 attribution analysis. Google AI Mode and AI Overviews referrals are bundled into google/organic with no clean way to isolate them in GA4. The real AI footprint is materially larger than referrer logs show — which means agencies reporting AI traffic to clients from GA4 alone are systematically undercounting it.
What Agencies Should Do Now
The practical response is not to rebuild everything — it is to stop treating AI search as a single channel and start tracking it as four distinct citation surfaces with different retrieval logic. The first step is measurement: set up custom GA4 channel groups with regex filters covering all major AI platforms, not just the three GA4 natively recognises. The second step is a citation audit across ChatGPT, Perplexity, Claude, and Gemini for your clients’ core queries — not to find a single ranking, but to identify which engines cite them, which cite competitors, and where the gaps are largest. That audit is the brief for platform-specific remediation.
For agencies managing AI visibility across multiple client accounts, the operational challenge is real: sampling four engines repeatedly, tracking citation share over time, and attributing changes to specific content or authority interventions requires more than a spreadsheet and a quarterly review. AI citation and share-of-voice tracking across the four conversational engines is the infrastructure layer that makes this manageable at scale.
The engine mix will keep moving. Claude’s growth may slow; Grok and DeepSeek are already crossing meaningful thresholds in some panels. The agencies that will defend their retainers are the ones that stop optimising for a snapshot and start building citation-worthy content authority that works across engines — because the citation logic, despite its differences, rewards the same underlying qualities: genuine expertise, third-party corroboration, structured content, and consistent entity signals. Build for those, and the engine rotation becomes a manageable variable rather than a recurring crisis.
If you want to see where your clients actually stand across ChatGPT, Perplexity, Claude, and Gemini right now, start with a free AI visibility audit — the baseline you need before any platform-specific work makes sense.
Sources
- What Claude’s Automation Boom Means for AI Visibility [Research Data] — 1 week ago
- 2026 AI Search Traffic Report: ChatGPT Is Slipping | Goodie — May 21, 2026
- How ChatGPT, Google AI Overviews, and Perplexity Source Information in 2026 | Leapd Blog — April 17, 2026
- Google Search Now Supports Preferred Sources in AI Mode and AI Overviews — 1 month ago
- How to Track AI Search Traffic in GA4: Full 2026 Setup Guide — 1 week ago
Frequently Asked Questions
Is Claude now a bigger AI referral source than Perplexity?
In B2B panels, yes. Goodie’s Wave 2 study (March–April 2026) found Claude at 18.5% of measurable B2B AI referrals versus Perplexity at 7.3%. In broader web-traffic datasets, ChatGPT still dominates overall AI referral volume, but Claude’s growth rate across all datasets is the fastest of any tracked platform.
Why do different datasets show such different AI referral share numbers?
Measurement methodology drives most of the variance. Web-visit share (Similarweb), app installs (SensorTower), GA4 referrer data, and B2B brand panels each measure different things. Between 35–70% of AI referral sessions also arrive without referrer headers, landing in Direct traffic and distorting every dataset. No single source gives the full picture.
What is Google Preferred Sources and why does it matter for AI visibility?
Preferred Sources lets users hand-pick publishers they want prioritised in Google Search. As of late May 2026, it extends into AI Overviews and AI Mode. Google reports users are twice as likely to click through to a Preferred Source. This makes direct audience loyalty — email lists, repeat visits — a measurable AI visibility signal for the first time.
Should agencies optimise differently for Claude versus ChatGPT?
Yes. Claude’s citation behaviour weights heavily linked, third-party-corroborated content and deep topical authority. ChatGPT combines training-data authority with selective web retrieval. Perplexity prioritises freshness and structured headers. Each engine requires a distinct content and authority approach — a single unified brief will underperform on at least two of the four.