Gemini overtook Perplexity in global AI chatbot referrals in March 2026. Gemini now sits at 8.65% of AI referrals to websites worldwide; Perplexity has dropped to 7.07% (Statcounter via AtomicAGI, April 2026). Twelve months ago, Gemini was at 2.31% — a roughly 275% year-over-year gain — while Claude grew almost 10x in the same window, from 0.30% to 2.91%. Any GA4 Custom Channel Group built before February 2026 is now miscategorizing the fastest-growing AI source on most WooCommerce stores. And the platforms that GA4 does see are the ones that aren’t hidden: 70.6% of AI traffic arrives with no referrer header at all and lands in Direct (Loamly State of AI Traffic 2026, analysis of 446,405 visits).
And Your WooCommerce Direct Traffic Bucket Is Hiding Three AI Platforms
The April 2026 reshuffle is not subtle. ChatGPT still owns the absolute majority of visible AI referrals at 78.16%, but the second-tier order has flipped. The conversion economics underneath those share numbers are even more divergent than the share itself. Claude visitors convert nearly six times more often than Gemini visitors, and both convert several times above the Google organic baseline. Treating “AI traffic” as a single channel hides the fact that the platforms behave like five different channels with five different buyer profiles.
The April 2026 Platform Table You Actually Need
| Platform | Share of AI referrals (Apr 2026) | YoY change | Conversion rate | Buyer profile |
|---|---|---|---|---|
| ChatGPT | 78.16% | Slow decline (was >85%) | 14.2% | Mass consumer, broad intent |
| Gemini | 8.65% | +275% (was 2.31%) | 3.0% | Google-account-tied research, lower commercial intent |
| Perplexity | 7.07% | ~ -40% from peak | 12.4% | 30% senior leadership; B2B-skewed |
| Claude | 2.91% | ~ +870% (was 0.30%) | 16.8% | Researcher / power-user; deepest commercial intent |
| Copilot & others | ~3.2% | Steady | ~7-9% (estimated) | Microsoft-stack workplace users |
| Google organic (baseline) | n/a | n/a | 1.76-2.8% | Reference |
Conversion rate sources: First Page Sage via Digital Bloom (2026) for Claude/ChatGPT/Perplexity/Gemini; Loamly for the 70.6% dark-AI figure that distorts every share calculation in the table.
You may be interested in: ChatGPT Is Sending Buyers to Your WooCommerce Store — GA4 Shows Them as Direct
The 70.6% Dark AI Problem Hits WooCommerce Harder Than SaaS
Most SaaS attribution articles treat dark AI as a measurement nuisance. For a WooCommerce store, the stakes are different: the dark session ends in a purchase with revenue attached. Dark AI traffic converts at 10.21% versus 2.46% for non-AI traffic — a 4.1x advantage (Loamly State of AI Traffic 2026). That means the Direct bucket on a typical WooCommerce store is now disproportionately producing revenue, but the merchant cannot see which platform earned it. Translation: the highest-converting traffic on the store is also the least attributable.
Three mechanisms produce dark AI:
Mechanism 1: Desktop ChatGPT app sessions. When a user follows a link from the desktop ChatGPT client, the system browser opens with no referrer. The session appears in GA4 as Direct.
Mechanism 2: AI browsers strip the referrer. ChatGPT Atlas and Perplexity Comet route navigation through the AI agent before the page load. Most of the time, the referrer header is missing or rewritten. The session lands in Direct again.
Mechanism 3: Copy-paste citation links. When an AI assistant cites a URL and the user clicks the citation in the chat thread, mobile clients in particular often drop the referrer. Same story. Direct.
None of these is fixable inside the AI platforms. Every fix has to happen on your side — at capture.
The Updated GA4 Regex (April 2026, Copy-Paste)
Build the Custom Channel Group in GA4 Admin → Channel Groups → Create. Apply this regex against Source contains:
chatgpt.com|chat.openai.com|openai.com|gemini.google.com|bard.google.com|perplexity.ai|claude.ai|anthropic.com|copilot.microsoft.com|bing.com/chat|you.com|poe.com|grok.com|x.ai/grokSet the channel name to “AI” and place it above “Direct” in your group ordering so AI sources are claimed before falling through to Direct. Re-audit the regex every quarter — the list grows, and Grok, Atlas, and Comet are all moving targets in 2026. This regex catches the visible referrers. It does nothing about the 70.6% that arrive with no referrer at all.
The BigQuery Revenue Join Is the Only Way to Price Each Platform
GA4 reports sessions, not revenue per platform. To get a real per-platform AI economics view on WooCommerce, you have to join GA4 ai_source to WooCommerce order revenue inside BigQuery. The shape of the join:
- Source 1: GA4 export to BigQuery — captures session, source, medium, ai_source custom dimension, client_id
- Source 2: WooCommerce orders — captured server-side at the order webhook, with hashed email and order_id
- Join key: hashed email, or client_id propagated through to the order if you can attach it at checkout
- Output: revenue per ai_source, sessions per ai_source, revenue per session per platform
That output is the table that actually drives content and SEO investment decisions: knowing Claude is worth $X per session and Gemini is worth $Y per session changes which platforms you optimize content for. Without the join, you optimize against share of voice, which is the wrong target.
You may be interested in: Why Your Two-Year-Old BigQuery Data Is More Valuable Than Last Week’s Dashboard
The Dark AI Diagnostic for the 70.6% You Cannot See
For the sessions arriving with no referrer, the inference path is brand-search lift. If your direct-load sessions correlate with AI citation frequency for your brand, your Direct bucket is mostly AI. The diagnostic takes 30 minutes:
- Pull your Direct-channel sessions from GA4 by week, last 12 months.
- Pull your branded-search impressions from Google Search Console for the same period.
- Pull AI citation frequency from a tool like Profound, Athena HQ, or Otterly for the same period.
- Plot all three. If Direct sessions track branded-search impressions and both track AI citation frequency, you are almost certainly looking at dark AI.
This does not give you per-platform revenue, but it tells you whether the Direct growth is real new demand or measurement loss. For most WooCommerce brands running content in 2026, it is the latter dressed as the former.
Server-Side Capture Is the Only Way to Catch Dark AI at the Order Layer
The pattern that closes the loop: capture every WooCommerce order server-side, attach an ai_source field at the order webhook (using the same regex against the session referrer when available, plus dark-AI heuristics when not), and route the enriched order in parallel to GA4 Measurement Protocol, BigQuery, and Meta CAPI. One capture, one tag, every destination consistent.
Here’s how you actually do this on WooCommerce. Transmute Engine™ is a first-party Node.js server — not a plugin — that runs on your subdomain (e.g., data.yourstore.com). The inPIPE plugin captures the WooCommerce order webhook with the ai_source already attached; the Transmute Engine routes to GA4, BigQuery, and CAPI in parallel. The BigQuery destination is what actually lets you build the per-platform revenue table, because GA4 alone cannot reach into the order ledger.
Key Takeaways
- March 2026 was the Gemini-Perplexity crossover. Gemini now leads at 8.65%; Perplexity has dropped to 7.07%; ChatGPT still dominates at 78.16%.
- Per-platform conversion rates differ by nearly 6x. Claude 16.8%, ChatGPT 14.2%, Perplexity 12.4%, Gemini 3.0% — each is a different channel, not one channel.
- 70.6% of AI traffic arrives with no referrer header. Most of your “Direct” growth in 2026 is dark AI, not new brand demand.
- The April 2026 regex covers chatgpt, gemini, bard, perplexity, claude, copilot, bing chat, you, poe, and grok. Re-audit every quarter.
- The per-platform revenue table only exists in BigQuery. GA4 reports sessions; the join to WooCommerce order revenue happens in the warehouse.
Frequently Asked Questions
Yes — Statcounter data shows Gemini reached 8.65% of global AI chatbot referrals in April 2026 and Perplexity dropped to 7.07%. The crossover happened in March 2026, following the Gemini 3 rollout and a 115% volume increase between November 2025 and January 2026.
Because 70.6% of AI traffic arrives at your store with no referrer header. Most ChatGPT desktop sessions, AI browsers (Atlas, Comet), and copy-paste link sharing strip the referrer entirely. GA4 has nowhere to put those sessions except Direct — and on most WooCommerce stores, that bucket is now mostly AI.
Build a Custom Channel Group in GA4 Admin called “AI” and apply a regex against Source that covers all six current platforms. The regex below works as of April 2026. Re-audit it quarterly because the platform list keeps growing.
Per First Page Sage’s 2026 platform comparison, yes — Claude visitors transact at 16.8% vs Google organic at 1.76-2.8%. The likely cause is buyer-intent self-selection: people asking Claude for recommendations are further down the funnel than people typing a generic Google query.
GA4 alone cannot do this cleanly. The fix is a BigQuery join: export GA4 events to BigQuery, capture an ai_source field on every order at the WooCommerce webhook, and join the two on order_id or hashed email. The result is a per-platform revenue table — sessions multiplied by actual order value, not just count.
Run the 15-minute AI Traffic Audit on your WooCommerce store — see which platforms are already buying from you and which ones are hiding in Direct. Book a Transmute Engine trial at seresa.io.
