The same week voice AI grabbed headlines, a data report quietly punctured the most comfortable assumption in the whole GEO world. Goodie's 2026 AI Search Traffic Report found that eight months ago, ChatGPT held 89% of B2B "AI referrals"; today it holds 63%. In other words, optimizing only for ChatGPT now reaches roughly a third less of the AI traffic map than it did a year ago. The headline is blunt — ChatGPT is slipping.
Why now? Over the past 12 to 18 months, the "answer layer" has moved from one dominant player to a multi-way split. On July 10, 2026, Google made Gemini 3.5 Flash the default across all searches, so every query now opens with an AI-generated answer and the classic blue links get pushed below the fold — publisher clicks fell about 58%. Meanwhile Anthropic's Claude, Perplexity, and various vertical assistants are all competing for the same "research and commercial-investigation" queries — exactly the high-value searches that used to send organic traffic to content sites. The market is no longer small: generative-AI sites received about 9.5 billion monthly visits from June 2025 to May 2026, up ~70% year over year.
The sharpest cut in Goodie's report is the speed of fragmentation: Claude jumped from 1.4% to 18.5%, Gemini quadrupled, and Perplexity more than doubled. Its methodology isn't a single self-serving source — it uses a GA4 referrer panel of anonymized brands, triangulated against SimilarWeb's 25.77 billion visits across the top five AI surfaces (Jan–Apr 2026) and SensorTower's iOS rankings. The one-line conclusion: "The Big 1" has become "The Big 4," and the category is moving from monopoly toward a split.
Will this hit SMBs? Yes — and it's half bad news, half good. The bad: you can no longer watch just one engine. The good: most competitors haven't started multi-engine GEO, so movers get a head start. This article unpacks how to watch all four at the lowest possible cost, without buying an expensive enterprise monitoring subscription.
Event Details and Full Numbers
Lay out Goodie's key numbers and you see a market being reshaped:
- ChatGPT referral share: B2B AI referrals fell from 89% eight months ago to 63%.
- Challengers rising: Claude 1.4% to 18.5%; Gemini roughly quadrupled; Perplexity more than doubled.
- Query volume: per SimilarWeb's 2026 AI Search data, ChatGPT Search processes ~250–500 million weekly queries, Perplexity around 50 million.
- Overall traffic: generative-AI sites average ~9.5 billion monthly visits, up ~70% year over year.
- Methodology: GA4 anonymized brand panel + SimilarWeb top-five surfaces (25.77B visits, Jan–Apr 2026) + SensorTower iOS rankings, triangulated.
Who wins, who loses? Winners are sites whose content is formatted to be cited — research shows lists, tables and step-by-step formats get cited far more than paragraph-only content. Losers are brands that bet all their GEO on one engine and only watch overall rank: when four AIs cite different sources, you can be first in engine A and simply absent in engine B.
Immediate Actions for Three Reader Types
- Brand owners / SMB owners: First make "which AIs actually cite you" a measurable number. Today, set up four referrer segments in GA4 (chatgpt, perplexity, gemini, claude) and check the distribution in two weeks — data tells you where to spend, instead of chasing ChatGPT on gut feel.
- Marketing / SEO operators: Stop "one piece rules them all." For each topic, prepare a citable, structured version: a 40-word direct answer up top, tables and lists in the body, a verifiable source on each claim. This works across all four AIs at near-zero cost.
- Developers / agencies: Turn "AI traffic attribution" into a deliverable service. GA4 doesn't capture AI sources cleanly by default; you can build clients' referrer rules, UTM standards, and a dashboard that separates the four engines. It's the most painful and least-served gap right now.
SaaS Tool Comparison Table
| Tool | Focus | Pricing tier | Best for |
|---|---|---|---|
| Goodie | AI citation/traffic reports, cross-engine visibility | Subscription (mid-high) | Brands tracking citation trends across all four |
| Profound | Enterprise AI visibility monitoring | Enterprise (high) | Mid-large brands with budget |
| Scrunch / Evertune | Brand appearance-rate tracking in AI answers | Subscription (mid) | Marketing teams doing AEO/GEO monitoring |
| DIY: GA4 + referrer segments | Self-built four-engine source tracking | Near-zero (labor) | Budget-sensitive SMBs |
What They Won't Tell You
1. "Share falling" is not "ChatGPT shrinking." ChatGPT's absolute query volume is still growing (250–500M weekly); it just no longer owns the incremental growth. Reading "relative share" as "absolute decline" makes you abandon a still-largest single source too early. The real signal is "don't put all eggs in it," not "it's collapsing."
2. Panel data has built-in bias. Goodie's sample is a GA4 panel of specific brands, skewed B2B and unevenly distributed by industry. It tells you the direction of the trend, but your industry's four-way split may look completely different. No third-party report should replace your own GA4 data — that's the only sample 100% correct for you.
3. "Multi-engine GEO" easily becomes "4x the work." Forcing four versions of the same piece is the dumbest approach. All four AIs actually prefer the same underlying traits: clear structure, verifiable, answer-first. Doing the right structure once beats surface-optimizing four times. Don't let vendors scare you into four subscriptions.
The "No SaaS Subscription" Alternative for SMBs
- Build four GA4 segments yourself: use referrer contains chatgpt.com, perplexity.ai, gemini.google.com, claude.ai to create four segments or Explorations — two weeks shows who cites you.
- UTM and internal-link standards: for any page you want cited, ensure clean titles, summary metas and structured data (Google structured-data docs) so all four can parse it.
- One manual tracking sheet: every two weeks, ask each of the four AIs three questions you care about and log "cited or not, who was cited." Low-tech, but enough to replace most entry-level monitoring subscriptions.
- Template your content structure: build a "direct answer + table + list + source" writing template so every piece is born citable.
Should I abandon ChatGPT optimization and go all-in on the other three?
No. ChatGPT is still the largest single source; it just shouldn't monopolize your effort. The right move is "keep ChatGPT while adding Perplexity, Gemini and Claude to tracking," and let data decide the mix.
With no budget for monitoring tools, how do I know if AI cites me?
Build four GA4 referrer segments plus a bi-weekly manual question test — that covers most needs. Upgrade to paid tools only once data proves your AI traffic is large enough.
Do the four AIs need very different optimization?
Their underlying needs overlap heavily: clear structure, answer-first, verifiable sources. Differences are in details (Perplexity leans on real-time sources; Gemini on Google-ecosystem signals). Nail the common structure first, then fine-tune.
My Take
The mainstream reads this report as "go do multi-engine GEO now." My call is more contrarian: "the Big 4 split" is a transition state, not the endgame. The answer layer's historical pattern is fragment first, then re-consolidate — as Google locks Gemini into its ecosystem and engines start charging "pay when cited," share could well re-concentrate to two or three within 18 months. So I would not advise SMBs to spend on four monitoring subscriptions chasing every fluctuation; that's the old rank-chasing anxiety under a new name. What you should actually do is nail your content's "citable structure" and stand up AI-source tracking in your own GA4 — two things that won't be wasted whether the endgame is Big 4 or Big 2. For ScriptWalker, this is a clear productization opening: building clients' "AI-traffic attribution + citable content structure" infrastructure has more lasting value than selling a rank report that expires.
Sources
- Goodie — 2026 AI Search Traffic Report: ChatGPT Is Slipping (first-hand report)
- Google — Search I/O 2026 updates (Gemini default and AI Mode)
- Google Search Central — Structured Data docs (citable-content practices)
- SimilarWeb 2026 AI Search Report (top-five surface visits and query volume)
- SensorTower iOS rankings (triangulation source)