Research
Retrieval Layer Intelligence Report
Monthly benchmarks and strategic guidance for navigating the shift from Search Engines to Action Engines.
Published February 10, 2026 · Last updated March 12, 2026
Research Methodology
This report was put together by Hayden Bond, an independent AI SEO practitioner with no vendor relationships and no affiliate arrangements. Data is triangulated from SimilarWeb, First Page Sage, and a16z consumer app rankings. This page is updated monthly as the market moves.
Market Update: March 2026
- •ChatGPT drops below 65% globally for the first time, now at 61.7% web traffic share per SimilarWeb February 2026 data
- •Copilot surges to 10.4% North American market share, up 5.4 points week-over-week, driven by GPT-5.2 integration and Fortune 500 enterprise penetration
- •Claude hits #1 on the US App Store on March 1 - first time ever. Catalyst was the Pentagon dispute after Trump directed federal agencies to stop using Anthropic products. Free users up 60%+ since January, paid subscribers more than doubled in 2026
- •Grok and Claude overtake DeepSeek for 3rd and 4th place in global web traffic for the first time
The Multi-Model Reality
Web traffic share is no longer the whole story.
Enterprise Adoption
81%
of Global 2000 companies now run three or more AI model families concurrently. Picking one model is no longer the strategy. Source: a16z, January 2026.
Multi-Tenanting
20%
of weekly ChatGPT web users also use Gemini in the same week. Consumer behavior mirrors enterprise strategy - this market is not winner-take-all. Source: a16z, March 2026.
Embedding Economy
Copilot + Claude
Microsoft integrated Anthropic's Claude into Copilot Cowork for agentic M365 workflows - choosing a competitor's model because it was better at the task. The biggest enterprise AI platform in the world runs multiple models. Source: Microsoft 365 Blog, March 9 2026.
ChatGPT Global Share
61.7%
down 14pp year-over-year
Gemini Global Share
24.4%
up 18.7pp year-over-year
Copilot NA Share
10.4%
up 5.4pp week-over-week
Claude App Store
#1
US iOS, March 1 2026
Agentic Ecosystem Status
When a user asks an AI agent to research a topic, book a trip, or compile a report, the agent clicks, reads, and acts. The human may never see your site at all. These platforms are where that agent activity originates. Scale tells you how much of it is happening. Bounce rate tells you how much intent the human behind the agent actually had.
ChatGPT
Monthly Visits
5.35B
Bounce Rate
32.3%
Traffic Channel
Direct (76%)
Dominant by volume. Visits down from January peak of 5.72B as competition intensifies.
Gemini
Monthly Visits
2.11B
Bounce Rate
29.9%
Traffic Channel
Direct (77%)
More than quadrupled web share in 12 months, from 5.7% to 24.4%.
Grok
Monthly Visits
298.5M
Bounce Rate
N/A
Traffic Channel
Direct (est.)
Overtook DeepSeek for 3rd place globally for the first time in February 2026.
Claude
Monthly Visits
290.3M
Bounce Rate
26.7%
Traffic Channel
Direct (75%)
Crossed 3% global web share for the first time in February 2026. Lowest bounce rate in the cohort.
DeepSeek
Monthly Visits
246.4M
Bounce Rate
N/A
Traffic Channel
Direct (est.)
Dropped to 5th globally after Grok and Claude surpassed it for the first time.
Perplexity
Monthly Visits
153.8M
Bounce Rate
32.2%
Traffic Channel
Direct (73%)
Stable niche in AI-native search. High-intent research audience.
Copilot
Monthly Visits
~95M (web only)
Bounce Rate
31.3%
Traffic Channel
Direct (82%)
Web traffic severely understates actual market position. 85% of Fortune 500 use Microsoft AI tools. 15 million paid seats as of Q1 2026.
The Infrastructure Shift: From Search to Action
The market pivots toward agentic action with Google's Universal Commerce Protocol (UCP) representing a Tier 2 Infrastructure signal.
Bypassing the Click
UCP enables agents to execute purchases via a standardized API layer, removing the need for users to visit merchant sites.
Transactional Authority
Competition shifts from a battle for clicks to a battle for transactional authority within agent ecosystems.
Agent-Ready Brands
Agentic readiness becomes critical for direct commercial outcomes as AI agents mediate more transactions.
Referral Metrics
When an LLM does send a user to an external site, this is where that traffic lands. Referral share measures each platform's slice of total AI-driven referral traffic arriving on third-party sites. Small numbers, high intent.
Gemini
24.4%
12-Month Trend: +18.7pp
More than quadrupled share from 5.7% in February 2025. Structural market shift confirmed.
ChatGPT
61.7%
12-Month Trend: -14.0pp
Still dominant but 12-month erosion is substantial. Two-player race with Gemini now established.
Grok
3.4%
12-Month Trend: +2.9pp YoY
Entered top 3 for the first time. Mobile-driven growth and X integration.
Claude
3.3%
12-Month Trend: +1.8pp YoY
Crossed 3% mark for the first time in February 2026.
DeepSeek
3.2%
12-Month Trend: —
New entrant in tracked cohort.
Perplexity
1.8%
12-Month Trend: -0.2pp
Stable niche position.
Copilot
1.1% (web only)
12-Month Trend: —
Copilot's 1.1% web share is the most misleading number in this report. Microsoft AI tools are deployed across 85% of the Fortune 500 with 15 million paid seats as of Q1 2026. Web traffic does not capture enterprise deployment. If your buyers work in large organizations, Copilot is not a Tier 2 platform.
Optimization Guidance
AEO Strategy
ChatGPT and Gemini now control 86% of AI referral traffic combined. Optimize for broad informational coverage on both. For Claude and Perplexity, focus on citation quality. Structured, authoritative content is what agents pull from when compiling answers. Lower bounce rates on both signal high-intent users worth capturing.
GEO Priority
The next competitive layer is transactability. Build API-first architecture for product catalogs and checkout so AI agents can complete purchases on a user's behalf without a site visit. Brands that are not agent-ready will be invisible in agentic commerce.
Modern SEO
Schema markup, clean product feeds, and accurate technical documentation are the signals quality-focused platforms like Claude and Perplexity use to determine what to trust and cite. Get the structured data right and the representation follows.