AI Product Rollouts — Adoption & Retention
What the data actually shows from OpenAI, Google, Microsoft & Anthropic
Big-lab AI products have hit unprecedented scale. Retention diverges sharply: ChatGPT leads on stickiness, Gemini leads on subscriber growth, Copilot battles a seat-use gap, and Claude trades consumer volume for developer depth.
Key Findings
ChatGPT's month-12 desktop retention (50%) is roughly double Gemini's (25%). But Gemini's paid-subscriber growth (≈300% YoY) outpaces ChatGPT's (≈155% YoY). More users are trying the alternatives — fewer are staying.
What this means for product rollouts:
- Raw user counts are misleading. A product can add tens of millions of sign-ups while losing the majority by month 3.
- Paid conversion locks in retention. Paid users retain at 57–68% by month 12, free users at 25–50%.
- Feature launches rarely break through. None of ChatGPT's 2024–25 drops (Tasks, Study Mode, Shopping, Record) showed clear retention signal.
Scale — Monthly Active Users (est., mid-2025)
Data
| Product | |
|---|---|
| ChatGPT | 1000M |
| Gemini | 750M |
| Copilot | 420M |
| Claude (consumer) | 30M |
Sources: OpenAI CEO via press; Alphabet Q4 2025 earnings; secondary estimates; Anthropic press. All indirect; treat as order-of-magnitude.
Engagement: DAU/MAU Ratio
Data
| DAU/MAU (higher = more habitual) | |
|---|---|
| ChatGPT | 36% |
| Gemini | 21% |
Source: Yipit data cited in a16z State of Consumer AI 2025. Copilot and Claude ratios not publicly available.
The 36% vs 21% gap means ChatGPT users return far more habitually than Gemini users, despite Gemini's larger raw sign-up base.
Retention at Month 12
Data
| Free users | Paid users | |
|---|---|---|
| ChatGPT | 50% | 68% |
| Gemini | 25% | 57% |
Source: Yipit panel data, a16z 2025 report. Figures reflect desktop web cohorts; mobile and API cohorts differ.
Free-to-paid conversion is the single largest retention lever. The paid gap (68% vs 57%) is narrower than the free gap (50% vs 25%). Gemini locks in paying users even as casual users churn.
Subscriber growth YoY
Data
| Paid subscriber growth (YoY, 2024→2025) | |
|---|---|
| Gemini Pro | 300% |
| ChatGPT Plus | 155% |
Source: a16z 2025 consumer AI report. Gemini's higher growth reflects a lower base; absolute ChatGPT Plus subscribers still outnumber Gemini Pro.
Microsoft Copilot — The Pilot-to-Production Gap
Microsoft Copilot is the clearest case of adoption without use in enterprise AI:
- 70% of Fortune 500 had "adopted" Copilot as of FY25 Q1 — but most deployments were pilots or phased rollouts, not company-wide.
- 79% of surveyed enterprises reported deploying Copilot (Morgan Stanley/RSM, Jul 2025), with half past pilot.
- Seat use gap: one estimate put active use at ~8M of 15M paid seats (≈53%) as of Aug 2025. Microsoft has not confirmed this.
- DAU doubled QoQ during FY2025 — a growth-rate claim, not a retention figure.
- No published renewal or churn rate for enterprise Copilot exists in public disclosures.
The 70% Fortune 500 figure conflicts with other estimates (64%, >90%) by source. All Copilot figures are approximate until Microsoft publishes seat-use data in earnings.
Anthropic Claude — Developer-First, Consumer-Thin
Claude's adoption pattern is inverted: smaller consumer base, stronger developer and enterprise depth.
- Consumer web: ~18.8M unique visitors (Nov 2024) → ~54M (early 2025) → directional estimates of 220M+ MAU in Q1 2026 (methodology unclear).
- Mobile app: 2.9M MAU (Jan 2025) → 7.4M MAU (mid-2025), a 2.5× rise.
- Enterprise: Anthropic reported 300,000+ business customers (Sep 2025). Claude Code weekly active users doubled since Jan 2026.
- No public retention rate for consumer Claude exists. Developer and API stickiness appears high based on revenue, but no cohort data is public.
Feature launches vs. retention
No major big-lab feature launch in 2024–25 showed a durable retention uplift. The pattern:
- Spike at launch — power users and press drive a DAU surge.
- Decay by week 4 — casual users ("AI tourists") churn. Median Day 30 app retention is 4% across all apps.
- Stabilization — the metric that matters is where the curve flattens, not the peak.
a16z recommends measuring retention from Month 3 (not Day 1) to filter out launch tourists.
Data quality scorecard
| Claim | Source type | Reliability |
|---|---|---|
| ChatGPT 1B MAU (May 2026) | CEO statement, unaudited | directional only |
| DAU/MAU 36% vs 21% | Yipit panel · a16z | best available |
| Month-12 retention 50% / 25% | Yipit panel · a16z | best available |
| Gemini 750M MAU | Alphabet earnings write-ups | not verified in primary filing |
| Copilot 70% Fortune 500 | Microsoft investor call | "adoption" ≠ company-wide deployment |
| Claude 220M MAU (Q1 2026) | Third-party aggregator blogs | methodology unclear |
| Copilot 8M/15M active seats | Secondary blog | unverified; no primary source |
Takeaways
- Retention is the real launch metric. MAU is easy with distribution. Month-3 and month-12 retention separates a product from a feature.
- Paid conversion is the biggest lever. Free-user churn is structurally high. Getting users to pay cuts churn significantly.
- Enterprise adoption is not use. Copilot shows you can sell seats to 70% of the Fortune 500 and still have most seats unused.
- Feature launches rarely retain. Big-lab feature drops show no durable cohort lifts. Retention comes from the core loop.
- Primary data is scarce. The best public retention figures come from one panel (Yipit) in one report (a16z 2025). Treat all numbers as indicative until corroborated.