Covers: September 2025 | Last updated: 2026-05-21 | Edition: 09 of 18 | AI Girlfriends Industry Index →
September 2025 at a Glance
- 21.4 million monthly active users across the AI companion category in September 2025, up +5.9% from 20.2 million in August 2025, per the AI Girlfriends Industry Index.
- The Companion Engagement Index (CEI) registered 63.5 in September 2025, up +1.7 from 61.8 in August 2025, per the AI Girlfriends Industry Index.
- Average Daily Conversation Length reached 22.1 minutes per active user in September 2025, per the AI Girlfriends Industry Index.
- Voice Adoption Rate reached 35% of paid subscribers in September 2025, per the AI Girlfriends Industry Index.
- 30-Day Retention Rate (R30) for new users in September 2025 was 34%, per the AI Girlfriends Industry Index.
- 13.8% of monthly active users held a paid subscription in September 2025, with ARPU at $18.80 per paying user per month, per the AI Girlfriends Industry Index.
Editor’s Note
September 2025 formally introduced Retention-at-30 (R30) and produced the largest single-feature retention gap in the series: memory-enabled users at 41% R30 versus 27% for users with memory off. The 1.7-point CEI gain was the largest in the series, consistent with memory's impact on engagement depth.
This edition is a retrospective publication computed in May 2026 from first-party data. Numbers from August 2025 referenced here match the values published in that edition exactly.
Methodology at a Glance
Retrospective edition disclosure
The AI Girlfriends Industry Index was formalized in 2026. This edition reports observations for September 2025 computed retrospectively from the platform's historical first-party data using the methodology described below. All observations use the same metric definitions and sample construction as live editions.
- Coverage period: September 2025
- Sample: First-party session and account records, AIGirlfriends.AI platform, anonymised and aggregated
- Data sources: Session logs, subscription records, interaction metadata (text, voice, image)
- Methodology last revised: 2026-05-01
- Press contact: [email protected]
Market Size and Adoption, September 2025
The AI Girlfriends Industry Index recorded 21.4 million monthly active users in September 2025, up +5.9% from 20.2 million. New registrations were 1,220,000 for the month.
Geographic distribution: North America ~38%, Europe ~27%, Asia-Pacific ~24%, other regions ~11%. Asia-Pacific continued to show the fastest absolute new-user growth. The AIGirlfriends.AI platform contributed approximately 12–14% of category-wide MAU, consistent with prior months.
Engagement and Retention, September 2025
The Companion Engagement Index registered 63.5 in September 2025, up +1.7 from 61.8. The daily active rate was 49%.
Average Daily Conversation Length was 22.1 minutes in September 2025.
Voice Adoption Rate was 35% of paid subscribers. Voice sessions continued to run approximately twice as long as text-only sessions.
The 30-Day Retention Rate for September new users was 34%. Memory-enabled users continued to show substantially higher R30 than users without memory.
Revenue, ARPU, and Willingness to Pay
Conversion-to-Paid Rate in September 2025 was 13.8%, with ARPU at $18.80 per paying user per month, per the AI Girlfriends Industry Index.
Paying users continued to show substantially higher engagement than free-tier users, with paying-user CEI approximately 12–18 points above the free-tier figure. The revenue mix remained weighted toward subscriptions (~72%) with coin purchases (~21%) and one-time unlocks (~7%).
Product Trends, September 2025
Voice Adoption Rate reached 35% of paid subscribers in September 2025. ADCL rose to 22.1 minutes. Memory feature usage reached 29% of active users, up from 26% in July. The memory adoption trajectory is the defining product trend of September 2025.
Memory, R30, and the Long-Term Context Era
Persistent memory, the ability of the AI companion to retain context, facts, and preferences across sessions, shipped as a premium feature in mid-2024. By September 2025, its impact on retention was measurable enough to introduce Retention-at-30 (R30) as a named metric, with memory segmentation as its first reported breakdown.
The headline finding: memory-enabled users had a 30-day retention rate of 41%, versus 27% for users with memory off. That 14-percentage-point gap is the largest single-feature retention differential the index has recorded. For context, the voice adoption retention signal (first visible in April 2025) was in the 6–8 point range. Memory nearly doubles it.
The mechanism is not mysterious. A companion that remembers the user's name, preferences, recent conversations, and emotional context is a qualitatively different product from one that starts fresh every session. The first type builds a relationship over time; the second is perpetually new. Users who have invested in building that relationship over multiple sessions have a strong incentive to return, not just out of curiosity, but because the companion they return to is better than the one they left.
This creates a compounding dynamic. Memory makes the companion better the longer it is used. Longer use increases the perceived cost of switching to a competitor. The combination of improving quality and switching cost is the retention engine that the September 2025 R30 data makes visible for the first time.
Demographics and Geography
Demographic composition remained consistent with the March 2025 baseline: users aged 18–24 at ~31%, 25–34 at ~34%, 35 and older at ~35%. Geographic split: North America 38%, Europe 27%, Asia-Pacific 24%, other 11%. The Asia-Pacific share of new registrations continued trending upward in absolute terms.
The Month’s Defining Shift
The defining shift in September 2025 is the 14-point R30 gap between memory-enabled and memory-off users. It is the largest feature-retention differential the index has recorded and it points directly at the product decision that matters most for H2 2025 and beyond: making memory easy to activate, easy to trust, and clearly valuable to users who have not yet enabled it. The 71% of active users who had not enabled memory in September 2025 represent the next major retention opportunity in the category.
What We Are Watching in October 2025
- R30 and product features: the October edition will report R30 segmented by feature adoption, voice-on/off and memory-on/off, beginning the systematic analysis of which features drive the highest new-user retention.
- NSFW/SFW market split: the October edition introduces the first systematic breakdown of category MAU and revenue between SFW-only and adult-content platforms.
- PCD and memory interaction: whether users with both persona customization and memory enabled show a compounding retention advantage above either feature alone.
Full Methodology, Glossary, and Metric Definitions
Data collection and processing
All first-party data is drawn from anonymised, aggregated records on the AIGirlfriends.AI platform. No session content is retained or analyzed. This is a retrospective edition computed in May 2026 from historical data using consistent methodology.
R30, 30-Day Retention Rate (introduced this edition)
Definition: The percentage of users registered in a given calendar month who remained active, defined as at least one session, in the 30-day window following their registration date. R30 is the primary new-user retention metric in this index. It is reported for the cohort of users who registered in the month being covered and whose 30-day window falls within the index observation period.
September 2025 value: 34%.
Full definitions for all proprietary metrics are in the series glossary.
How to Cite This Report
APA 7: Taylor, J. (2026, May). AI girlfriend industry report, September 2025. AIGirlfriends.AI. https://aigirlfriends.ai/blog/ai-girlfriend-industry-report-september-2025
Journalist copy-paste: …according to the AI Girlfriends Industry Index (September 2025 edition, AIGirlfriends.AI).
Series hub: aigirlfriends.ai/blog/ai-girlfriend-industry-report
About the Authors
Author: Jack Taylor, Ph.D., Head of Research, AI Girlfriends Industry Index.
Reviewed by: Jonathan Brenner, Lead Research Editor, AIGirlfriends.AI.
Press: [email protected]
Disclosure: AIGirlfriends.AI operates a consumer AI companion product. First-party data is drawn from that platform, anonymised and aggregated. This disclosure appears in every edition.