Merriam-Webster Premium subscribers had the highest paid-AI overlap
Share with a paid ChatGPT, Claude, Perplexity, Grammarly, or Gemini receipt, January 2024–August 2026
Share with a paid AI receipt
Source: YipitData U.S. e-receipt panel. Merriam-Webster Premium n=517; Merriam-Webster book or Kindle buyers n=1,829; general panel n=1,207,958.
Chart data
| Series | Share with a paid AI receipt |
|---|---|
| Merriam-Webster Premium subscribers | 45.5% |
| Merriam-Webster book buyers | 26% |
| General panel | 12.4% |
How quickly did paid-AI overlap rise in the subscriber cohort?
Cumulative paid-AI adoption among the 517 Merriam-Webster Premium subscribers rose from 8.1% in the first quarter of 2024 to 43.9% in the second quarter of 2026. This measures whether an AI receipt had appeared by each point, not whether every subscription remained active.
Did subscribers who adopted paid AI while active retain less often?
In a directional 2024-subscriber comparison, 71.7% of 99 users who adopted AI by or while active had a recent Merriam-Webster receipt, versus 61.9% of 197 subscribers with no paid-AI receipt. The cohorts were not randomly assigned, so the difference does not show that AI improved retention.
Does the analysis cover Britannica.com subscriptions?
No. Direct Britannica.com consumer subscriptions produced only about 335 observable users over two and a half years, so this article is limited to Merriam-Webster Premium, the measurable consumer subscription asset. Paid-AI receipts indicate wallet presence rather than product use.
Assumptions & Methodologies
We analyzed established U.S. e-receipt panelists from January 2024 through August 2026. Merriam-Webster Premium required a subscription receipt; the comparison included one-off Merriam-Webster dictionary or Kindle buyers and the general panel. Paid AI covered ChatGPT, Claude, Perplexity, Grammarly, and Gemini. The directional retention split used 2024 subscribers with at least 20 months of observation and a recent renewal-receipt proxy. Cohorts were modest, app-store templates can add noise, the panel skews slightly male and upper-middle income, and results measure association rather than usage, substitution, or causal retention effects.