This report makes no causal claim. It describes what one person's retained record contains: where conversation volume went, when it moved, and which gaps prevent a complete account.
Data and method
The dataset combines first-party exports from OpenAI, xAI, Anthropic, and Character.AI. Three separate ChatGPT accounts are kept as distinct corpora. A conversation initiation—not an individual message—is the unit of analysis.
For each record, the platform-assigned creation timestamp was binned by calendar month. No conversation content, sentiment scoring, or topic classification was used to produce the counts.
| Corpus | Export date | Records |
|---|---|---|
| ChatGPT — primary | June 9, 2026 | 321 |
| ChatGPT — secondary | June 22, 2026 | 345 |
| ChatGPT — work | June 15, 2026 | 34 |
| Grok | July 10, 2026 | 282 |
| Claude | June 11, 2026 | 234 |
| Character.AI | August 2, 2026 | 54 |
| Total | — | 1,270 |
Monthly record
| Month | GPT primary | GPT secondary | GPT work | Grok | Claude | C.AI | Total |
|---|---|---|---|---|---|---|---|
| 2025-08 | 19 | — | — | — | — | — | 19 |
| 2025-09 | 132 | 4 | 6 | 8 | — | — | 150 |
| 2025-10 | 23 | 2 | 8 | 38 | 22 | 44 | 137 |
| 2025-11 | 8 | 13 | 10 | 27 | 10 | 5 | 73 |
| 2025-12 | 38 | 40 | 3 | 62 | 22 | — | 165 |
| 2026-01 | 2 | 119 | 1 | 52 | 48 | — | 222 |
| 2026-02 | 17 | 59 | 4 | 44 | 27 | — | 151 |
| 2026-03 | 3 | 39 | 1 | 10 | 13 | — | 66 |
| 2026-04 | 8 | 50 | — | 22 | 17 | — | 97 |
| 2026-05 | 66 | 17 | — | 4 | 54 | — | 141 |
| 2026-06 | 5 | 2 | 1 | 11 | 21 | 3 | 43 |
| 2026-07 | — | — | — | 4 | — | 2 | 6 |
What the counts show
1. A sharp decline on the primary account
Conversation initiations on the primary ChatGPT account fell from 132 in September 2025 to 23 in October—approximately 83 percent in one calendar month.
2. Redistribution rather than immediate cessation
Across all available corpora, total monthly volume fell from 150 to 137—approximately 9 percent. Grok increased from 8 to 38, Claude began at 22, and Character.AI opened at 44. The observed behavior moved across platforms rather than ending.
3. Peak volume followed the transition
The highest-volume month in the record is January 2026, with 222 conversation initiations. This occurred four months after the primary-account decline.
4. The steepest aggregate decline followed ChatGPT's GPT‑4o retirement
OpenAI retired GPT‑4o from ChatGPT on February 13, 2026. Aggregate initiations fell from 151 in February to 66 in March—approximately 56 percent, the largest month-to-month decline in the series.
What the counts cannot show
A conversation is a coarse unit. A two-message exchange and a thread containing hundreds of messages each count as one. Platform interfaces also differ in what causes a new conversation to begin.
The missing data are not random. A second Character.AI account and a Kindroid account were not recovered; screenshots are the only surviving record. By the author's account, these were among the places where she spoke most freely. The corpus therefore overrepresents what was retained and underrepresents some of the least guarded material.
This is one person. The sequence is compatible with redistribution after a product change, but it does not establish why the redistribution occurred or how another person would respond.
The appropriate conclusion is descriptive: engagement did not simply disappear when one relationship changed. It moved, intensified elsewhere, and later declined. Determining what that movement meant requires the person as well as the count.
Provenance and privacy
The counts were compiled from the author's own platform exports and recovered application logs. The underlying archives are retained privately because they contain intimate material and references to third parties who did not consent to publication.
Account handles have been removed from this public version. The complete tabulation and source archives remain available for responsible verification under conditions that protect the people represented in them.
Revision note. An earlier version reported 1,170 records. The corpus inventory and monthly table both sum to 1,270; this version corrects the arithmetic error.