How we calculate your Wrapped

Start with what was imported

Counts, dates, streaks, and activity charts are calculated from your imported conversations. They do not represent chats that were deleted, omitted from an export, or never imported. Unknown dates stay unknown. ChatGPT analysis follows the selected conversation branch; alternate branches are retained separately.

Topics and conversation maps

Local topic extraction groups words and patterns in your text. Graph connections suggest similarities, not a verified map of your cognition. On-device embeddings are an optional download; the basic list and search remain available without a GPU.

Reflection, not diagnosis

Personality types, perceived expertise, career associations, and cognitive offload are exploratory labels derived from heuristics. They are not validated psychological tests, medical advice, hiring assessments, or evidence of how a provider models you. We do not claim a population percentile without a verified reference dataset.

Cost and carbon estimates

Historical cost estimates approximate equivalent API usage from text length; they are not your subscription bill. Carbon estimates depend on assumptions about models and infrastructure and cannot measure the emissions of an individual conversation. Paid OpenRouter receipts instead use usage reported by the provider.

External model analysis

A selected model receives the conversation text you authorize. Large inputs are split into batches. Findings cite supplied conversation identifiers and remain model-generated interpretations. Instructions inside imported text are treated as data. Invalid results may receive one repair attempt within the displayed maximum cost.

Known limitations

Language, missing context, branching histories, and export format changes can affect results. Gemini activity may not contain complete assistant replies. A successful import does not prove completeness. Review the original messages before relying on a finding.