QUICK ANSWER
ChatGPT Token Counter
You can count the visible text copied from a ChatGPT conversation, but that is not the exact total processed by ChatGPT. A pasted transcript omits message formatting and may omit system instructions, tool definitions, images, files, reasoning tokens, and other context. For OpenAI API requests, use OpenAI’s input-token endpoint or the usage returned with the response.
What a pasted ChatGPT transcript can tell you
When you copy the visible messages into a tokenizer, you get the token count for that copied text under the selected encoding. This is useful for comparing prompt length or identifying unusually large messages. It is not a complete measurement of the conversation state used internally by ChatGPT.
Why the exact total can be different
OpenAI’s documentation explains that a model request can include structural tokens for message roles and boundaries. A request may also include tools, schemas, images, files, earlier turns, and non-visible generated tokens. These items are not reconstructed simply by copying the text displayed in a chat window.
How to count an OpenAI API request
For a Responses API request, OpenAI provides an input-token counting endpoint that accepts the same request format, including supported messages, files, images, tools, and conversations. It returns the count the model will receive before generation. After a real request, the response usage fields are the appropriate source for actual input and output token usage.
Which counter should you use?
- To inspect copied ChatGPT text, use the OpenAI Token Counter.
- To plan token-based API cost after you have a count, use the Token Cost Kit calculator.
- To count a complete API request, follow OpenAI’s official token-counting guide.