Token usage is the count of input and output units a language-model request consumed, reported alongside the cost of that request. Tokens are the small chunks of text a model reads and writes, and their total is what pricing is metered against.
| Term | Token Usage |
|---|---|
| Category | AI, Content Generation and Experiments |
| Also known as | Tokens |
| Where it appears | Ask Claude answer details |
What it means in RankGear
When you send a prompt through Ask Claude, RankGear records how many tokens the request used and surfaces the figure in the answer details for that response. The count is split between the input side, which covers the prompt plus any page context or data you attached, and the output side, which covers the text the model generated back. Because providers bill per token, this number is the direct link between what you asked and what the request cost.
How to interpret it
Read token usage as a size and cost signal, not a quality signal. A larger count means more text moved through the model in either direction: a long document you pasted in, a broad question, or a lengthy generated answer will all push it up. Input and output tokens are usually priced differently, so a request that reads a lot but replies briefly costs differently from one that reads little but writes at length. If a response is truncated, checking the output token count against the model’s limit often explains why.
| Component | What it tells you |
|---|---|
| Input tokens | Size of the prompt and any attached context sent to the model |
| Output tokens | Length of the generated answer returned to you |
| Total | The billed unit count for the request |
Example
You ask Claude to critique the topical coverage of a 2,000-word draft. The answer details report roughly 3,100 input tokens (your instructions plus the pasted draft) and 640 output tokens (the critique). Trimming the draft to the section you actually care about before the next pass would lower the input count and the cost, without changing the model or the question.
Important considerations
- Token counts describe request size and cost only. They say nothing about whether the answer is correct, and a high or low count carries no ranking meaning.
- Input and output tokens are typically billed at different rates, so the total alone does not tell you where the cost came from.
- Token accounting is comparative within a provider’s own pricing and limits, not a Google or search metric; do not read it as anything a page does to rank.
- AI output can be incomplete or wrong. Verify factual claims, statistics, citations, and experimental isolation before acting on a response, regardless of how many tokens it used.
Related terms
Part of the RankGear glossary · how RankGear measures · the 870 factors.