Workflow
LibraryBrowse guides
Workflow utility · runs in your browser

Count the words before you prompt.

A calm, local word and character counter for shaping AI prompts before you send them. It counts the text you provide; it does not claim to calculate model-specific tokens.

Prompt scratchpad

Drop .txt or .md

Length planning by model

Approximate context use
GPT-40 / 8,192 tokens
Within planning range
Start with a preset
Customize this template

0Words
0Characters
0Lines
0Est. tokensApprox. for GPT-4
0 minReading timeAt about 200 words/min
0 minSpeaking timeAt about 130 words/min
$0.0000Est. API cost Illustrative input rate: $30.00 per 1M tokens for GPT-4. Actual provider pricing, output tokens, and billing rules may differ.GPT-4 planning rate

Planning estimate only: based on the approximate input token count and an illustrative input rate. Provider pricing, output tokens, and billing rules may differ.

Token estimate is for rough planning only; actual tokenization varies by model and tokenizer.

Use the count as a check.

A longer prompt is not automatically a better prompt. Look for the smallest context and constraint set that lets another person review the request.

Brief-first check

  • Is the task specific?
  • Is the audience named?
  • Are constraints visible?
  • Can the output be reviewed?

Why this belongs in the workflow.

Small tools are useful when they make a review step clearer, not when they promise to replace judgment. This utility keeps the work visible and processes the input locally in your browser.

Privacy note. The current version does not upload your text to a server. Avoid pasting confidential material into any browser tool unless you understand the device and browser you are using.

Continue with a small utility or a practical guide that supports the same step in your workflow.

GuideCreate a meeting agenda from notes

Turn confirmed decisions and open questions into a focused next-meeting plan.

GuideWrite a meeting follow-up email

Confirm owners and dates without turning an ambiguous remark into a commitment.

GuidePlan weekly priorities

Sort a crowded project list while keeping dependencies and trade-offs visible.

GuideCreate a decision log

Preserve options, evidence, owners, and unresolved questions for future readers.

How to use the AI prompt counter.

  1. Paste a draft prompt into the scratchpad.
  2. Compare words, characters, and the approximate token count.
  3. Use the result as a planning signal, then check the final prompt in the model you actually use.

Worked example: size a prompt before a code-review task

Scenario: You need to ask an AI to triage a batch of GitHub issues into a prioritized table for the on-call engineer. The raw notes are scattered across Slack threads, a Notion page, and a few issue comments.

Input:

Task: turn these support notes into a triage table Audience: the on-call engineer Constraint: mark unknowns instead of guessing Output: issue, evidence, next check Raw notes: [paste 800 words of messy notes here]

Counter result: ~950 words, ~2,400 chars, ~600 estimated tokens (GPT-4). The token estimate tells you the prompt fits comfortably in GPT-4's 8k context with room for the model's response.

Verify: the count helps compare versions, but the person sending the prompt still checks whether the source notes are complete, whether any confidential data slipped in, and whether the output shape matches what the engineer needs.

When to use this pattern: before any prompt that exceeds ~1,000 words, when you need to estimate API cost, or when you're comparing two prompt variants to see which is leaner.

When not to use it: for very short prompts under 200 words where the overhead isn't justified, or when the prompt contains no variable content (e.g., a fixed template you've already validated).

What this tool does not promise

It does not produce official token counts, guarantee a prompt will fit a model, or judge whether the prompt is factually correct. Tokenization and context limits vary by model and may change. The estimate uses a character-per-token heuristic (4 chars/token for GPT-4, 3.7 for Claude 3.5, 4.1 for Gemini 1.5) which is accurate to within ~15% for typical English prose but can drift for code-heavy or non-Latin text.

Important: All processing happens locally in this browser. Token counts are estimates, not official model billing measurements.

When to reach for this tool

  • Planning a long prompt: you have a multi-part task (context + constraints + output shape) and want to know the rough token budget before sending.
  • Comparing prompt variants: you're iterating on a prompt and need to see whether a rewrite actually reduced length.
  • Estimating API cost: the planning-rate cost estimate gives a ballpark for input tokens; multiply by your expected volume for a rough budget.
  • Checking context safety: the model-length progress bar warns when you're near or over a model's context limit.

When this tool is the wrong choice

  • You need exact token counts — use the provider's tokenizer (e.g., tiktoken for OpenAI models).
  • The prompt contains sensitive data — even local browser tools can be inspected by browser extensions or device management software.
  • You're prompting a model with a very different tokenizer (e.g., multilingual models with non-Latin scripts) where the 4-char heuristic is unreliable.

Continue the workflow

when the prompt still lacks a clear task, audience, constraint, or output shape.

when the counted prompt is preparing progress notes for another reader.

when you need to inspect headings, lists, or a handoff structure before sharing.

Frequently asked questions