Prompt Diff & Comparer

Overview

Compare two versions of a prompt line by line and see exactly what changed, with added, removed and unchanged line counts plus a token delta. When you are iterating on a system prompt and results shifted, the first question is what actually changed - this answers it in one paste, entirely in your browser.

How It Works

Paste the current version into the Version A field and the candidate into Version B. A summary line shows plus-N added, minus-N removed, unchanged lines, and the token delta with its sign. Below it, the unified-style output marks each line: plus for added, minus for removed, plain for unchanged. Both counts update as you edit either side. Nothing is uploaded, so this is safe for proprietary system prompts.

Step-by-Step Usage Guide

1. Paste the current prompt into Version A.
2. Paste the candidate version into Version B.
3. Read the summary: added, removed, unchanged and token delta.
4. Inspect the marked lines to see the change block by block.

Technical Specifications & Standards

The comparison is positional: line N of version A is compared with line N of version B, and a mismatch is emitted as a remove of A's line followed by an add of B's line. That is a deliberate trade-off against a full LCS (longest-common-subsequence) diff - it cannot detect a pure insertion shifting later lines down, so it will report those lines as changed rather than as one insertion. For prompts, which are usually edited block by block and line-aligned, the positional view is often clearer than a minimal diff that interleaves plus and minus in confusing ways. The token delta is computed by encoding both full prompts with the in-browser GPT byte-pair tokenizer and subtracting, so it counts real tokens, not characters - a change from a long word to a short symbol can show a negative delta even when the line count grew. Line splitting accepts CRLF and LF endings, so copy-paste between Windows and Mac editors does not produce phantom changes.

Targeted Use Cases

• Reviewing a teammate's edit to a shared system prompt before merging.
• Understanding why a prompt tweak changed model behaviour by seeing the exact lines.
• Tracking token budget drift across prompt iterations.
• Documenting a prompt change in a pull request description.

Notes & Gotchas

• Keep both versions line-aligned where possible; the diff is positional, not a minimal edit script.
• Watch the token delta when close to a context limit - small wording changes can cost hundreds of tokens.
• Paste from a plain source; rich-text copies can carry invisible characters that show as changes.
• Use the unchanged-line count to confirm the two versions really differ only where you intended.

Frequently Asked Questions