Prompt Compare

Prompt Compare takes two prompt texts — Version A and Version B — and runs a deterministic Myers diff algorithm to compute every added, removed, and unchanged line between them. The result is displayed in a side-by-side split view or an inline diff view with line numbers, word-level highlights on changed lines, and a summary of added/removed/unchanged counts.

No. Comparison runs locally in this browser. Your prompts are not uploaded or sent to a model. The diff algorithm runs entirely in vanilla JavaScript with no network calls, no API access, and no analytics on your input text. You can verify this by opening browser DevTools → Network while using the tool — you will see zero outbound requests related to your content.

Prompt Compare uses the Myers diff algorithm — the same algorithm used by Git, GitHub, and most professional code-review tools. It computes the minimum edit distance between the two inputs and produces a stable, deterministic output for any given pair of texts. The result is the same every time you compare the same two texts with the same options.

Side-by-side (split) view shows Version A and Version B in two columns. Removed lines appear in the left column only with a − marker; added lines appear in the right column only with a + marker; unchanged lines appear in both columns. Inline view shows all lines in a single column with colour background and +/− symbols, similar to a traditional unified diff.

When Version A line 5 and Version B line 5 differ, Prompt Compare shows both lines and highlights the specific words that changed within each line using a word-level diff algorithm. Words removed from Version A are marked with a red background; words added in Version B are marked with a green background. This makes it easy to see that, for example, "summarize" changed to "briefly summarize" rather than showing the entire line as changed.

Ignore case treats uppercase and lowercase letters as identical when deciding whether two lines match. A line that says "You are an expert" in Version A and "you are an expert" in Version B will be considered equal. Ignore whitespace trims leading and trailing spaces and tabs from each line and collapses internal whitespace runs before comparing — useful when one version was pasted from a different editor that added extra indentation. When either option is active, a "normalized comparison" label appears in the diff summary to indicate that comparison keys were normalized even though your original inputs are not mutated.

When enabled, long runs of unchanged lines are folded down to a single row showing "… N unchanged lines …" with only 3 context lines shown above and below each changed region. This keeps the diff output compact when you are comparing two prompts that share most of their text.

Type in the search box on the status row to highlight every occurrence of your search term across both the split view and the inline view. Matches are highlighted with a yellow background. The search uses a literal text match with regex special characters escaped, so you can safely search for brackets, dots, and other punctuation found in prompts.

Yes. Each input pane has a File button that opens your local file picker. You can load any text file — .txt, .md, .json, .yaml, .html, or similar — directly into Version A or Version B. The file is read in the browser using the FileReader API and never uploaded anywhere.

Use the Download .diff button to save a standard unified diff file (prompt-compare.diff) that can be shared, stored, or applied with git apply. Use the Download Report button to save a plain-text report that includes both input statistics, the diff summary, and the full unified diff in a single file named prompt-compare-report.txt.

Yes. Each input pane has a Copy A or Copy B button that copies that version to the clipboard. The toolbar has a Copy unified diff button that copies the unified diff text. All copy operations use the Clipboard API with a textarea fallback.

Clicking the Swap button exchanges the contents of the Version A and Version B editors and immediately re-runs the comparison. After a swap, lines that were additions become deletions and vice versa. This is useful when you want to see the diff from the opposite direction.

Prompt Compare is designed for reviewing AI prompt iterations — it uses the same diff engine but opens with prompt-specific sample pairs, labels inputs as Version A and Version B rather than Text A and Text B, and removes code syntax highlighting which is not relevant for prompt text. Text Compare is designed for comparing general text documents, code files, and configuration and includes syntax highlighting for JSON, YAML, JavaScript, Python, HTML, CSS, SQL, and Bash.

Prompt Compare shows you what changed between two existing prompt texts. It does not evaluate, score, or improve either text. Prompt Optimizer analyzes a single prompt and applies rule-based transformations (whitespace normalization, section detection, duplicate removal) to produce a cleaner version of that one prompt. If you want to review the changes the Prompt Optimizer made, you could paste the original into Version A and the optimized version into Version B in Prompt Compare.

Prompt Compare is a focused two-input diff tool for comparing exactly two texts in a single session. It does not store versions. Prompt Version Manager stores labelled revisions of a prompt across a session and can generate diffs between any two stored versions. If you want to track more than two versions or label each revision, use Prompt Version Manager.

No. Prompt Compare is a text diff tool. It shows you what is added, removed, or unchanged between two prompt texts. It does not score prompts, predict model performance, or claim that a prompt with more words is better, more effective, or more secure than one with fewer words. Evaluating prompt quality requires model testing outside this tool.

Yes. The diff algorithm treats each line as a Unicode string. Line-level comparison, word-level highlighting, and character statistics all work correctly with non-Latin scripts including right-to-left text, CJK characters, and emoji. Word tokenization uses whitespace boundaries, so scripts that do not use spaces between words may show word-level highlights at the whole-line level rather than within individual tokens.

Template variables are treated as ordinary text. The diff will show if a variable name changed (for example {{topic}} to {{subject}}), if a variable was added or removed, or if surrounding text changed. Variables are displayed literally and safely escaped so they cannot inject HTML or execute JavaScript.

Each input side has a hard limit of 500,000 characters (about 100,000 words). Inputs above this limit are rejected with a status message. For inputs below this limit but above 5,000 lines, a performance warning is shown. The Myers diff algorithm has O(ND) complexity — for very large, highly divergent inputs the diff may take a few seconds. Inputs of typical prompt length (under 5,000 characters each) compare in under 10 milliseconds.

No. The Recent comparisons panel at the bottom of the tool stores the last 8 comparisons in memory only — the list is cleared when you close or reload the tab. No comparison data is written to localStorage, sessionStorage, IndexedDB, cookies, or any server.

The ignore-case and ignore-whitespace normalization is applied only to the comparison key used to decide whether two lines match — it is never applied to the displayed text. If you have a JSON snippet in your prompt and enable ignore-whitespace, lines like " "key": "value"" and ""key": "value"" will be treated as equal in the diff, but the original indented text is shown in the output unchanged.

Yes. A common workflow is to paste the previous version of a prompt into Version A and the revised version into Version B, review the diff to confirm the changes, then click Download .diff to save a standard diff file that can be attached to a ticket, pull request, or design document. For storing multiple labelled revisions in session, use Prompt Version Manager.