Passive voice creeps into LLM output constantly — “the report was generated by the system” instead of “the system generated the report.” It is wordier and vaguer about who did what. This tool flags likely passive-voice sentences with a fast local heuristic, and — with your own OpenAI or Anthropic key — rewrites the flagged ones in active voice while keeping the meaning intact.
How it works
Detection is local and free: the tool scans each sentence for a form of to be (is, are, was, were, been, being) followed by a past participle, optionally with a by agent, and highlights matches with a count. No key is needed for this step. If you want rewrites, add your API key and run the rewrite — the tool sends one direct request from your browser asking the model to convert only the flagged sentences to active voice and return the full revised text.
Your key never reaches a Gera server — it is held only in the tab and sent straight to the provider (with the official direct-browser-access header for Anthropic). Refreshing clears it.
Passive voice isn’t always wrong
The detector finds candidates, not crimes. Passive voice is the right choice when the actor is unknown, unimportant, or intentionally downplayed, and it is standard in much scientific and formal writing. Use the flags as prompts for judgement, not a mandate to rewrite everything.
Tips
- Detect first, read the highlights, and only rewrite when the active version is genuinely clearer.
- Cheaper models (gpt-4o-mini, claude-3-5-haiku) handle straightforward rewrites well and minimise cost.
- After rewriting, re-run detection to confirm the passive count actually dropped.