Get a second opinion: the audit
The audit asks two other AI models to read your rewrite and quote anything that still sounds generated.
Before you start
- Your provider must be OpenRouter. The audit picks its models from OpenRouter's catalogue, so it does not run with other providers.
- The audit runs on Rewrite and Recompose, not on Polish.
Turn on the audit
- Tick Audit with two other models below the Rewrite button.
- Select Rewrite.
When the rewrite finishes, the Audit panel on the right fills in. Each audit costs about two cents. PlainSpeak only picks models that cost at most $1 per million tokens of input.
How the verdict works
Three models vote on whether the rewrite reads clean. The model that wrote the rewrite counts as the first, because it was told to remove every tell. The two audit models are the other two.
| Verdict | What it means |
|---|---|
| PASS | 3 of 3 clean: neither audit model found anything. |
| WARN | 2 of 3 clean: one audit model flagged something. |
| FAIL | 1 of 3 clean: both audit models flagged something. |
Each audit model gets its own card in the panel. It shows CLEAN, or the phrases it quoted, up to five, each as an AI-ism, jargon, cliché or filler. The audit models judge wording only.
How the audit models are chosen
- Never from the same model family as the model that wrote the rewrite. A model is poor at spotting its own habits.
- Two different families, picked at random from a list including DeepSeek, Qwen, Google, OpenAI, Anthropic and Mistral.
- The same two models stay with a piece through every round, so a later round checks the same eyes' findings. New source text gets new models.
Act on what it found
| Button | What it does |
|---|---|
| Fix | Rewrites only the quoted phrases and leaves everything else alone. |
| Audit again | Asks the same two models to read the current copy again. |
| Audit with new models | Picks two different models from other families. |
After a fix, the panel says how many of the earlier findings are gone. A phrase flagged three rounds running is marked Author's call, and after three fix rounds a note suggests editing the rest by hand: what is left is often a matter of taste. When both audit models find nothing for two rounds in a row, the panel says the piece reads clean to them.
If the audit does not run
- Audit timed out. Your rewrite is saved. Select Audit again.
- This piece is too long to audit. None of the available models can read it in full. Shorten it, or keep the rewrite as it is.
- No audit after "The rewrite didn't land." The audit only reads rewrites that were kept.