# Tone of voice: make the bot write like you

> Set the tone Crutor's AI uses with candidates: formality, form of address, emoji, reply length, your own style rules, and learning from your chats.

The tone decides how the AI writes to your candidates: formal or casual, with or without emoji, short or a bit longer, and, in Russian and Ukrainian, whether it uses the informal or formal “you”. You set it with `/tone` in the bot. The four basic settings apply with a single tap; your own style rules are checked first, and the bot shows two sample replies before you save them. You can set one tone for all your jobs and a different one for a single job. There are six steps below, all optional: do only the ones you need.

## What the tone changes and what it doesn't

The tone changes only **how** the AI's messages sound: wording, warmth, length, emoji. It doesn't change **what** the AI asks a candidate, what it promises or how it scores them; that comes from the questions and the job text. The tone applies in the conversation with the candidate, in replies after the application is complete, and in the message drafts the AI writes for `/tell` in a candidate's topic.

## Steps

1. **Open the tone settings.** Send `/tone` to the bot (in the menu: “Tone of voice for the AI”). The bot sends the “🗣 Tone of voice” card: four settings, your style rules and buttons.
2. **Pick the basic settings.** Each button shows its current value, for example “Formality: neutral”. A tap moves to the next value; after the last one it goes back to “not set”. The change applies right away, with no check.

   - **Form of address**: informal (ты) or formal (вы). The AI uses that form of “you” in every reply where the language has the choice, even if earlier messages in the conversation used the other one.
   - **Formality**: formal (formal and polite), neutral (friendly and professional) or casual (casual and relaxed, like a colleague).
   - **Emoji**: none, a few (an emoji now and then is fine) or lots (emoji are welcome).
   - **Length**: short (1–2 sentences) or medium (up to 4 sentences).

   When a setting is not set, the AI uses its default. For the form of address that means it keeps the form it has already used in this conversation; there is no default ты or вы. In English the form of address makes no difference.
   ![/tone opens the tone card; each tap on a setting switches it to the next value at once.](https://crutor.app/scenes/tone-card.en.webp)

3. **Add your own style rules.** Tap **➕ Add a rule** and write one rule, for example “Be warm, but don't over-praise” or “Never use corporate jargon”. The bot answers “🔍 Checking the tone and writing two sample replies…”. It then sends the verdict and two samples: how the AI, in the new tone, answers “Hello! I'm writing about the “…” job. Is it still open?” and “What's the salary, and what are the working hours?”. If you like them, tap **💾 Save**; if not, **✖ Discard**. Candidates notice nothing until you save.
   ![A new style rule is checked first, with two sample replies; it applies only after 💾 Save.](https://crutor.app/scenes/tone-rule.en.webp)
4. **Or let the bot learn from your chats.** Tap **🎓 Learn from my chats** or send `/learn_tone`, forward messages from real conversations with candidates (yours and theirs, at least 8 of yours and no more than 600 at a time), then send `/done`. Messages the bot sent under your name with `/say` and `/tell` count as yours. Contacts and names are removed before the AI reads the text, and it learns only how you write, not what. The rules it finds go through the same check with samples, and you save them the same way. You can run the learning up to five times a day.
5. **Give one job its own tone, if needed.** On the job card, tap **🗣 How the AI writes to candidates**. Until the job has its own tone it uses your general one, and the bot marks the values “(as in general)”. Change a setting and the job gets its own tone: the job's settings override the general ones one by one, and the job's own rules, if it has any, replace the general rules. **Use your general tone for this job** puts everything back.
6. **Edit in the app.** **✏️ In the app** on the tone card opens “Settings → AI tone of voice” in the app. It has everything on one page: what it applies to (all jobs or one job), the basic settings (applied right away), style rules with an on/off switch for each, **Check and show sample replies**, an upload for chat exports from Telegram Desktop (result.json files), and the version history with **Restore**.

## Example

Say the AI asks a candidate, in a Russian conversation, «Сколько лет вы занимаетесь тестированием?» (formal “you”). Pick “Form of address: informal (ты)” and “Length: short”, and from then on it writes roughly «Сколько лет ты занимаешься тестированием?»: the informal “you” in every reply, even where it used the formal one earlier in the conversation, and no more than one or two sentences. The question stays the same; only the delivery changes. The exact wording depends on the conversation, so look at the samples the bot shows during the check.

## Rules that won't pass

A style rule is only about how messages sound. The check rejects a rule that:

- changes what the AI does rather than how it sounds;
- asks candidates for sensitive data or payment;
- sends candidates somewhere else (links, @mentions, phone numbers);
- treats candidates differently depending on who they are;
- misleads candidates.

The bot may rewrite an unclear rule so that it's only about style; the rewritten version is what gets saved. If a sample reply contains a link, a request for payment, a hiring decision or the wrong language, the tone fails too. When a rule is rejected, tap **Save only the style, without this rule** or **✏️ Write it again**. Limits: up to 15 rules and 1000 characters for all rules together.
