> ## Documentation Index
> Fetch the complete documentation index at: https://docs.trydear.app/llms.txt
> Use this file to discover all available pages before exploring further.

# Voice blending: mix styles in AI suggestions

> Blend up to four named writing styles, including your own voice, by setting percentage weights that Dear applies to every AI suggestion.

Voice blending lets you shape the stylistic character of AI suggestions without writing a prompt. You assign a percentage weight to each active voice, and Dear combines them into a blend bar that it passes as a style instruction alongside every generation request.

The system is **on by default** and lives in the **Custom voices** section of AI preferences.

## The four built-in voices

| Voice                 | Style                                                            | Default weight | Default state |
| --------------------- | ---------------------------------------------------------------- | -------------- | ------------- |
| **My voice**          | Your own baseline — updated continuously from your writing       | 40%            | On            |
| **Essay — Didion**    | Spare, observational. Short declaratives. Personal but detached. | 39%            | On            |
| **Product memo**      | Direct, structured. Leads with the decision. No hedging.         | 21%            | On            |
| **Newsletter — warm** | Conversational, second-person. Warm openers, light asides.       | 0%             | Off           |

The defaults sum to 100% across the three active voices. The Newsletter voice ships off; toggling it on adds it to the blend and rebalances the weights.

## The blend bar

The blend bar at the top of the Custom voices panel is a horizontal bar divided into colored segments — one per active voice. Segment widths correspond to each voice's weight:

* **My voice** — dark neutral
* **Essay — Didion** — blue
* **Product memo** — green
* **Newsletter — warm** — purple

Inactive voices contribute no segment. The bar updates in real time as you drag sliders.

## Setting weights

Each voice has a **Weight** slider ranging from 0 to 100. Dragging one slider adjusts the others automatically so all active voices always sum to 100%.

The adjustment distributes the offset equally across the other active voices. If one voice would go below 0%, Dear clamps it there and redistributes the remaining difference across the others.

<Tip>
  You can also type directly in the percentage display next to each slider if you want a precise value.
</Tip>

### Normalize

The **Normalize** button distributes weight equally among all active voices. With three voices active, each gets approximately 33% (the last voice absorbs the rounding remainder).

Use Normalize to quickly reset to an even blend after experimenting.

## Toggling individual voices

Each voice card has an on/off toggle. Turning a voice off:

* Removes its segment from the blend bar.
* Sets its percentage to 0%.
* Rebalances the remaining active voices to sum to 100%.

Turning it back on re-adds it to the blend. Its weight is initialized from wherever the other voices redistribute.

<Warning>
  If you turn off all four voices simultaneously, all weights drop to 0% and Dear sends no blend instruction. At least one voice should be active for blending to have any effect.
</Warning>

## My voice and voice learning

**My voice** is anchored to your [voice learning](/ai/voice-learning) progress. The percentage learned from your human-written words (0–100%) appears on the My voice card alongside a note that it updates continuously. The higher your voice learning progress, the more accurately Dear can represent your baseline style in the blend.

## Creating custom voice presets

The **Create a new voice** form at the bottom of the Custom voices section lets you save a named style description that Dear echoes in suggestions alongside the built-in blend.

<Steps>
  <Step title="Enter a name">
    Give the voice a short, descriptive name. Example: "Technical blog" or "Cover letters".
  </Step>

  <Step title="Write a description">
    Describe the voice in plain language. Example: "Write in short sentences. Avoid adverbs. Lead with the thesis. Use plain words over jargon." The more specific you are, the more consistently Dear can apply it.
  </Step>

  <Step title="Save the voice">
    Click **Create voice**. Your preset appears in a **Your saved voices** list below the form.
  </Step>
</Steps>

Saved presets are applied in addition to the four built-in voices — Dear includes them as supplementary style references. You can delete a saved voice at any time by clicking the **×** button on its card.

<Note>
  Sample uploads for custom presets ("Add samples") are visible in the UI but not yet active. Support for training voices from your own writing samples is coming in a future update.
</Note>

## Enabling and disabling voice blending

The master **Custom voices** toggle at the top of the section turns the entire system on or off. When disabled, Dear sends no blend instruction and generates without a named style reference. Individual voice states and weights are preserved so re-enabling restores your previous configuration.

## Related

* [Voice learning](/ai/voice-learning) — how Dear builds your My voice baseline
* [Inline completions](/ai/inline-completions) — where voice blending is applied to Tab suggestions
* [AI guardrails](/ai/guardrails) — control unwanted patterns alongside voice settings