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

# Training

> Feed Beat examples of your writing so it sounds like you.

Training is the highest-impact thing you can do to improve Beat's output quality. Voice descriptors and tone guidelines tell Beat how you sound in the abstract. Training samples show Beat exactly how you sound in practice — and that difference is significant.

## Why training matters

AI models are pattern matchers. When Beat has real examples of your writing, it can match your rhythm, sentence length, transition style, how you open paragraphs, how aggressive your CTAs are, and dozens of other micro-patterns that make your voice recognizable. Without training samples, Beat defaults to polished-but-generic. With 2–3 strong samples, the difference in output quality is immediate.

## The three weights

When you add a training sample, you assign one of three weights:

| Option                                      | Badge shown in list | What it means                                                                                  |
| ------------------------------------------- | ------------------- | ---------------------------------------------------------------------------------------------- |
| **This is exactly how I want to sound**     | On voice            | Your voice at its best. Beat actively learns from and emulates this style.                     |
| **This is okay but not ideal**              | Okay                | Acceptable but not representative. Beat understands you write this way but won't lean into it. |
| **This is what I DON'T want to sound like** | Anti                | What you want Beat to actively avoid.                                                          |

You also tag each sample by **content type**: Blog post, Email, Bio, About page, Social post, Sales copy, or Other. The tag helps Beat apply the sample's voice patterns in matching contexts.

## How to add a training sample

<Steps>
  <Step title="Go to the Training tab">
    Navigate to `/app/brand` and click the **Training** tab.
  </Step>

  <Step title="Click Add sample">
    Click the **+ Add sample** button to open the sample input.
  </Step>

  <Step title="Paste your content">
    Paste a piece of your writing. A full blog post, a LinkedIn article, an email newsletter, or a long-form social post all work well. The sample should be long enough (200+ words minimum) to show Beat your natural patterns.
  </Step>

  <Step title="Tag the content type and pick a weight">
    Choose the content type (Blog post, Email, Bio, About page, Social post, Sales copy, Other) and the weight (**This is exactly how I want to sound**, **This is okay but not ideal**, or **This is what I DON'T want to sound like**).
  </Step>

  <Step title="Click Add sample to save">
    The sample is saved immediately and Beat begins using it for your next generation.
  </Step>
</Steps>

## What types of content to use

The best training samples are real pieces you've already published. Good options:

* A blog post or article you're proud of
* A LinkedIn article that got strong engagement
* An email newsletter that felt authentically you
* A long-form social post with your signature voice

Avoid using: pieces that were heavily edited by someone else, content you wrote in a rush, or anything that doesn't reflect how you'd write today.

## How many samples you need

You only need **2–3 "exactly how I want to sound" samples** to start seeing a meaningful improvement. You don't need to upload your entire content archive.

* **2–3 samples** — Beat gets a baseline understanding of your voice
* **5–10 samples** — Beat's voice accuracy becomes noticeably more consistent
* **10+ samples** — Best results, especially if your content spans multiple formats

## When to add more

Add new training samples when:

* Beat's output starts drifting from your style (common after a few weeks of heavy use)
* You've evolved your voice or brand positioning significantly
* Beat keeps producing a pattern you don't like, even after tweaking your voice descriptors

## Anti-reference examples

Anti-reference samples are just as useful as positive examples. Use them for:

* **Generic AI output** — paste a piece of AI-generated content that sounds robotic or over-polished. This trains Beat on what to avoid.
* **Competitor style you dislike** — if there's a writing style in your industry that feels off-brand for you, add it as an anti-reference.
* **Your own weak content** — a piece you wrote when you were trying too hard, being too formal, or using jargon you've since dropped.
* **Passive voice or hedged language** — if you want Beat to be more direct, paste examples of overly cautious writing as anti-references.

<Tip>
  If Beat keeps using a phrase or structure that feels wrong, find a good example of that pattern and add it as an anti-reference. This is often faster than tweaking voice descriptors.
</Tip>

## What's next?

<CardGroup cols={2}>
  <Card title="Brand Beat Overview" icon="id-card" href="/brand/overview">
    Learn how to maintain and update your full brand profile over time.
  </Card>

  <Card title="Beat AI Overview" icon="robot" href="/beat/overview">
    Understand how Beat uses your brand profile across every page.
  </Card>
</CardGroup>