Help Center
Go back to application
Help Center
Go back to application
1. Getting Started
2. Agents
3. Data Streams
4. Branding and Voice
5. Sites
6. My Content
7. Integrations
8. Subscription and Billing
9. Prompt Studio & Expert Tips
10. FAQs & Troubleshooting
My Agent Didn’t RunCan’t Find Your Brand VoiceMy Article Looks Off – How to ImproveAgent Output Sounds Too AI — What Now?Why Isn’t My Image Generating?Image Upload ProblemsCommon Errors Explained
11. Product Updates
12. Microdocs & Reference

My Article Looks Off – How to Improve

What’s going on

The article generated by your Agent technically ran — but the output doesn’t feel right. Maybe it sounds generic. Maybe the tone doesn’t match your brand. Maybe the structure is flat, or the formatting isn’t what you expected.

This doesn’t mean anything is broken. But it does mean something upstream can be improved — and it’s usually easy to fix.

What to check

Start here:

  • Is the Agent designed for this content type?
    Using a listicle generator for a story-driven article (or vice versa) may produce awkward results.
  • Is the Brand Voice well defined?
    A short or vague Brand Voice makes it hard for the AI to match tone. The system needs clear direction.
  • Was the temperature set too high?
    A setting of 0.9–1.0 may result in looser structure or more variation between runs.
  • Are you using a lightweight model?
    Models like O1 Mini or Claude Haiku are fast but can lose detail or nuance.
  • Was a Datastream involved?
    If yes, low-quality or irrelevant source material can affect the writing quality.

How to fix it

  1. Rerun the Agent
    Sometimes a second pass with the same setup produces stronger results. That’s the nature of LLMs — they vary slightly with every run.
  2. Check the Brand Voice tone
    Go to Branding → Brand Voices. Make sure your tone, personality, and structure guidelines are specific and written in complete sentences. Add examples if needed.
  3. Lower the temperature
    Set it to 0.6–0.7 for cleaner structure and more reliable formatting.
  4. Upgrade the model
    If you’re using a fast/low-cost model (like O1 or Nano), switch to GPT-4.1 or Claude 3.7 for higher editorial quality.
  5. Review the Agent type
    If you're forcing a non-list format into a Listicle Generator (or vice versa), try a different Agent that fits your intended output.
  6. Use Prompt Studio for testing
    If the issue is persistent, run a quick test in Prompt Studio to isolate whether it’s the prompt, the voice, or the source material that’s affecting the tone.

If you're still not satisfied

Send us:

  • The Agent name
  • A short note on what feels “off”
  • An example of your ideal tone or structure (if available)

We’ll help review your configuration and recommend edits to voice, Agent type, or prompt that bring the output closer to what you want.

Powered by Product Fruits