Skip to main content

Tips & Tricks

Maximize the quality and consistency of your AI responses with these best practices for Response Settings.

Getting the best out of Rewo is often a careful and gradual process. While the AI is designed to hit the ground running, it can sometimes take a bit of fine-tuning to capture your unique brand voice exactly the way you want it. The following strategies have proven useful for teams seeking to achieve the highest level of professional consistency.

1. Start Simple, Then Refine​

Don't try to perfect everything at once. Configure the basics:

  1. Fill out Brand Voice and General Context
  2. Set Personality sliders to match your gut feeling
  3. Generate a few responses and see how they feel
  4. Adjust one or two settings at a time based on results

2. Test Your Settings​

Generate responses for a variety of review types to make sure your settings work well across different scenarios, for example:

  • Positive reviews (5 stars)
  • Negative reviews (1-2 stars)
  • Questions about features
  • Bug reports
  • Feature requests

3. Document Your Decisions​

Keep notes on why you chose certain settings. This helps when training new team members or revisiting settings months later.

4. Update Regularly​

Your app evolves. Update these settings when:

  • You add new features
  • You fix major bugs
  • Your brand voice changes
  • You notice the AI missing important context

5. Understand AI Limitations​

Managing expectations is key to a successful AI workflow. Language models have inherent behaviors that can affect response quality:

  • Hallucinations: AI models can sometimes generate incorrect information or confidently state things that are simply not true.
  • Parroting: The AI may echo specific parts of the user's review in its response. While the reply can still be perfectly helpful, this behavior can occasionally make the output feel repetitive if the AI mirrors the reviewer too closely.
  • Recency Bias & Instruction Drift: The model might over-prioritize newer instructions or lose track of older ones over long prompts.
  • Few-Shot Poisoning: A few poor historical examples or inconsistent settings can "poison" the AI's understanding of your desired style.
  • Non-determinism: AI responses are probabilistic; you may get slightly different replies even with the exact same input and settings.

We've implemented numerous measures to mitigate these issues, however, AI behavior will never be 100% perfect. You can take several steps to manage these risks yourself and ensure the highest quality for your users:

  • Utilize Auto-Scheduling Lead Time: Rewo is built with auto-scheduling and configurable lead times specifically to give you a buffer. Use this window to review drafts and catch any hallucinations or awkward phrasing before they are published.
  • Avoid Conflicting Instructions: Be meticulous when configuring your Personality sliders and Custom Prompts. Providing contradictory instructions—such as asking for extreme formality alongside deep conciliation—can lead to instruction drift and unpredictable behavior.
  • Maintain a Clean History: Since the AI learns from your experience, your published history is its primary textbook. Regularly audit your responses and adjust the Experience Influence if you feel the AI is starting to mirror outdated or repetitive patterns.

We are continuously refining Rewo to minimize inherent AI limitations, but we know that technology alone has its bounds. If you feel you are battling persistent inconsistencies or hitting a deeper issue that doesn’t seem to resolve through settings alone, please reach out to us. We take these cases seriously to improve our core logic and help you achieve the results you expect.