Skip to main content

Topic Prompt Optimizations [Beta]

Topic Prompt Optimizations helps you answer the question “Do I have the right prompts for this topic—or too many?” And gives you a clear, data-backed way to fix it.

Updated this week

What is this?

Topic Prompt Optimizations helps you find the right number and mix of prompts for a topic—so you maximize coverage without unnecessary overlap.

It analyzes which prompts contribute unique value vs. which are redundant, then recommends an optimal set to keep.

Note: this feature is currently in private beta. If you are interested in learning more, please reach out to support or your customer success manager.

How it works (at a glance)

  1. Overshoot
    Start with more prompts than you think you need.

  2. Patience
    Ensure prompts have been running for at least two weeks before using the tool.

  3. Optimize
    Archive prompts that don’t add meaningful new coverage in a topic.

Key concepts

  • Coverage
    % of unique URLs cited by at least one prompt
    → Measures breadth

  • Resilience
    % of URLs cited by two or more prompts
    → Measures redundancy / consistency

Anticipated changes in Presence, Sentiment, and Position are also considered when recommending prompts that can safely be archived.

Getting started

1. Select a topic

Choose a topic and platform(s) to analyze. The system uses recent data (last ~14 days) to evaluate prompt performance and by default will run across all active models.

2. Run optimization

Click Run Optimization to generate results.

3. Understanding your results

Summary metrics

  • Prompts → # of prompts remaining in your topic

  • Coverage % → how much of the URL space you capture

  • Resilience % → how well URLs are backed up

4. Archive Prompts

Optionally, archive suggested prompts to cut.

System recommendations

You may see messages like:

  • ✅ “Your prompt setup is already efficient”
    → Every prompt adds value

  • ⚠️ “Pruning not recommended”
    → Removing prompts would reduce coverage

Exporting results

Click Export CSV to download:

  • Which prompts to keep vs. cut

  • Closest replacement prompts for anything removed

  • The configuration used to generate the result

Best practices

  • Start with more prompts than needed, then optimize down

  • Use Coverage 50 / Resilience 50 if unsure

  • Watch for a flattening coverage curve before pruning

  • Don’t prune aggressively if every prompt adds unique URLs

For the data nerds among us!

Optional Model Configurations

Constraints

  • Coverage Floor
    Keep at least X% of unique URLs after pruning

  • Resilience Floor
    Preserve redundancy across key URLs

  • Budget
    Set a max number of prompts to keep

Pruning Strategies

  • Protect Coverage
    Prioritize unique URL discovery

  • Protect Resilience
    Prioritize backup coverage

  • Coverage / Resilience blends (70/30, 50/50, 30/70)
    Balance both (50/50 is a strong default)

  • Least Connected First
    Removes prompts that cite very few unique URLs first

  • Most Redundant First
    Removes prompts whose URLs are all heavily covered by other prompts

Charts

Greedy Coverage Curve

  • Shows how many new URLs each prompt adds

  • If the curve flattens → you’re hitting diminishing returns

Pareto Frontier

  • Shows the best possible tradeoffs between coverage and resilience

  • Points on the line = optimal choices

  • Faded points = suboptimal (avoid)

Retention Charts

  • Show how coverage and resilience change as prompts are removed

  • Helps you pick the right balance for your goals

Did this answer your question?