Skip to main content
Solved

A favorite Mira studio Prompt that no longer works

  • January 20, 2026
  • 2 Replies
  • 53 views
A
Explorer
Forum|alt.badge.img

I learned this prompt on one of the AI challenges last year and loved it! However, it no longer works as it used to :( Any suggestions on how to get it to work is it used to?

The super prompt: Using #[a saved Explore Search that covers a year of data across social data] identify the top 5 trending social media conversations from a year ago, the top 5 trending social media conversations from 6 months ago, and the top 5 trending social media conversations from the past two weeks. Include 3 linked results documents (with the highest reach) for each. Only use each document as one example.

Response I get: No distinct documents from six months ago were returned in the current results. The above conversations may have persisted, but there is no specific evidence from that period. This answer is based on available data only.

Best answer by kelly.bebenek

​@anamaracha I spoke with the Mira Product Team ( Thank you ​@Giovanni ).

 

This prompt isn’t very reliable right now. In practice, it sometimes works and sometimes doesn’t, especially when it’s asking the AI to analyze the same saved search across multiple time ranges in a single request. What’s happening behind the scenes is that the agent doesn’t always run three distinct analyses. In many cases, it runs one broader pass and then tries to infer the rest. When that happens, certain periods, like six months ago or the last two weeks, may not be sampled clearly, which leads to gaps in the results.

 

How to get more consistent results today

We are improving how Mira deal with more complex prompts like this one, and your feedback helps us prioritize fixes that matter in real workflows. Keep the examples coming, they’re incredibly useful. 

 

In the meantime, these two approaches tend to work best:

 

Option 1: Split the prompt by time period
Run the same prompt three times, each with a single, explicit timeframe:

  • One year ago
  • Six months ago
  • Past two weeks

This helps ensure the AI actually analyzes each period separately.

 

Option 2: Narrow before expanding
Start with the shortest time range first (for example, the past two weeks) and confirm you’re getting distinct documents. Then repeat for the older periods. This reduces the chance of the model defaulting to one general data sample.

 

      2 replies

      kelly.bebenek
      mEmployee
      Forum|alt.badge.img+10

      ​@anamaracha Let me speak to that product team to see what’s going on.  Thanks for flagging.

       

      Sr. Director of Customer Ed @ Meltwater (DM me to discuss: Academy, Community & Help Center)
        kelly.bebenek
        mEmployee
        Forum|alt.badge.img+10
        • kelly.bebenekmEmployee
        • mEmployee
        • Answer
        • January 21, 2026

        ​@anamaracha I spoke with the Mira Product Team ( Thank you ​@Giovanni ).

         

        This prompt isn’t very reliable right now. In practice, it sometimes works and sometimes doesn’t, especially when it’s asking the AI to analyze the same saved search across multiple time ranges in a single request. What’s happening behind the scenes is that the agent doesn’t always run three distinct analyses. In many cases, it runs one broader pass and then tries to infer the rest. When that happens, certain periods, like six months ago or the last two weeks, may not be sampled clearly, which leads to gaps in the results.

         

        How to get more consistent results today

        We are improving how Mira deal with more complex prompts like this one, and your feedback helps us prioritize fixes that matter in real workflows. Keep the examples coming, they’re incredibly useful. 

         

        In the meantime, these two approaches tend to work best:

         

        Option 1: Split the prompt by time period
        Run the same prompt three times, each with a single, explicit timeframe:

        • One year ago
        • Six months ago
        • Past two weeks

        This helps ensure the AI actually analyzes each period separately.

         

        Option 2: Narrow before expanding
        Start with the shortest time range first (for example, the past two weeks) and confirm you’re getting distinct documents. Then repeat for the older periods. This reduces the chance of the model defaulting to one general data sample.

         

        Sr. Director of Customer Ed @ Meltwater (DM me to discuss: Academy, Community & Help Center)
          Terms & ConditionsCookie settingsAccessibility statement