Skip to main content

ChatHarmoni Prompt Library - Coming Soon

Updated

Asking a question of ChatHarmoni is a great way to get to the heart of the story and produce the analyses you need. Writing a prompt to do this efficiently is important and as you use ChatHarmoni more, you'll become an expert. This article provides some tips and suggested prompts to get you started.

    Tags:
  • Creator
  • Explorer

 

IpsosIcon_LightBulb_WhiteTealBg_Rotate.svg

Coming Soon

 

In this article

 

1. Prompt Library

Included in this article are some suggested prompts to help you create and develop analyses. Keep in mind these general tips when you interact with ChatHarmoni.

  • Keep prompts within the scope of Harmoni's capabilities
  • Finite, targeted or controlled instructions with clear expectations typically lead to better outcomes
  • If you would like to change ChatHarmoni’s style, run a Quick Action prompt option first or Ask a question before defining stylistic preferences
  • Be patient, especially if the result isn’t what you wanted on the first try
  • Some things may take a while to process. Consider how long the manual process would take while you sit back and let ChatHarmoni do the work.

 

Simple Prompts

Here are some examples of some straight forward prompts. You can include more than one question, splitting out requests into different steps if required.

Cross [variable] with [variable] and filter by [variable]
Show me the trends of [variable] for the last 5 years
What are the top 5 [variable] across [variable] filtered by [variable] & [variable] as a stacked bar
Show me a ranked table of [variable] measured by [variable]
Discover [variable] [element] - This will use the Discover function, setting the named element as the target.
Analyze the relationship between x and y
Draw a plot chart of [variable] and [variable] for [variable]
Show me [variable] under [variable]. Using the words 'under' or 'over' creates a nested analysis. 

 

Comprehensive Prompts

You can write very comprehensive prompts that provide guidance to ChatHarmoni and help to shape the types of analyses included in your requests. Keep in mind that when you do this, it's better to refine your list and not include too many rules. LLMs can become confused when there are multiple considerations and steps to follow.

Here is an example of a comprehensive prompt to return 10 analyses.

A valuable analysis is one that highlights patterns and relationships in the data. It will contribute to an understanding of the real world and will help with strategic decision making.

Please find 10 valuable analyses from this project. Rank them in terms of value and display the highest ranked analysis. Explain the rationale for each analysis and also state the chart type and the dimensions used for each.

Follow all 25 of these rules:

  1. Assess what the main topics of the survey are and find analyses that highlight those main topics.
  2. Never have more than 10 elements in a graph legend and, unless it is a time series axis, never have more than 20 elements in the X-axis.
  3. Each analysis can include a filter but does not have to. When a filter is added, always display the analysis with only the first filter element selected.
  4. Never start with a time series analysis of a variable that does not have data for all time periods.
  5. If a tracker has fewer than three time period elements, display the data as a bar chart.
  6. When there is more than one axis representing a variable, always use the axis with the fewest elements. Favor ages axes with fewer elements. Favor summary axes of rating scale variables, for example, top and/or bottom box elements.
  7. Only put time series axes in the across drop zone.
  8. In time series analyses, only use variables for the down axis that have data for the majority of the periods.
  9. Always put demographic variables in the down axis to show the demographic profile of key variables.
  10. Never include an analysis of one demographic crossed by another unless you also add a measure or a filter.
  11. If the number of bars in a Bar Graph would be > 70, display the analysis as a table.
  12. If the analysis is a time series, always use a line graph.
  13. When you display a bar chart or a stacked bar chart where the elements of a down axis are not sequential, rank the chart on the first element in the legend. When the down axis elements are sequential (income groups or age groups for example), do not rank on the first element of the legend axis.
  14. Never include the Total element in a graph.
  15. Always have only one variable in each dimension. Never create a group.

 

Quality Assurance Prompts

When using ChatHarmoni to interpret data, either in the Analysis area or in a Story View, you can use the following prompt or something similar to check an analysis.

Prove to me that the following is accurate: <include text that ChatHarmoni shared with you>

 

Verbatim Analysis Prompts

After creating a verbatim analysis in Harmoni, you can use #interpret to ask ChatHarmoni to analyze the open-ended comments. You can also ask your own questions. Here are a few examples of prompts that may be useful.

Perform a thematic coding of this data. Group the responses into themes and sub themes. Summarize the key details of each theme.
Perform a sentiment analysis.

 

Where to from here?

Learn more about Harmoni:

Up Arrow.png

Related articles