All Collections
Storytelling
Write great prompts for CoWrite
Write great prompts for CoWrite
Jennifer Funk avatar
Written by Jennifer Funk
Updated over a week ago

CoWrite is an AI data storytelling assistant that helps you overcome the dreaded blank page syndrome by quickly generating a first draft to get you going. Like any assistant, CoWrite performs better with clear and focused instructions. So, here are a few things to keep in mind as you experiment with new prompts.


Have a clear objective.

Before creating a prompt, decide what kind of story or storytelling device you need. Is it a headline, a key takeaway, or maybe a strategy for addressing the problem the data illustrates? A clear objective will allow you to provide more specific directions in your prompt and will produce a better result.

Example:

Prompt with a vague objective: Write a paragraph about what’s happening here

Prompt with a clear objective: Suggest three short headlines

If you’re having trouble deciding what you need, go back to the basics: consider your audience. Policymakers often seek actionable insights, so a headline or key takeaway is more likely to support their needs, while the general public craves understanding and relevance and may benefit from additional context to help them understand why the data matters. Let your audience be your guide, and use that to guide CoWrite.


Identify an angle.

Every data point tells a story, and every story has multiple angles. Whether it’s a headline, an explanation of a trend, or a persuasive call to action, telling CoWrite what to emphasize will yield a more compelling result.

The “Barriers” suggested prompt is an example of this. A common and effective way to frame stories about health and wealth inequities is to highlight the inequity itself and provide context about the structural and systemic barriers that cause the inequity.

Example:

Discuss the various systemic barriers that may create geographic inequities.

To determine your angle, one useful technique is to consider the larger strategic outcome your organization is working toward. For example, you might tell CoWrite to focus on the relationship between an indicator like “Occupied Housing Units Without A Vehicle” and an outcome like “Reducing food insecurity.”

Example:

Describe the relationship between lack of transportation and food insecurity.

Or, if the data visualization shows a trend with a clear inflection point, you might prompt CoWrite to focus on the inflection point, and you can add relevant local context.

Example:

Summarize the trend that started in 2014 with raising awareness about the harms of overprescribing.


Be specific.

Set clear expectations for tone, format, and/or length. mySidewalk has already provided baseline guidance to CoWrite in each of these areas, but reminders don’t hurt. CoWrite is especially responsive to style instructions, as long as the style is relatively distinct and familiar. (For an exaggerated demonstration, try telling it to write in a pirate's voice.)

It’s also responsive to length instructions, though there is some nuance here. As an AI feature, CoWrite conceptualizes the world in tokens rather than words and it’s not great at reasoning. As a result, it will not count words or sentences, but it will be directionally accurate. An instruction to write 50 words will seldom be exactly 50, but it will keep CoWrite on the pithier side (for an AI).

mySidewalk has given CoWrite a set of baseline instructions, or things it should try to do in every output - things like use plain and inclusive language. Learn more about what we're trying to achieve with these instructions here. Even with these instructions in place, you have a lot of control over the end result through your own prompting, which is why we encourage you to experiment!


Set boundaries.

While generative AI generally performs better with positive instruction (don’t we all?), we’ve found it helpful to strike a balance between “dos” and “don’ts". For example, in some of our suggested prompts, we tell CoWrite what to focus on (à la tip 2) and what to avoid (e.g. repeating data values). Depending on your need, you might use the inverse instruction, “List the relevant data values using bullets. Don’t provide an explanation.” While results can be a little bit hit-and-miss with this tip, it’s worth testing out, particularly if you notice the outputs are failing repeatedly in a specific way.


Encourage creativity (within limits).

Idea generation is one of generative AI’s strengths. So, while guiding CoWrite is crucial, allowing some creative freedom can lead to unexpected and valuable results. The suggested "strategies" prompt is an example: given a topic and relevant data, simply asking it to share “community-based strategies” allows the AI to leverage its vast library of knowledge to return a variety of ideas.

Pro Tip

CoWrite is single-pass—one input, one output—but you can always string together multiple outputs to tell bigger stories. For example, you might generate three outputs: 1) a key takeaway, 2) barriers to better outcomes, and 3) community-based strategies for addressing the barriers.


Crafting a prompt is an art, balancing clarity with creativity. As you experiment with different prompts, we encourage you to use the suggested prompts as a starting point, tweaking and tailoring them to fit your unique needs, and then doing your own thing. Every data visualization holds a story; with the right prompt, CoWrite will help you tell it. Let us know what you discover.

Did this answer your question?