CoWrite is a Chart feature available to all Chart subscribers. If you have access to reports and/or dashboards, you are a Chart subscriber.
What is CoWrite?
CoWrite — mySidewalk’s generative AI writing assistant — generates text to help explain your data visualizations in clear and engaging language. CoWrite can accelerate and simplify your writing process, helping you quickly produce high-quality data stories. It works on visualizations powered by mySidewalk data as well as user-uploaded data.
In this article: |
How to use CoWrite
Currently, CoWrite is available in Chart for all visualizations
except for correlations and custom charts.
We are in the process of continuing to optimize CoWrite outputs for maps. In the coming days and weeks, expect map outputs to improve, including the ability to reference normalized data.
CoWrite is designed to be intuitive and user-friendly, so you can easily incorporate it into your data storytelling workflow. Use CoWrite by hovering over a data visualization in a Report or Dashboard page and clicking the CoWrite icon. You can use CoWrite with visualizations built using mySidewalk's data and the data you've uploaded.
A form with suggested prompts will appear. Click on any of the suggested prompts to apply that prompt, click the blue edit icon next to a suggested prompt to edit the prompt, or click "Write a Custom Prompt" to write your own prompt.
Submitting a prompt will generate text in a new block above the visualization. You can then edit the text until you're happy with the results. No matter what, always review the text for accuracy and language preferences before publishing it since generative AI will sometimes produce less-than-optimal results.
Pro Tip: As you're iterating toward a satisfying result, you can delete the text block and start over with CoWrite, or keep part of the original result and have CoWrite generate additional copy to help you keep drafting.
How to submit feedback
Please give us feedback on CoWrite using the link, which will appear at the end of each newly generated text block. You can also share your feedback with your mySidewalk representative on how this tool might better meet your needs.
Remember that CoWrite is an experimental feature in active development. It marks the beginning of our commitment to leveraging advancements in AI to help our customers find, visualize, and communicate data. For this reason, your feedback is essential and will help shape how we evolve this and future tools.
Capabilities, use cases, and limitations
CoWrite can help you create text that frames, contextualizes, and highlights key insights in your data visualizations. It’s designed to quickly produce a first draft of text to review and refine before sharing with external audiences.
In particular, CoWrite can assist with the following tasks:
Preparing a presentation to policymakers
Creating reports on high-interest topics
Apply for funding
Reporting progress on strategic initiatives
Engaging followers on social media
While we plan to expand CoWrite’s capabilities over time, it is not currently well-suited for the following tasks:
Doing the math to find insights in your visualization
Generating alternative text for site accessibility
Completing outputs in other languages
Answering questions about a data finding
Additionally, it's important to understand that while mySidewalk employs various prompt engineering techniques and tools to meet quality standards on readability, the use of plain and inclusive language, and factuality — at this time, CoWrite (like generative AI generally) will perform better on some measures than others. For example:
CoWrite is directionally good at responding to length-related instructions, but it will rarely hit specific word or sentence count targets. There are a couple of reasons for this. First, AI conceptualizes the world in tokens rather than words or sentences. In other words, its unit of measurement is different from ours. Second, generative AI is notoriously verbose and is prone to overusing certain words and phrases. Often, you'll find that simply editing outputs for brevity and repetitiveness will take the writing from a 6/10 to an 8/10.
CoWrite struggles to complete outputs at specific reading levels. While we've made some progress on this metric via various prompt engineering techniques, CoWrite frequently underrates reading levels by about two grades. While we are continuing to experiment with strategies to systematically optimize CoWrite's reading level capabilities, a couple of things you can do: ask CoWrite for a lower reading level than you actually want and use a tool like Readability Analyzer to evaluate and refine outputs until they meet your standards.
CoWrite may occasionally produce inaccurate information. While mySidewalk has taken steps that significantly reduce the likelihood of this happening, it can happen. We will continue to leverage advancements in AI programming to optimize CoWrite for accuracy, and we also strongly recommend that you manually review each CoWrite result for accuracy, appropriateness, and compliance with your organization’s guidance on acceptable use.
Using CoWrite responsibly
At this time, mySidewalk uses OpenAI to build AI features and products. Like all current foundation models, OpenAI’s models have been trained on large swaths of the Internet, including websites, news articles, books, research articles, and more. Because this data reflects societal biases, the models can have racial, gender, cultural, and other biases, too. OpenAI is taking steps to address the potential for harm, and mySidewalk closely follows their progress. We also recognize the responsibility of our company, customers, and individual users to do our part. As it has always been, it is only through collective effort that we can combat biases of all kinds. That is why, for the ethical use of AI, we must all work together. With this in mind, we encourage our customers to:
Make sure you use CoWrite in accordance with your organization’s guidelines around the use of AI.
Manually review each AI-generated result for accuracy, appropriateness, and compliance with your organization’s guidance on acceptable use.
Let audiences for your data stories know if the content was generated by AI.
Inform mySidewalk of AI-generated results that include inaccuracies, biased ideas or language, or otherwise concern you for any reason.
Read about the steps mySidewalk is taking to combat bias, as well as our full set of principles for the ethical use of AI here.
Frequently asked questions
What is mySidewalk's approach to the ethical and responsible use of AI?
mySidewalk believes that anyone who cares about a place should be able to use data to create positive change. We also believe that the thoughtful application of AI to community data makes this possible. Read our AI principles in full here or on our website. If you'd like to discuss further, please reach out to your customer representative or email us at [email protected].
I'd like to learn more about how CoWrite works technically.
Sure, the summary of our approach to building CoWrite is a great starting place. And if you have additional questions, just let us know!
Will data or data conclusions in a CoWrite output always be accurate?
Often, but we can't guarantee always. CoWrite passes a summary of your data visualization to OpenAI to help facilitate data accuracy and reduce the likelihood of hallucination. However, CoWrite may occasionally produce inaccurate data or conclusions. We strongly recommend that you review CoWrite outputs for accuracy and appropriateness.
Why do outputs consistently score at a higher reading level than directed?
CoWrite (like most generative AI) frequently underrates reading levels. While we continue experimenting with strategies to systematically optimize CoWrite's reading level capabilities, a couple of things you can do: 1) Ask CoWrite for a lower reading level than you actually want and 2) Use a tool like Readability Analyzer to evaluate and refine outputs until they meet your standards.
Why can't CoWrite meet target word and sentence count guidelines?
CoWrite is directionally good at responding to length-related instructions, but it will rarely hit specific word or sentence count targets. There are a couple of reasons for this. First, AI conceptualizes the world in tokens rather than words or sentences. In other words, its unit of measurement is different from ours. Second, generative AI is notoriously verbose and is prone to overusing certain words and phrases. Often, you'll find that simply editing outputs for brevity and repetitiveness will take the writing from a 6/10 to at least an 8/10.
What are the best practices for writing my own custom prompts?
Great questions! We have a whole tip sheet on that.
Where can I provide feedback on CoWrite?
Please give us feedback on CoWrite using the link, which will appear at the end of each newly generated text block. You can also share your feedback with your mySidewalk representative on how this tool might better meet your needs.
Related resources