Objective

• As a user, I want to ask an AI bot a question I am curious about having to do with climate science and have it provide me a succinct answer.

• As a user, I want to see suggested questions The Post think would be a helpful starter.

• As a user, I want to ask a question and give feedback on how well it was answered, the sourced articles and their relevance to my interests.

• As a user, I want to ask an AI bot climate related questions and when it can't answer it, I want to be given an explanation as to why and a relevant CTA to continue my discovery journey.

Given the experimental and relatively new nature of AI utilization, our objective was to meticulously select language that effectively communicates its operational mechanisms, our strategies for ongoing model enhancements, and current risk factors. In this project, we prioritized comprehensive language that is:

Educational;

• Educate the user that this involves AI and is experimental
  Always present and prominent throughout the flow

Informational:

• Educate that any response is solely based on our published reporting (and only ours)
  At least present on initial landing but important throughout as well

• Show a clear methodology section aka “how it works” section that goes a bit deeper into how the tool works
  Can be treated as an appendix and secondary

• Imply if not outright state there may be mistakes in answers “e.g. This is a beta experience. Please verify the auto-generated answers by consulting the full articles. Notice a mistake?”
  Always present and prominent throughout the flow

1. Queries per user

2. Qualitative feedback

3. Return visits

4. Pageviews per session

Our previous VQA process was hindered by the absence of a streamlined system that could be utilized by the entire team to ensure effectiveness and timeliness. In response to the tight deadline for this project, my primary objective was to optimize the VQA process between design and engineering, ensuring that all items were accounted for while delivering a highly functional, accessible, and delightful product. I developed and implemented a new VQA document featuring a tagging system designed to:

• Clearly define all VQA issues and bugs

• Explicitly indicate the priority level of reported issues

• Accurately translate items for the engineering team

• Effectively communicate approvals

• Clarify any remaining issues

Through this project, I have realized the critical importance of meticulously informing users about the involvement of AI in experimental engagements. We dedicated a considerable amount of time to refining our communication to ensure clarity regarding functionality and associated risks.

Furthermore, I conducted extensive deliberation on optimizing the VQA process. Given the project's rapid turnaround requirement, I reviewed past successes and challenges to develop a comprehensive document for tracking reported issues and approvals. This document was instrumental, as it was shared among designers, product managers, and engineers, becoming an essential component of our daily stand-ups to monitor progress leading up to the launch.