AITA app

Reddit´s “Am I the Asshole” or AITA for short is one of the biggest sub-reddits on the platform. It is a space where users can post their inter-relational conflicts and get opinions and judgement from other users.

Background

The combination between intense drama and moral exploration has led to these reddit posts to be shared across social media platforms such as TikTok, X, YouTube, podcasts and more.

This led my team and I to consider how we can improve on the experience in engaging with this format as well as how we can enable an even larger user-base to participate.

Design opportunities

In this case study, I will highlight three key opportunities we uncovered through interviews with current AITA users.

User experienced related opportunities:

• The forum uses tags to reveal what the user consensus is of a given post once it has been up for a while.
• • These tags may trigger bias in some users even before entering the post.
  • The tags only displays what the most represented vote is and in doing so potentially misrepresenting the discussion as one-sided.
• The forum lacks a way for users to vote on posts and is subsequently done via typing their vote in their comment
• • There is no way to get an accurate overview of how users have voted
  • If a user only wants to submit a vote, they still have to write their vote and submit a comment.

We were also curious in challenging:

• AITA is entirely in English, and although there are some similar forums catering for speakers of other languages, this language barrier disallows potential interesting cultural differences to be revealed and discussed

Solution

Tags

To account for the bias problem, we created a different set of tags indicating what kind of discussion is going on in the post without revealing the outcome, thereby allowing users to form their own judgement.

In taking this solution back to the users we saw it testing well for most of our testers with only smaller adjustments required. However, we also saw that a significant number (~20%) of testers preferred to get more information up-front.

To cater to their preference, we included an option to see a diverging bar chart illustrating how current votes are distributed instead, in doing so giving users a more nuanced and accurate preview.

User voting

By separating voting from commenting, we made voting more accessible while additionally enabling users to filter comments on the votes of their posters.

Since commenting and voting were the two main actions that users take after reading a post, we decided to attach these options to the navigation bar, allowing users to engage with the post regardless of where they are in the page.

The diverging bar chart from the post cards is visible front and center, although for users who opted for preview tags, the chart will first be revealed after they have voted.

In testing, we noticed a false affordance in the bar chart with users attempting to click it to vote. Instead of altering the chart sections, we chose to rely on user intuition and make them clickable in the next iteration.

While 'Not the asshole' and 'You are the asshole' are the most common votes, other options are also accessible by clicking the button below these two.

A full vote break down is also available for when users want additional granularity. The breakdown will remain visible across all posts until the user chooses to return to the simpler chart, allowing for some more customizeability.

Globalization

In order to facilitate conversations taking place between people speaking different languages we wanted to translate posts and comments to readers’ interface languages.

Established platforms like Youtube and Instagram do this by having a button allowing users to translate posts or comments one by one. However, with this approach users are unable to scan foreign language comments and decide what threads they want to participate in. Therefore we ventured to automatically translate posts and comments.

This approach presents a different set of challenges, although more on the technical side. So if you are interested, read ahead for a developer perspective on the issue, otherwise you can skip to the end.

With the recent advances in AI models, machine translation has been significantly improved and can increasingly provide translations that seem natural, with some models being able to translate slang accurately as well (we successfully tested Swedish to English with Claude 3.5 Sonnet).

However, running and using AI models is costly, therefore we wanted to reduce the number of API calls as much as possible

We did this in four ways:

1. Avoiding unnecessary translations:

By asking the users during onboarding which languages they know in addition to their native language we are able to disable translation for all comments that users can understand on their own.

2. Store data in the database

By storing translations alongside the original texts in our database, we only have to translate a text once per language and subsequent users can benefit from faster loading times.

3. Restricting languages

Currently we are restricting the language pool to a select few languagues (English, Hindi, Spanish, Arabic, Bengali, French, Russian, Portuguese, Indonesian, German, Japanese) hoping to cover most users. Once we see significant use in an area not covered by these languages we will work to include them as well.

4. Translate comments in batches

Comments in the app are fetched in batches in order to reduce load time. By only translating the comments in the current batch, we make sure that we don’t translate any comments unnecessarily.

This feature is highly experimental and we suspect that a lot of parameters will have to be tweaked along the way, but we believe that the gamble could be worth it!

This project is still unannounced to the public and is a work in progress.

However, if you want to read more you can do so here:
https://amitheasshole.app/about

Thank you for reading!