"Please Provide a Source/Citation" Prompt When the User Pastes Content

סיכום מבוסס בינה מלאכותית

Tris20 proposed a feature to prompt users for a source or citation when they paste text blocks exceeding 100 characters. The goal is to improve traceability for AI-generated content, documentation, and other pasted material. The suggested implementation involves a modal window asking users to categorize the pasted text as “Another website” (adding a URL label), “AI Generated or Assisted” (adding a specific label), or “I wrote this elsewhere” (pasting as-is). While this approach encourages proper attribution and allows honest users to disclose AI use, Tris20 acknowledged it relies on user honesty and suggested it should be an opt-in setting. Tris20 also highlighted an open debate on whether to include a “skip” option, balancing the friction of citation against the burden placed on readers to verify sources.

In response, awesomerobot argued that labeling AI content is an oversimplification that creates unhealthy community dynamics. They contended that since not everyone will be honest about AI use, such a system disadvantages honest users who label their content, while those who hide AI usage gain undue credit. Additionally, awesomerobot warned that asking users to declare potential AI use implies suspicion, potentially damaging trust and making interactions adversarial. Instead of shifting focus to labeling, awesomerobot suggested that communities should evaluate content based on its quality and value relative to its length, regardless of its origin, as this fosters a healthier foundation for trust than policing honesty through labels.

Problem

Large pasted blocks are increasingly common: AI output, documentation, articles, text from other sites. Readers often have no way of knowing where the content came from, which makes it harder to judge its accuracy and context.

Proposal

Prompt the user for source/citation if they paste more than 100 characters.

Example Solution using a Modal window

Modal window might not be the right method in the long run, but it’s easy to mock up and illustrate how this solution could look. Essentially when a user pastes a bunch of text a modal asks the users “Where is this text from” and provides options like:

Another website

User is asked to provide the url which is added as a label to the quote(see image)

:backhand_index_pointing_up: Useful for traceability regardless of the desire to label AI content

AI Generated or Assisted

This adds a small label to the content

I wrote this elsewhere:

e.g. I wrote this in another editor. Just pastes like currently.

You can play with the mockup here :backhand_index_pointing_down:

Benefits

  • Encourages proper attribution of quoted material in general
  • Lets honest users disclose AI use easily

Drawbacks and limits

  • It relies on honesty. Users can choose a misleading option, or retype the text.
  • Some communities may not want an AI label at all, so this would make sense as an opt-in site setting.

Open points

I’m 50/50 about this having a “skip” option. I worry it would be abused, but one the other hand this process adds friction and not all pasted content needs citation. Then again, not adding source applies burden to the reader, and the number of readers typically exceeds the
number of authors so surely the burden should sit with the author?


Related topic: Improved AI content labelling in the composer for post content - #16 by Tris20

5 לייקים

This thought has crossed my mind too, and I think it answers what people are asking for but oversimplifies the problem to the point of creating unhealthy side-effects.

I’ve come to believe that asking people to label AI content is not the answer, and actually ends up fairly unhealthy in the long run. The main issue being: not everyone will not be honest about it, and when they’re not, it is increasingly difficult to prove.

This creates an imbalance where the honest people are hurting themselves in the eyes of their readers by adding a “produced by AI” label (no one is thinking “yay, AI content!”)… and the people successfully lying about their content being AI produced are reaping credit that isn’t fully theirs.

There’s also an unintentional secondary effect where prompting “we think this could be AI” to everyone who copies and pastes tells them that “we think you’re doing something suspicious” which impacts how they think you feel about them.

If content is of low value relative to the length, that’s the problem regardless of origin. By making it instead a problem of labeling, it’s shifting the discussion away from evaluating quality to a potentially adversarial conversation about honesty and labeling. That may not be a great foundation for forming the sort of trust people seek in a community.

3 לייקים