-1

TL;DR: Our recent comment experiments concluded, and we gained valuable insights: Users want to say thanks, and ask follow-up questions on existing answers. We know that current options aren't ideal for enabling follow-up questions, especially when code blocks or images are needed, so,we are now planning a new experiment explicitly focused on enabling these follow-up questions.


For context, this post builds on our previous communications and experiment regarding comments:

Our goals included enabling users to express thanks more subtly than in past experiments. Additionally, we experimented with redirecting users to Discussions to learn if subjective conversations with lower participation barriers could flourish in places other than comments. There will be a follow-up post later discussing those specific findings.


What We Learned from Recent Experiments

'Say Thanks' Feature

The adoption we saw was higher than anticipated, reaching 1200 uses per week, which is a bit more than one and a half times the expected 750 uses. Our analysis shows that it did help in reducing thank you comments. In total, we saw a ~10% decrease in comments that included some form of 'thanks' from Feb to Mar 2025. To that end, we will be releasing the ‘Say Thanks’ feature at a later date once we have made some additional improvements to it. For instance, creating a direct path to give Thanks, such as its own button. Feedback throughout the experiment made it clear that the multi-item menu on the comment link was not ideal.

The Need for Follow-Up Questions

Originally, Discussions was chosen as the path to create new content because we had a hypothesis that users had subjective questions related to existing questions. We found that many interactions weren't broad discussions but specific, technical follow-up questions related to an existing question and its answers. Reviewing and labeling over 100 posts from the experiment revealed the common types of follow-ups users created:

  • Asking a direct follow-up question about the existing Q&A (49 times).
  • Sharing a slight variation to the provided answer (14 times).
  • Asking "How does this work?" or seeking clarification (10 times).
  • Stating "I still need help" or noting the answer didn't work/is outdated (8 times).
  • Other categories included discussing the answer, asking about best practices, clarifying the original question, asking a different question, or sharing "what worked for me" (19 times).
  • About 50% of these follow-up attempts included code blocks or images. These takeaways led us to believe that despite finding answers to questions they were investigating, users were still looking for help on the topic. Help they can’t easily get in comments due to the 500 character limit and a less than ideal space to ask follow up questions related to an existing piece of content.

Confusing Content Types

One theme that emerged is that users find it unclear how Comments, Discussions, and Questions differ. Users trying to ask a legitimate follow-up question were sometimes unsure where to post it. This placed a burden on them to understand Stack Overflow's specific content structures, which can lead to confusion for a number of reasons, such as a lack of proper guidance and unclear site rules. While not directly related to this experiment, many of these issues were reflected in responses to a conversation we had about closed questions facilitated by our research team.

The Core Issue

Currently, if you have a question that builds on or is related to an existing Q&A, there isn't a straightforward and clear path for new users, and sometimes even experienced users, to ask that question. Often, such a question ends up closed as a duplicate. We do have to reopen flows, and in theory, a truly unique question shouldn’t be closed if it meets other appropriate standards. But we know that’s not always the case. Users often lack the background knowledge that experienced members possess regarding existing answers. This knowledge gap creates a significant barrier when determining if their question might be a duplicate. Comments have limitations for this purpose (length, formatting, no images/code formatting, site norms). This friction point is something we want to address with our next set of experiments.

What's Next? Focusing on Enabling Follow-Up Questions

This team's primary focus now shifts to exploring the best way to allow users to ask these follow-up questions directly on existing posts. We plan to run a new experiment specifically designed for this use case.

Our experiment of sending users from Q&A to Discussions to ask follow-up questions showed us a different need that Discussions doesn’t quite solve either: that users want to have more subjective conversations and inquiries surrounding existing Q&A. Discussions will continue to be treated as a separate feature, and there is a separate post sharing learnings specific to the Discussions product.

Next Steps & Your Feedback

Based on these findings, we're concluding the current comment experiments (though the 'Say Thanks' feature will need a little work before it's rereleased). Our primary focus now shifts to designing and launching a new experiment specifically aimed at providing a better way to get those follow-up questions on existing posts. We believe this will help more people solve their problems effectively while maintaining the quality of Q&A. Our intent is to look at evolving comments for this purpose.

We'll share more details about the upcoming follow-up question experiment soon. We welcome your thoughts and feedback on these learnings and this planned direction. We will keep an eye on this post for feedback till May 6th, 2025

8
  • 23
    How does that 10% decrease in "thanks" comments compare to the decrease in commenting in general during the experiment?
    – Kevin B
    Commented 2 days ago
  • 1
    @KevinB Hopefully they normalized by the number of comments. Commented 2 days ago
  • 3
    @KevinB I will have to double check to confirm, but we didn't see a concerning drop in overall comment volume during that experiment.
    – Hoid StaffMod
    Commented 2 days ago
  • 4
    "'Say Thanks' did reach 1200 uses per week" - how many times is the "Add a comment" button used during one week, for comparison? (And how often did people press the button but leave neither a comment nor a thanks?)
    – Bergi
    Commented 2 days ago
  • 4
    Just to note... another goal of these experiments should be to not drive existing established old rott... I mean, visitors, to a point of frothing at the mouth. Helping out folks to acquaint better with the site is all good, but not at the expense of the people that helped to make the knowledge base actually valuable over the years. It should not come as a surprise that changing something that has been the way it is for nearly two decades and was not broken is going to not be received well. Separate button, duh.
    – Gimby
    Commented yesterday
  • @KevinB The first week we saw a 9% decrease in total commenting, which was 2% above what we expected, around 7%. After that, we saw an average of 3% decrease in commenting on a weekly basis.
    – Hoid StaffMod
    Commented yesterday
  • @Hoid, Is the 3% relative to the first week or relative to before the experiment? Does 3% decrease mean that in the later weeks, there were even fewer comments than in the first week (i.e., in later weeks, the number of comments was 12% less than before the experiment), or that in later weeks there more comments than in the first week but less than before the experiment (i.e., in later weeks, the number of comments was 3% less than before the experiment)?
    – D.W.
    Commented yesterday
  • 2
    What if you added "say thanks" to the upvote button itself ?
    – Criggie
    Commented yesterday

7 Answers 7

17

To that end, we will be releasing the ‘Say Thanks’ feature at a later date once we have made some additional improvements to it. For instance, creating a direct path to give Thanks, such as its own button.

That defeats the purpose. You want to reduce the number of "thank you" comments. A separate button doesn't really work that good; and we've been down that road: Feature test: Thank you reaction

Feedback throughout the experiment made it clear that the multi-item menu on the comment link was not ideal.

Yes! I was amongst users that did not like the pop-up menu for the comments. I still don't like it. But, you're missing the point. An established user (let's say 500 reputations, for lack of a better metric) does not need to deal with that menu and its side effects such as broken userscripts and keyboard shortcuts. Yet, a new contributor can benefit from it. They will learn about "thanks" being considered fluff, they will know about voting (if they have gained the privilege, that button would add an upvote for them), and more.

That little annoyance (little, because they're not leaving as many comments as a moderator for instance), is worth it.

10
  • 1
    It isn't explicit enough though, the only thing that people might learn from it (if their mind is open to it) is that saying thanks is something different from posting a comment. But are people really triggered to care to know why? Nah, there is an annoying popup in the way.
    – Gimby
    Commented yesterday
  • @Gimby we can add a message that pops up after hitting Say Thanks which explains those. But let me raise this; does a separate button address these concerns?!!1
    – M--
    Commented yesterday
  • 1
    I am not sure I agree. I think the 'say thanks' button is different from the emoji reaction experiment. One is a visible expression of thanks on the post, and the other is private. I think we saw enough here that the option, even when hidden in a menu, does decrease the number of comments. However, that is a little fuzzy because some comments can include a thank you and a relevant clarification that makes it not truly a thank you comment. Regardless, when it gets rereleased, there will be some testing with it to determine its effectiveness as a button.
    – Hoid StaffMod
    Commented yesterday
  • @Hoid A/B testing, maybe?
    – M--
    Commented yesterday
  • How about the "thank you button" being an alias for the up-vote button, both of them having exactly the same mechanic?
    – Lundin
    Commented yesterday
  • @Lundin So you'd need 15 rep to say thanks?
    – PM 2Ring
    Commented 22 hours ago
  • 1
    @PM2Ring You can press the vote buttons all you like no matter rep. If something meaningful happens is another story.
    – Lundin
    Commented 22 hours ago
  • @PM2Ring it was like that. If you had enough rep, it'd have upvoted and thank the OP, if not, just a thanks notification for OP, if I am bot mistaken.
    – M--
    Commented 22 hours ago
  • @M-- Probably, but a different team is picking up that work, so I am not gonna make commitments for them.
    – Hoid StaffMod
    Commented 16 hours ago
  • @Hoid as long as it's passed on; that's all I can ask for. Cheers.
    – M--
    Commented 16 hours ago
16

Do Stack Overflow Staff have any interest in helping clean up the mess that was created in the Discussions section of the site? I understand that there are 1,000's of flags in the queue, now, due to the volume follow up questions that were generated in there; and I don't even know if everything got flagged that should have (I know I personally gave up flagging).

It would be nice of Stack Overflow help clear up the mess that they helped create; my estimate is there's probably at least 500 posts that need deleting (discussions was sitting at a little over 600 post pre experiment, and it's now at over 1,250).

1
14

I think there is a decent chance that I might have totally messed up your experiment. Let me explain.

As you are clearly well aware, there are a lot of "thanks" comments. Many of them are really obvious, and I figured I could make an SEDE query to find them with more or less 100% accuracy, and even find only ones that were nuked with only 1 flag, to avoid bothering the mods.

Turns out I could and I pretty quickly did. I then used said query to flag thousands of comments, and I also linked many other users to the query, who likely used it to flag many more comments themselves.

Based on my work with my comment bot, I'd estimate there is roughly 9,000 or so "No Longer Needed" comments a month, of which a significant amount if not a majority are not "thanks" comments (ex. "You're welcome" or "Please upvote").

I am aware of (roughly) what you were looking to figure out how "thanks" comments were affected (although that was told to me in a private chat by a staff member, so I think I'd need permission to post it publicly here) and I'm also aware of your cutoff data for the baseline (also told to me in the private chat), which unfortunately happens to be roughly around when I stopped doing this. Based on what I know, it seems its quite plausible that my comment flagging alone accounted for most or all of this 10%, and that's ignoring the unknown number of flags cast by users I shared the query with. These comments likely wouldn't have been flagged/deleted otherwise. Comments are so hard to moderate, particularly "No Longer Needed" ones, that the vast majority seem to remain undeleted, while my query makes it quite easy for someone to spend 10-15 minutes and flag 100 comments daily.

I did alert a Community Manager about this, so did you wind up doing something to fix/avoid this problem?

7
  • 1
    related in terms of assumptions and skew: meta.stackoverflow.com/a/432669/11107541
    – starball
    Commented yesterday
  • 2
    If the experiment looked at the number of 'thanks' comments posted then it should be fine. Only if they looked at the comments that are still visible (i.e., non-deleted) at the end of the time period then the analysis would be affected by your cleanup efforts.
    – Marijn
    Commented yesterday
  • @Marijn As previously noted above, I have information from a private chat which I'm not going to share publicly without permission (particularly since such information was not included in this post). Based on this information, I do think that my flagging would affect their analysis. Commented 19 hours ago
  • @Marijn In general, though, the challenge of measuring the number of thanks Comments is probably that the vast majority are never flagged/deleted, so there is no way to know which comments are thanks comments, short of looking through all comments. How exactly would one determine the number of thanks comments posted without that information? Commented 19 hours ago
  • 1
    I am waiting for confirmation, but the analyst was made aware of this, and I do believe they accounted for it.
    – Hoid StaffMod
    Commented 16 hours ago
  • @Starship there is a regex for single-flag deletion of thanks comments, you can use the same regex to count from all posted comments how many thanks comments there are.
    – Marijn
    Commented 15 hours ago
  • @Marijn Except said regex doesn't account for all thanks comments, and not all comments with said regex are thanks comments. Commented 15 hours ago
13

For follow-up questions, from an askers point of view it could be quite difficult for them to tell what they should really be doing. What I mean by this is that a user could have a follow-up question for several reasons, such as (but not limited to):

  1. The question they asked wasn't quite complete, and missed out some context; this doesn't change the question entirely, but does impact the solution(s). For example they may have omitted information on (some) edge cases.
  2. The answer is actually just plain wrong, but the reader of the answer lacks the ability to identify that
  3. The reader is in a "me too!" position, and failed to implement the problem properly
  4. They genuinely do have a follow-up question; perhaps because they have implemented that solution and are now stuck on a latter (but related) part.

These scenarios likely all need different actions by the reader, and not all need a follow up question. Using the above number points, my expected behaviours would be

    1. Provided that the change doesn't significantly change the asked question edit the question to improve it and provide the missing context. If an existing answerer chooses to update their answer is up to them.
    2. If the missing context is significant, a new distinct question should be asked. (I've seen many times a user repost the entire question with changes, which often results in duplicate closer). The answers to the existing question should be judged on their attempt to answer the asked question (not the unasked question)
  1. Downvote the answer (if able).
  2. A comment may be appropriate. Follow-up questions, depending on the users attempt, can often result in duplicate closure as they don't actually link to the answer they tried to implement (incorrectly), and then get pointed back to the post that they failed to implement. Ideally, the user should cite the original answer, demonstrate their attempt and explain why it didn't work with a full MRE.
  3. Ask a new question.

All of these different actions need good awareness of how the site, community, and curators work. If we do want follow-up questions (which I'm not against), I think that the user should be directed on being asked why they want to ask a follow-up, and this should differ depending on if the user is the author of the question or not. This means that the user might be directed to comment, or downvote, rather than asking a follow-up question. If they are directed to ask a follow up question, then a box like shown in Discussions in the new question, linking to the answer they were looking at, is important. If it's a "Me too!" question, they should also be directed to included their attempt, and explain why it didn't work, as this will help users see that their question is distinct enough. It could still be a simple error, but at least it's not a "I'm having this problem too, but the answer didn't work" type post.

This dialogue could be presented a little like the flags screen, and then leads to a (more) bespoke ask wizard or advises the user on the "correct" action, such as "Please consider editing your question to include inforamtion on the edge case and leaving a comment to let the answerer know you've added some additional context". If they are sent to a question wizard it should include appropriate prompts for the user to ensure that they demonstrate the relevant information, based on their follow-up question "type."

5
  • 1
    "If the missing context is significant, a new distinct question should be asked" -> Yes. As someone who answers questions, I don't mind answering a few clarifying follow-up comments (even years later). But I don't want to be committed to helping the OP further if substantial effort may be required. Commented 2 days ago
  • (from my comment above) "to helping the OP further" nor some other visitor who happens upon an answer months or years later. Though I would not object to a mailbox notification about such follow ups. Commented 2 days ago
  • @snakecharmerb I would be fine if the op (or anyone really) could simply indicate in such cases via a button that the post should instead be a separate standalone question
    – Kevin B
    Commented 2 days ago
  • You bring up some really good points about finetuning it. Right now, our focus is just on where we could put these that make the most sense and get good enough visibility. Discussions was not the right answer, and we think that evolving comments more intentionally is the right direction, but we may be proven wrong. Next week on MSE, I am posting a lot more details about the upcoming experiments around this, and I think it's a good start, but probably will need more fine-tuning down the road to incorporate your points here, which, to be clear, are things we have talked about internally as well
    – Hoid StaffMod
    Commented 16 hours ago
  • I would like to think we can find a middle ground where someone asks for clarification for whatever reason in the comments, gets an answer, and as a community user, can decide that this should be a separate question, and we can make that happen. However, it's sometimes unclear to askers when their clarification should be a question, resulting in duplicate closures or closure for low-quality content. Its not too high of a burden, but when someone just doesn't know what they don't know it frequently ends up in a closure cause they believe their question is actually the same thing.
    – Hoid StaffMod
    Commented 16 hours ago
10

In total, we saw a ~10% decrease in comments that included some form of 'thanks' from Feb to Mar 2025. To that end, we will be releasing [...]

How did you run your numbers? Did you normalize for the number of days per month? Did you normalize against overall change in influx of comments? (which, by my calculation was about -8.08% for february).

Currently, if you have a question that builds on or is related to an existing Q&A, there isn't a straightforward and clear path for new users, and sometimes even experienced users, to ask that question.

No, I'd say it's been pretty straightforward. Just make a new question post and link to the related one, explaining the connection in words as needed. The Linked Questions section shows the relation on both sides. But this experiment made things more complicated and confusing by suddenly throwing discussions into it.

14
  • Possibly the linked question section could be divided into from/to links to make a bit clearer who cited whom. Or there could be a submenu of the ask question button on a page with an existing Q&A that offers to create some skeleton for a "related" question. Commented 2 days ago
  • 1
    Note that SEDE has no option(s) for deleted comments, so your numbers could easily be skewed by the lack back.
    – Thom A
    Commented 2 days ago
  • 1
    @ThomA that's why I used the lowest comment ID for each month and got the diff between each month.
    – starball
    Commented 2 days ago
  • "Just make a new question post and link to the related one, explaining the connection in words as needed" is obvious to you, but it's not obvious to someone who hasn't been using the site for years. An "ask followup question" that automatically creates the link will be a major benefit for new users.
    – Mark
    Commented 2 days ago
  • @Mark I don't disagree. I think it's good to explore making it more obvious to users what to do with site features in common user scenarios. I just can't understand why discussions was chosen as the place to route users to instead of Q&A for... Q&A.
    – starball
    Commented 2 days ago
  • We were measuring on a 7 day period from the start to end date. We expected to see a decline in commenting somewhere between 3-7%. In the first 7 days we did see that hit 9%. However, the average from there was hovering slightly above or below 3%.
    – Hoid StaffMod
    Commented yesterday
  • This is the main reason why when we return the feature, it will get a dedicated button. Clearly the drop down menu deterred some people from commenting. So when its brought back it will be a separate button.
    – Hoid StaffMod
    Commented yesterday
  • 1
    @Hoid so why is my number different than yours? that doesn't explain how you (the company) did your calculation. "Clearly the drop down menu deterred some people from commenting" people's feedback on meta supports that, but to what degree the data "clearly" shows that... again, did you normalize for context like the existing, consistent downward trend in site activity?
    – starball
    Commented yesterday
  • @starball Possibly this experiment caused people who would have posted comments to post non-answers which were later turned into comment and hence got a later ID, which (maybe?) explains the discrepancy. Or possibly they did something to account for the problem in meta.stackoverflow.com/a/433654/21440243 Commented yesterday
  • 1
    @Starship I can't imagine that to make a statistically significant difference. I have a sort of good thumb on the pulse of non-answers and I didn't see any uptick. and the way I calculated things by taking the lowest comment ID per next time period, I'm pretty sure my calculation is fairly immune from that anyway.
    – starball
    Commented yesterday
  • @starball Perhaps the company calculation is not immune to that. And I like to think I know about how many NAAs there are as well...but it seems that the number of Natty and Dharmanbot reports points to you being correct about there being no uptick in NAAs. Perhaps what I was noticing was an uptick in the time NAAs stay (becuase most people involved with dealing with NAAs, particularly from SOBotics and NAABot, disappeared or their activity was far lower) Commented yesterday
  • @starball I couldn't say for sure why your number is different. I know they used a 7-day moving average to review the data. I would need to schedule some time with them to review that, and I'm not sure when that would be. In terms of deterring commenting, we see that in the data we reviewed, there was a decline in commenting, especially as indicated in the first week. In particular, we saw comment levels return to relatively normal after the drop-down menu was removed. So it's not airtight reasoning, but I do believe it to be the most likely.
    – Hoid StaffMod
    Commented 19 hours ago
  • @Hoid what's not airtight reasoning? I'm saying that if there's an overall trend of decline in commenting in the same period you observe decline in thanks comments, you should take that into account. if you didn't already, your "10%" decline in thanks comments should be adjusted to be relative to the general decline, which would make it less than 10%.
    – starball
    Commented 18 hours ago
  • @starball I am saying my reasoning is not totally airtight, with comments returning to normal at the conclusion of the experiment. I am not up to date on commenting declines as a whole, but the snapshot in the time before the experiment and after, I believe, are sufficient enough to identify harm we may have inflicted with the change outside of any general ongoing decline over the last few years.
    – Hoid StaffMod
    Commented 16 hours ago
8

The adoption we saw was higher than anticipated, reaching 1200 uses per week, which is a bit more than one and a half times the expected 750 uses....

That doesn't tell me much because I have no idea what you expected nor do I assume that my expectations would be the same. I would be more interested in if the total rate of thanks increased. You report 10% less thanks in comments and these 1200 additional uses per week correspond to how many percent? Please provide either only absolute numbers or only rates (per total number of comments I guess) or both but not a mixture of them.

All in all, if that feature is used 1200 times per month and reduces some (don't know how many) comments and has its own button, I would accept it, even though there is no real benefit to it (it's not a vote). But in general I still think this feature is not needed and distracts more than helps. I won't use it and vote instead.

Btw. did voting suffer in any way by people thanking?


Regarding the follow ups: they often read like comments that should be comments not follow-up questions.

  • "Sharing a slight variation to the provided answer (14 times)." Either put as new answer or as edit to the answer.
  • "Asking "How does this work?" or seeking clarification (10 times)." Would be best directly under the content it is concerned with.
  • "Stating "I still need help" or noting the answer didn't work/is outdated (8 times)." Answer didn't work is useful directly below the answer. Still need help can be anything really.

And if there really is a follow-up question, why not ask it as a question and refer to the original? That could be done already in the past. If they just end up as discussions I fear they might take away some potential Q&As. Everything that can be a Q&A, should be a Q&A. Discussions or chats should be second level to that (just my opinion).

Discussions would be fine, if curation would be possible of the discussions. So far it's not well supported as far as I know. This seems like a big gap. Did you maybe check how the created discussions fared? How many of them resulted in some sort of satisfying conclusion?

But as long as there isn't another extra menu just to comment I'm fine with everything.

0

I'm concerned about encouraging follow-up questions. I don't mind clarifying some detail in my answer, and of course we all should repair any errors when they're pointed out. However, writing an answer should not be an open-ended task.

If someone has a related / follow-on question they can ask a fresh question, linking to any existing questions or answers that they feel are relevant. If they also post a comment on my answer asking me to take a look at their new question, that's ok, but I'm not obligated to answer it, or even read it.

This is closely related to the issue of chameleon questions. IMHO, encouraging follow-on questions has to be done with caution. Otherwise, it could encourage "help vampire" behaviour.

1
  • 1
    When considering encouraging follow-up questions, we want to avoid making the original authors of answers feel obligated to respond to every semi-related question. So it is something we have on our radar to think about, as it's also an existing issue with older answers when authors no longer keep tabs on their content. We would like to think we can find a way to elevate these follow-ups in a way that anyone feels comfortable chiming in to clarify things, without cluttering the question feed.
    – Hoid StaffMod
    Commented 18 hours ago

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.