Is there data or research on how effective suggested answers are in preventing people from asking duplicate questions on online forums and Q&A sites?

While some people make an effort to search, some people will click the "Ask Question" button (or whatever the button name is) and will start typing their question. To deal with this, many forums, including this one, dynamically suggest to users answers which match the subject of their post.

Two specific questions I am trying find some data about are:

  1. What % of people click on such answers (obviously it depends on quality, but some examples would be helpful).

  2. Does it actually stop people from posting their questions (that would probably require an A/B test with two pages, one with auto-suggest on and other with off, so that impact can be attributed only to auto-suggest).

Any other related insights or ideas would be appreciated.

  • Thanks AndroidHustle, fiar point. I tried to post my question on meta.stackoverflow.com. Commented Jun 3, 2013 at 12:07
  • 2
    @AndroidHustle, as specifically relating to Stack Exchange this is a MSO question. Not as it relates to other sites. The OP asked it here: meta.stackexchange.com/questions/182790/…
    – Ben
    Commented Jun 3, 2013 at 12:13
  • 7
    @AndroidHustle: this is surely a valid UX question; it just so happens it asks about a UX element that’s used by this site, but it’s asking about the general UX aspects, not anything stackoverflow-specific.
    – PLL
    Commented Jun 3, 2013 at 13:29
  • @PLL Ok I see what you mean. But it happens to be SE specific in this question, "What % of people click on such answers", this is a question that is very SE centric and does therefore not fit outside the meta compound IMO. Commented Jun 3, 2013 at 13:45
  • 1
    Could you please clarify the reason of your question - why it is important for you? I think, any abstract % data will not be relevant to your particular case. I can suggest you solution for 100% reading suggested answers before asking your own - allow 'Ask' button only after reading all your suggested answers.
    – Serg
    Commented Jul 16, 2013 at 7:56

3 Answers 3


I think the answer is not about having precise stats on suggested questions, but about providing the best exit points to your users in that context in general (known UX patterns). Suggested questions only being one of them.

You can still ask "yeah but are suggested questions effective?". It still comes down to what your users are doing on a case by case basis. I mean, what works for Quora, might not work for SE, etc.


Here is an interesting article that generated some interesting ideas on this: Link

In response to your specific questions - You could run a small sample usability test with a simple prototype on some users (as you've suggested) and take down your own findings. I haven't come across any specific data on this as it's very specific to Q&A sites but doesn't mean you can't do your own research that would be valid.

It may also be interesting to look at help-desk request data, or even data relating to tourism centers or information points at train stations. The reason for this is that this may be considered as a human behavioural and psychological pattern. Information centers provide detailed printed information of a variety of things to do. Most things someone needs to know will be in there, but it may be considered more effort to search, look through and then identify the right answer. Some people may be happy to do this by my initial inclination is that most people would ask the staff at the desk their specific question even though its been asked thousands of times and regards of clear signage giving the answer. This is very similar to the Forum scenario and could give you some interesting insights.


If your goal is to understand broadly what % of people would click on such answers, I suggest to follow Pareto's Law and seek support on the rest of the principles of network dynamics.

Pareto's Law says that roughly 80% of the effects come from 20% of the causes. Following the principle, you should have a +/- 20% of clicks to the suggested answers.

The % will vary depending of your understanding of the behavior of the users and the interactions and contents proposed to them.

PS: Although this is not a specific answer, I think that it gives a concrete start point to analyze the problem.

  • 3
    The 80-20 number is just an example. Guessing that every intervention will have 20% effectiveness doesn't help the OP.
    – Dan Hulme
    Commented Jul 20, 2013 at 10:11
  • I understand your point, but I have to disagree with you: 80-20 is not an example, but one of the rules that mold our universe. I suggest you to go deeper on the normal distribution and statistics. Commented Jul 20, 2013 at 10:42
  • 5
    @marianogoren: That is wrong on too many levels to list in 1 comment. Most importantly, "normal distribution" applies to continuous unbounded distributions. "Does or does not" is discrete, not continuous. Furthermore, the wikipedia link shown explicitly mentions that the 80-20 law comes from the power law, which is quite distinct from the normal (Gauss) distribution (e.g. left-bounded at 0). Next is your choice of variables: with equal validity (i.e. none), I could argue that 80% of new questions are answered by 20% of existing answers, and therefore you should have 80% clicks, not 20.
    – MSalters
    Commented Sep 2, 2013 at 9:27
  • @MSalters thanks for the contribution, I'll study what you're commenting here Commented Nov 1, 2013 at 19:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.