Talk:Readers/Information Retrieval/Phase 1

Feedback regrading Android app onboarding

Hey there,

I just got the onboarding message in the Android app and have two thoughts I wanted to share:

1) The page (which I got to from the pop up) mentions that the Hybrid Search experiment is opt-out / opt-in. (It's a bit unclear which one it is supposed to be.) From the onboarding message itself, it doesn't seem like the user has an option, the only buttons are "Learn More" and "Get Started". Maybe "Get Started" leads to some option, but unfortunately the app got closed in the background and I can't access that screen anymore, so I can't know. (Also, the feature seems to have been activated without me being able to decide.) In any case, having to click "Get Started" feels a bit coerced, especially when one has suspicions regarding the feature. (I assume that's far from the intention here, I just wanted to give my user perspective.)

2) It's very unclear to me how AI (and, separately / more specifically, LLMs and cloud-hosted models) is used for this. As an user, I'd really like to know about this before enabling the feature / having it enabled for me. (Again, I know this is just an experiment and I assume best intentions, but I think it's helpful to consider this perspective in order not to alienate people.)

Also, feel free to point me elsewhere if this Talk Page is not meant for this kind of feedback.

Of course, even if this particular experience was a bit irritating, I'm very thankful for all the folks putting in work to make MediaWiki and Wikipedia great and even better! <3 Felurx (talk) 12:36, 28 March 2026 (UTC)

Hello @Felurx,

Thanks a lot for taking the time to share this.

On your first point: this feedback is very valuable, and we’ll make sure to share it with the team. One of the goals of running this as an experiment is to gather perspectives like yours and understand how the experience is perceived. The intention is not to make this feel forced.

On your second point: the current experiment does not use AI to generate new content. The results shown are based on existing Wikipedia content, and no new summaries are being created.

And yes, this Talk page is absolutely an appropriate place to share this kind of feedback.

Thanks again for sharing your thoughts, and for the kind words as well. it’s really appreciated. ARamadan-WMF (talk) 14:06, 3 April 2026 (UTC)

Feedback regarding in-app navigation (Android)

I did have a poor experience with the hybrid search experiment, and the main part of it relates to the process of navigating to an article.

If a user knows the exact or approximate title of the article they want to read (for example, the article I wanted was "List of chemical elements" and the search term was "list of elements"), there's now an extra step before they can reach the article. Although the article title appeared among the suggestions after typing in my query, tapping on the title of the article took me not to the article whose title I tapped, but rather to an interstitial page.

I had to scroll past the summary cards and find another link to the article in order to continue to my intended destination. The summaries were effectively an obstacle to information, rather than a source of it. The only way I could find to bypass the obstacle was by disabling the experiment altogether.

I didn't think about whether the summary cards were relevant to the information I was trying to retrieve. I already knew where I could find that information (the article page) and was at that time trying to go to that location. For that reason, I didn't even look in the summary cards for the information I wanted.

It seems like this workflow is designed for the use case of users who do not know what article contains the information they seek, nor which related articles would be likely to link to it. I can appreciate the intention to improve that experience, but it would be better if it does not come at the expense of adding unnecessary steps to the workflows of users who are already proficient at information retrieval.

Second, less importantly, I think that cards are a poor fit for certain article types, such as lists and disambiguation pages. I struggle to understand how an excerpt from the middle of these pages could be useful at all.

And, as a final minor point of feedback, when I realized that I was looking at summary cards instead of an article, I felt a momentary flash of dread and rage. I thought, "Oh no, is even Wikipedia pivoting to AI now?" The disclaimer that the content I was seeing was taken from human-written articles did not reassure me in that regard; it seemed like the kind of evasive phrasing that would conceal the fact that an LLM had summarized a human-written article. Only by following the "learn more" link and coming to the experiment page did I see that the summaries were excerpts, not paraphrases, to my great relief.

Using AI summarization for the contents of the hybrid search page would be a very serious error, I believe. AI summaries are less reliable than the text they summarize. Displaying an AI summary as the first result of a search would mean occluding reliable information with unreliable information, which I think would be contrary to the purpose of an encyclopedia. ~2026-22870-39 (talk) 19:25, 13 April 2026 (UTC)

Hello @~2026-22870-39,

Thank you for taking the time to share this detailed feedback.

I understand your point about the extra step, just to clarify one point: the goal of the summaries was to help with discovery. This was a short experiment, and it has already ended this week. You can follow future updates and developments on the project page here.

Your notes help us better understand how different types of content behave in this format.

Thanks again for sharing this so thoughtfully. ARamadan-WMF (talk) 13:27, 16 April 2026 (UTC)