Page MenuHomePhabricator

CentralNotice severely impacts CLS score
Open, Needs TriagePublicBUG REPORT


I've noticed a sharp deterioration of the "Page Experience" Score from Google Search Console (for the Hebrew Wikisource), starting September 1st:

image.png (261×908 px, 28 KB)

A quick investigation pointed to Central Notice banners, specifically the Wikimedia Soundlogo Contest campaign. According to Lighthouse estimation, the banner adds to the Cumulative Layout Shift about 0.6 seconds. The banner images lack width/height properties, which should be fixed.

The delayed CLS severely impacts page visibility.

Event Timeline

Fuzzy renamed this task from Central Notices badly impairs CLS to Central Notice severely impacts CLS score.Sep 19 2022, 12:26 PM
Fuzzy updated the task description. (Show Details)

Possible duplicate: T280476.

mpopov added a subscriber: jwang.

@jwang has looked into this but not for Hebrew Wikisource. Putting this into tracking for us for now as it's not clear what work will be required of Product Analytics. Will wait for more info from Olga.

@ovasileva I remember you and @jwang had looked at some of the Wikipedias, and wanted to surface this to you. Please let us know if you think we need to look into this further, or if you think we have enough information from prior investigations.

I also know we're looking into other SEO-related information - tagging @ahemmer so you're aware of this concern.

@greg also wanted to give you a heads up since I know you've been following the concerns around Central Notice.

Krinkle renamed this task from Central Notice severely impacts CLS score to CentralNotice severely impacts CLS score.Sep 20 2022, 5:56 PM

Discussed this with @jwang and we will look into it further.

Hiiii all, thanks so much for this... a few quick notes, in case they're helpful:

  • It's been known for some time that CentralNotice banners negatively impact Cumulative Layout Shift (CLS) scores, which are one component of Google's Core Web Vitals (CWV) metrics.
  • As far as I understand, what's less certain, and much harder to evaluate, is how those scores may impact our ranking in Google's organic search results, and, in turn, pageviews.
  • Though Google publishes details of how CLS is evaluated, how it factors into search result rankings is inside the "black box" part of Google's algorithms. There have been vague statements [1][2] from Google on this and at least one outside analysis [3]. It seems the biggest hurdle in figuring out how CLS might impact rankings and pageviews, based on the data we do have, is that so many other unknown factors are also at play (like social trends, other changes in Google's ranking algorithms, changes in non-organic search results, unknown delay between when CWV scores change and when the impact on rankings occurs, what the scores of our sites and other similar sites are on the other metrics that Google takes into account, etc.).
  • CLS doesn't measure seconds of delay per se. Rather, it's a "visual stability" score. Below 0.1 is considered "good", between 0.1 and 0.25 is "needs improvement", and above 0.25 is "poor".

See also T298733.

[1] From this blog:

While page experience is important, Google still seeks to rank pages with the best information overall, even if the page experience is subpar. Great page experience doesn't override having great page content. However, in cases where there are many pages that may be similar in relevance, page experience can be much more important for visibility in Search.

[2] From this discussion, cited here:

relevance is still by far much more important [than CWV scores]

[3] From this blog:

did anything happen? Actually, yes. But it’s more subtle. The real impact was felt for URLs that failed all three [of the CWV] tests.

Hebrew Wikisource is not in the campaign list in Please clarify the source of below statement.

A quick investigation pointed to Central Notice banners, specifically the Wikimedia Sound Contest campaign

Hebrew Wikisource was a target of the campaign first phase. I didn't keep screenshots of the page because I did not think this statement needs proof.

@Fuzzy, we need to know the timeline of the campaign on Hebrew Wikisource for analysis. That's the reason I am looking for the source of the statement. Do you know the timeline?

According to the Page Experience chart, the problem started August 30th. I cannot say when the campaign was over since we had holidays the past week. But I think it ended the last few days.

The sound logo campaign only started on 13 September so likely isn't the cause of this. (It did target Hebrew Wikisource though, I'm not sure why the calendar doesn't reflect that.)

There might be other causes which started earlier. Still, according to the Lighthouse analysis, the Central Notice added about 0.6 seconds to the CLS score. Also, the images in the banner lacked "width" and "height" properties, which also impacted the CLS.

Krinkle added a subscriber: Krinkle.

The general concern with any banner is indeed that CentralNotice is architected today such that it causes a layout shift. This is already tracked and investigated at T280476.

The issue about the Soundlogo campaign (both the banner, and the Soundlogo web site) will be handled elsewhere, but the banner is no longer live as of October 10th, so that's moot now.

@Fuzzy Can you confirm that the metric has recovered since then?

We did some initial analysis of WMF pageview metric on Hebrew Wikisource.
Pageviews referred by Google search engine counts for more than 90% pageviews from external search engine. [1]
However, we did not see pageviews referred by external search engine dropped as "Page Experience" Score dropped. [2] [3]

[1] search engine referral dashboard link

[2] Desktop pageviews on Hebrew Wikisource referred by external search engine, 2022 compared to 1 year prior (2021) and 3 year prior (2019). superset link

image.png (846×1 px, 280 KB)

[3] Mobile web pageviews on Hebrew Wikisource referred by external search engine, 2022 compared to 1 year prior (2021) and 3 year prior (2019) . superset link

image.png (864×1 px, 257 KB)