Google's "page loading speed" emphasis made our site slower in reality

[relicx]

I'm afraid that once again Google is making the Web unintentionally worse (happened many times before, for example when people were afraid to use anything that might cause Google to think they have hidden text on the page, even when it was useful for visitors, like in a "Spoiler" button for movie spoilers).

There is a huge problem with Google's crude attempt to use total page loading time for ranking long pages (that require scrolling down to fully view): it likely uses the total page loading time, not taking into account that in the user's browser the page could be visible long time before that, if he doesn't scroll down. We own very popular websites with long pages and we always tried to optimize the experience for the user by showing him what we can as soon as possible. That meant splitting images and Javascript into small parts that only load when they are actually used in that part of the page. This way the user can see the page on his screen as soon as possible. None of the current tools, such as YSlow, webpagetest.org (recommended by Matt Cutts), or Google's very own PageSpeed understand this, so there is absolutely no reason to think that Googlebot could understand it either.

Traffic from Google rankings is important to us, so we did what we think they wanted: we listened to the recommendations of these tools and combined images and Javascript to make the total page loading time quicker, making our pages appear to load slower to actual users. This is what happens when Google implements crude measures with a lot of secrecy about their methods - the Web becomes worse.

[tedster]

The page load speed is taken from toolbar users, not from googlebot's crawling experience. Also, there is no reason to assume that page speed actually IS a factor at this point.

We've been told that page sped might become an additional factor, and I'm sure Google does intend to do that. But it seems right now that the data is not accurate enough to use in the ranking algo.

Your page loading method sounds like a kind of progressive enhancement - is a new http request made when a visitor needs some feature?

[outland88]

Also, there is no reason to assume that page speed actually IS a factor at this point.

The fact is if page load speed were involved very little consensus would be reached in these threads because of the many variables involved. As with many things now Google may roll out enforcement piecemeal so a true consensus is often difficult. I seriously doubt Google will penalize its “fat cat buddies” either in regard to this matter.

I have had some unexplainable page drops that coincidentally involved Google reporting slow page speeds but it is impossible to say with 100% accuracy. It does however arouse suspicions.

As an example up until two days ago, like most, I didn’t believe the first page was for sale. The pricing schematic in Google Merchant Center tends to indicate otherwise.

[vordmeister]

Sometimes long pages with loads of images are very useful. At least I like to think so because they are my speciality. :-)

I have never split a page into several shorter ones as I find it a hassle pressing the next button at the bottom when all that stuff could have been loading while I was at the top.

Possibly Google will muck things up, but in the past they have tried to promote sites that are useful. I'm not changing my pages even though fully loading them might take a while. My own test is to see if they seem quick to me on 28k dialup.

Many sites are completely inaccessible on dialup or even my massive 256k broadband. News sites with adverts are the worst with over 1 minute before anything appears. It's those folk who should worry not us.

[ddogg]

Google's page speed tool is stupid anyway. It says my site takes 3.5 seconds to load on average. Give me a break, I have surfed it from many countries, it always loads in less than a second.

[Swanny007]

My main "widget" site has the content loading very fast, with the AdSense and other ads, Facebook widget, etc. all loading after the content. It's good and bad. The content loads quickly so the user can start reading right away but it's the extra features that slow the time of the entire page load. Google can't just say "oh, it takes 7 seconds for the page to load" and call it a day. There's more to it than the total load time.

[londrum]

i'm one of those sceptics who reckon this whole page speed thing has got more to do with keeping google's costs down. given the speed at which the web is growing day-by-day, their costs must rocket everytime they send their spider out. and they have to store all these pages as well. and run their scripts all over them to create content for their other sites, like google news. they want the web to go on a diet. so they've put this scare story into people's heads that their rankings will suffer if they don't cut their pages down.

ranking pages on speed is impossible. imagine a shop which just gives the details and a buy button. that will probably be pretty speedy. but imagine another one which includes a short video (like amazon do for their movies), more pics, customer reviews, similar items etc etc. which is the better site? if you listened to google they'd have you dumping all the extras.

[KenB]

It doesn't matter how many seconds Google Webmaster Tools Site Performance page says it takes your pages on average to load, because it is a relative figure based on users who have the Google Toolbar installed (many using IE6 which is very slow). What matters is how your site compares to others and what percentage of sites your site is faster/slower than. Whether the methodology is precise or slightly flawed, the methodology is consistent across all sites so the relative comparison is useful. It also helps in determining if certain changes made to a website are having an impact on end users.

In regards to the loading time of a webpage while using the Page Speed extension, this also is only useful as a relative benchmark from test to test as the simple process of measuring page loading times really slows down the loading of pages on Firefox. If you want a more consistent measurement of loading time you should be using Opera and its developer tools.

The idea should be to make a site as fast as possible without sacrificing functionality, unless that functionality is totally superfluous.

[relicx]

KenB, I'm not sure you understood what I described. After the changes we made, our pages are faster to load according to all existing testing tools and I'm sure Google Webmaster Tools will show an increase in speed. But these tools are not just slightly flawed, they are totally wrong and misleading, because they use the total page loading time. What matters is what users see on their screens and our pages filled out the user's screen very quickly and continued to load after that.

Think about a page that has 30 picture thumbnails arranged vertically. You have two choices:

1. Regular page that loads these pictures one by one, with one file for each thumbnail. First pictures will show up quickly and the user won't even know that the rest of the thumbnails are not yet loaded, unless he scrolls down immediately. Good user experience, but a lot of connections and some overhead for each picture, so the page speed testing tools show bad performance.

2. One huge sprite (look up CSS sprites for more info) and the page uses CSS to split it up into separate pictures. Horrible user experience because the user has to wait until the whole sprite loads completely before he sees anything, but the page speed testing tools show improvement.

[graeme_p]

I cannot take Webmaster Tools perfomance assessments seriously when I see stuff like this:

Details: Save up to 18.0 KB, 5 DNS lookups
Enable gzip compression
Compressing the following resources with gzip could reduce their transfer size by 18.0 KB:
http://example.com/page/ (4.36 KB)
[google-analytics.com...] (13.7 KB)
Minimize DNS lookups
The domains of the following URLs only serve one resource each. If possible, avoid the extra DNS lookups by serving these resources from existing domains:
[some stuff taken out here]
[pagead2.googlesyndication.com...]
[google-analytics.com...]

It is accurate in a way: my pages load fast until the first Google ad shows.

[hugh]

the metric is useless but my frustration with it has lead me to solve some very long standing slow mysql query problems so every cloud...

[KenB]

Compressing the following resources with gzip could reduce their transfer size by 18.0 KB:

Those suggestions on WMT Site Performance page in regards to Google Analytics are a known issue. There has been discussion about this in Google Groups from Google employees acknowledging the issue. They say that the Google Analystics scripts are being served gzipped to users, just not to Google. For now just ignore it and eventually they will make that message go away.

The domains of the following URLs only serve one resource each. If possible, avoid the extra DNS lookups by serving these resources from existing domains:

Notice the phrase "if possible? These are only suggestions of things that could help improve performance IF they can be changed. Obviously, you probably want your AdSense ads so that you can make some money. It is best, to reduce the number of DNS calls that have to be made by the browser. At the same time, however, google-analytics.com is called by so many websites that the odds are the DNS entry is already cached by the user's computer.

It is accurate in a way: my pages load fast until the first Google ad shows.

WMT Site Performance, is as you pointed out, accurate as to how long it takes a page to completely load. HOWEVER, there is no way that the Google Toolbar can measure how long it takes for only the stuff that is important to the user to load. Only the human eye can determine when a page is usable while the rest loads.

WMT's Site Performance page and the Google Page Speed add-on for Firefox are only tools. When these tools are used properly and with a little thought, they can help one make judgment calls as to what issues can and should be addressed. Not every issue impacting page loading times can be addressed as you must balance functionality against speed.

[KenB]

the metric is useless but my frustration with it has lead me to solve some very long standing slow mysql query problems so every cloud...

By helping you to solve a long standing issue, it was by definition, useful.

[Bewenched]

I'm frustrated as well... they've told us all along to build the site for our users.... well one of the things we want to start using to display larger product images is a requires the use of a scriptalicious javascript, we acutally have it being served by google through our website, but it still slows down the page because it's a fairly large file... it's only used when they click to see a bigger picture. However with this new speed factor in rankings should we now dumb down our pages just for the sake of search rankings, which in turn makes it less useful to our customers?

[KenB]

Page Speed is only a tool, it can't replace human judgment. Used properly it will help you make your pages load faster, but its suggestions need to be balanced against design and functionality requirements.