Ask HN: How Trustworthy is Google PageRank?

10 points by mcdowall ↗ HN
Hi guys,

Wondering if anyone can give me some feedback on their experiences with SEO and their site PageRank, we have been feverishly working on optimizing our commerce site : http://www.off2.com for the past few months and went live in the UK earlier this week.

Weve been crawled by Google yesterday (i think!) and returned a Pagerank of 7. We previously had managed to get a PR4 with our static holding page through inbound links, optimization etc. Obviously we are very happy about a PR7 but I really want to know whether others have found that PR can sometimes be incorrect and adjusted by Google. I've checked over our inbound links to find some .edu sites which of course carry a lot of weight with their PR8's.

Would love to hear the thoughts on any reversals or sudden changes others have seen.

24 comments

[ 4.0 ms ] story [ 64.4 ms ] thread
There are two PR calculations: the internal one and the publicly visible one you see on the toolbar.

Google uses the internal PR value as one factor (out of reportedly hundreds) in determining the rank of a page in the search results.

The publicly visible one is, for all intents and purposes, an entertainment channel like, for example, Comedy Central. It is never an accurate reflection of the current PR value Google uses internally.

So what to do? Use the Webmaster Tools and authenticate your site. The Webmaster Tools give you a ton of data about how Google sees your site which is a lot more valuable from a little green progress bar.

Pagerank (edit: the "ranking algo") is calculated from hundreds of factors? Wow. That seems overcomplicated. It's hard to imagine that the best way to rank a page is to consider hundreds of influences.
No, that's not what I said. PageRank is one factor out of hundreds that determine the rank of a hit in the search results.

PR is, very basically, a network analysis algorithm. A simple way of thinking about it is the random walk analogy: Imagine randomly browsing pages on the web by clicking on links, what's the probability of you landing on the page in question? The higher the chance of you randomly finding the page the higher its PR.

But again, this is a very very simplistic view and the actual algorithm is of course very very secret. The idea isn't though, and you can read the original paper by Brin and Page at http://cli.gs/qVDNYm .

PageRank is one factor out of hundreds that determine the rank of a hit in the search results.

I'm referring to the algorithm for calculating the rank of a hit in the search results. If there are hundreds of influences, then it seems overcomplicated.

I'm referring to the algorithm for calculating the rank of a hit in the search results.

That's not PageRank but the ranking algo.

If there are hundreds of influences, then it seems overcomplicated. Well it needs to be a general algo that works with all types of content, all fields of knowledge, work around all known types of spam, in all languages, while working on HTML quality.

Complicated? No, necessary.

That's not PageRank but the ranking algo.

That's what I'm referring to! Hehe. I feel insane. Am I really failing at communicating?

Let me try again: If the ranking algo considers hundreds of influences, then it is an overcomplicated mess!

Actual ranking at the time of your search does not involve too much computation, because a lot has been precomputed and stored in those Google indexes.

In theory, you can view every search as a function of your query and all the web documents out there. This function is partially evaluated for a fixed universe of documents and super optimized.

Evaluating hundreds of parameters is common sense for machine learning tasks. If you look at the papers explaining how the top Netflix prize competitors achieve their score, this is again what they do: they run literally hundreds of algorithms over the dataset and then blend the results. And this is for a domain where their data is only (user, movie, rating, date) quadriplets.

Yeah... But just because a design is also a scientific paper doesn't make it overcomplicated.
Don't get your point...

Mine was that the solutions to machine learning problems are inherently messy (at least this is the state of the art).

Typo, sorry. I meant, just because a design is a scientific paper, that doesn't make it a good or simple one.
Google keeps it very clean, and they tune the algo every week. But again, it's necessary because relevancy to a human being is continent on THOUSANDS of factors, and they try to incorporate as much of that in as simple of an algorithm as possible.
> If the ranking algo considers hundreds of influences, then it is an overcomplicated mess!

"overcomplicated mess" seems to imply "it's a bad solution".

While Google's ranking system may be "bad" (they believe that it isn't good enough), "bad" or even "overcomplicated" doesn't follow from "hundreds of influences".

Consider the "design" of a common house cat. Its design is the result of thousands of influences.

It may be difficult for people to think about systems with hundreds of influences, but that's a people problem and doesn't imply that such systems are inferior to more understandable systems.

Housecats weren't designed, they were evolved. That's a big difference. There can be cleanliness and simplicity in design. Also, a housecat's goal is to live and to reproduce, which involves solving many problems, whereas the ranking algorithm's goal is to match text to relevant pages, which requires solving far fewer problems than a living creature, so it's really not a fair metaphor.
> Housecats weren't designed, they were evolved. That's a big difference.

What properties of a design do you think are determined by the design process as opposed to its outcome?

> whereas the ranking algorithm's goal is to match text to relevant pages, which requires solving far fewer problems than a living creature

While "relevance" is the commonly used term, "utility" or "value" would be more accurate. And they're far more complicated than you seem to think.

There is no "best" search result for a given query. There's not even a best result for a given person. Add in their context, and you're getting close, but there are still other factors that are relevant to determining "best".

I suspect a large part of ranking algorithm's complexity comes also from evolution - they need to coevolve with all SEO stuff. Optimal ranking is a moving target, parasites break your fitness.
I wouldn't surprised if it is thousands or even millions of influences, many of which might be generated automatically.

Part of it might just work like a spam filter, for example, so even single words could be factored into the algorithm. Just guessing, but there could be a rule like "if site contains the string '$$$' reduce rank by 0.5%" and so on - all automatically generated, just as it would happen in a spam filter.

Also take a look at SpamAssassin, I have heard it let's you manually combine configure and weigth all sorts of algorithms. At the end of the day, most of Google's employees concerned with improving the search results are probably doing just that most of the time.

Even if some Google researcher comes up with a new algo tomorrow, it will probably just be factored in, it won't simply replace the previous algorithms.

(comment deleted)
I remember reading this a few years ago. We had an SEO firm perform some site analysis for a previous site and have just taken onboard the main principles and tried to replicate them.

I think thats a good assumption reading up on it in more detail, the relevance of the rank itself is a very small factor in the performance of the site and its keywords etc.

I think we may have some relevant high worth links coming in that i wasnt previously aware of, our own link building campaign rarely acheived more than a PR5 or PR4 site so I will look into the backward links.

Thanks for the info, I was somewhat sceptical we had just jumped from PR4-7 overnight, I have seen a large number of new inbound links but seeing as we only migrated the site over onto the new servers and propagated the URL last week we still have loads more work to get done on it.

Our keyword market in travel is as anyone would expect extremely competitive so hopefully SEO over time will prove to be a good investment of our time and effort, we were just rather shocked this morning at the new crawled Pagerank!

Well SEO on its own is not that useful. SEO to drive conversion is very very useful. To generalize, what's the final objective of doing SEO? Common answers are "get more buyers", "get more subscribers", "get people do download X", etc.

To give you an example: I have a PR5 page that ranks in the top 10 for a 50m results query. The page has links from really (really) awesome sites (think multinational companies). Traffic from this query? Minimal. Why? It's not a commonly used keyword and certainly doesn't drive that many subscribers to the site.

Exactly, were aware of the relevance of keywords and the optimization of our deep level pages with more detailed keywords.

We have started a review this week of our internal link structure and the use of more specific travel terms, ie 'holidays in tunisia' or 'package holidays in tunisia' for our deeper destination pages. Hopefully the relevancy of these search terms coupled with a well performing site should yield better results than we could ever anticipate for 'travel' or 'holidays' etc!!

Cheers for the input

Forget about your page rank. It won't get you anywhere. It helps you know where you are at, but what matters most is search traffic. You want to make sure that continues to climb. A high PR helps, but focusing on it won't help. Focus on building links with good anchor text, that match the keywords you are focusing on.

And if you do have a an actual PR 7 in Google that is great. It means you have a lot of link authority. Use it by targeting certain key pages on your site for the keywords that will drive traffic.

If any of your team went to university in the uk, get the company featured on their news page - they all want to push their enterprise credentials - with a link, thats a free way of getting good google juice.
We currently lease offices out of our local university (excellent subsidised rates!) so I am already over this, I've sent out a press release relevant to the local market and the university. We should get some links on their innovation centre website hopefully - its a .ac.uk tld so carries some weight with google and other engines.