In a nutshell, the NOFOLLOW attribute is not link-specific. If you use it on one link from page1 to page2, none of the other links to page2 from page1 will be followed either.

If this is true, then attempts to have Google favor a specific text link with good link text over some other image link will have serious and unintended consequences. Namely, no link juice will flow through either link!

I tested this myself. The results are pretty compelling.

From a TBPR4 site homepage, I created two links to page-two.html and one link to page-three.html. I used NOFOLLOW on one of the two links to page-two.html. After a week, only page-three.html had been indexed.

66.249.67.230 - - [15/Nov/2009:04:16:55 -0800] "GET / HTTP/1.1" 200 2940 "-"
66.249.67.230 - - [15/Nov/2009:16:55:19 -0800] "GET /robots.txt HTTP/1.1" 200 31 "-"
66.249.67.230 - - [16/Nov/2009:11:43:17 -0800] "GET / HTTP/1.1" 200 3002 "-"
66.249.67.230 - - [16/Nov/2009:13:28:38 -0800] "GET / HTTP/1.1" 200 3002 "-"
66.249.67.230 - - [17/Nov/2009:02:54:45 -0800] "GET /robots.txt HTTP/1.1" 200 31 "-"
66.249.67.230 - - [17/Nov/2009:02:54:45 -0800] "GET / HTTP/1.1" 200 3002 "-"
66.249.67.230 - - [17/Nov/2009:08:53:30 -0800] "GET / HTTP/1.1" 200 3002 "-"
66.249.67.230 - - [18/Nov/2009:00:31:05 -0800] "GET / HTTP/1.1" 200 3002 "-"
66.249.67.230 - - [18/Nov/2009:12:39:40 -0800] "GET /robots.txt HTTP/1.1" 200 31 "-"
66.249.67.230 - - [18/Nov/2009:12:39:40 -0800] "GET / HTTP/1.1" 200 3024 "-"
66.249.67.230 - - [18/Nov/2009:14:16:26 -0800] "GET /page-three.html HTTP/1.1" 404 532 "-"
66.249.67.230 - - [18/Nov/2009:20:20:01 -0800] "GET /robots.txt HTTP/1.1" 200 31 "-"
66.249.67.230 - - [18/Nov/2009:20:20:01 -0800] "GET / HTTP/1.1" 200 3024 "-"
66.249.67.230 - - [19/Nov/2009:06:34:14 -0800] "GET / HTTP/1.1" 200 3024 "-"
66.249.67.230 - - [19/Nov/2009:08:04:43 -0800] "GET /page-three.html HTTP/1.1" 404 532 "-"
66.249.67.230 - - [19/Nov/2009:20:43:46 -0800] "GET /robots.txt HTTP/1.1" 200 31 "-"
66.249.67.230 - - [19/Nov/2009:23:37:18 -0800] "GET / HTTP/1.1" 200 3024 "-"

In other words, despite putting a followable link on the site homepage to page-two.html, Googlebot never fetched it. Why? Because a second link to the same page used rel="nofollow".

So, if you’ve drunk the Pagerank Sculpting Kool-Aid over the past couple years — the temptation to link to some of the more egregiously bad pieces of such advice is nearly overwhelming — then “now” is a really good time to revisit your site nav and strip out all those nofollows.

A big hat-tip to Bruce Clay for not only investigating, but sharing this tip!

Counting the links on a page is therefore a useful tool for understanding how link juice passes through a page. There used to be a great link-counting plug-in for Firefox, but its author stopped updating it a long time ago, and it stopped working as of Firefox 3.

Benj Arriola of SEOreligion.com updated the add-on to work with Firefox 3, but his version expired as of 3.5.1. You can still download Benj’s modified XPI here if you need to.

The link-counting code still runs fine under Firefox 3.5.X, but won’t install without modification. Here are step-by-step instructions for modifying the add-on to work with FF 3.5.3 or later. For OS X:

1. Change the XPI file’s extension to .zip.
2. Double-click the file; OS X will recognize it as a ZIP archive and unpack it.
3. Open the resulting folder to show the archive contents.
4. Open the install.rdf file in a text editor.
5. Find the “EM:maxVersion” line and change the value to something greater than the current FF version, e.g. 3.7.5.*:
   em:maxVersion="3.7.5.*" />
   You might wish to increment the version number too, e.g. to 1.3.1.
6. Save the file.
7. In the Finder, select the modified install.rdf file and the other four items next to it (chrome.manifest, defaults folder, todo.txt, chrome folder).
8. CTRL-click to show the contextual menu for the Finder, and select the “make archive” item (or “Compress 5 items” depending on your OS version).
9. Rename the resulting .zip file to LinkCounter.xpi. You may need to do this within the archive’s Get Info window. Make sure the file has an .xpi extension, or Firefox won’t accept it.

To install the modified add-on:

1. Uninstall any previous version of the plug-in, if necessary; then restart Firefox.
2. Open the Add-Ons window.
3. Drag the new LinkCounter.xpi file from the desktop onto the Add-Ons window.
4. Confirm installation when prompted.

If you use Windows, the basic steps for modifying the XPI are the same, although the OS may not do the ZIP creation for you (not sure).

If you have a minute, vote for our Halloween ecards page at HalloweenDirectory.com:

http://search.live.com/results.aspx?q=wordle

Whoa, could my little website be ranking for “wordle” just a couple days after I mentioned it in a post?

Of course not. But the MSN/Live Search bot sends phony HTTP_REFERER strings when it crawls sites.

You know, I’ve seen faked referrers in my logs a lot. Here are a few actual examples to demonstrate the fine enterprises Microsoft is apparently emulating:

http://www.feelgoodpharma.com/product/c/57
http://www.viagraoverstock.com/
http://www.igsvmortgage.com
http://www.blacks-xxx.com/latina_sucks_monster_meat_rod.htm

(I know this is old news. The fact that Microsoft is still spamming websites with faked referrers a year later is confounding. One thing is sure — the day I run into bandwidth overcharges for this server, msnbot is going to be given a starring role in my robots.txt file.)

Will this actually help me evangelize SEO to the rest of the team? This remains to be seen. Perhaps the real value comes in comparing clouds from multiple data points over time, to visualize the change.

Included here is a sample tag cloud, generated for an old personal blog, just to illustrate the result. Read on for tips and a special tool for Omniture users.

There are a couple tricks to using Wordle to graph search terms effectively:

1. Replace spaces in multi-term phrases with tilde characters (because Wordle expects space-delimited lists)
2. For enterprise use, divide term frequency by something between 2 and the minimum count represented in your sample, unless you really want to wait while Wordle’s browser-hosted Java applet tries to process your 3MB list of search terms
3. And maybe just strip all the branding terms entirely. (I would hope you’re getting a ton of search traffic for your domain name.)

The other trick for enterprise use is getting the search terms in the first place. It isn’t practical to process tens of millions of lines of CLF (webserver log) data frequently. If you have Omniture, try the “Natural Search Keywords Report;” you can request a CSV file of the top 500 items. The CSV data looks like this:

474.,,no on prop 8 petition,1010,0.3%
475.,,acne diet,374,0.1%
476.,,puffy under eyes,215,0.0%
477.,,mac izle,167,0.0%

(Note: contrived data.)

Converting this CSV data into something you can drop into Wordle is as easy as submitting it to this tool. This is a PHP script I put together to optionally strip branding terms, add tildes, expand term counts, and so on.

If you don’t use Omniture, and it is reasonable to process Apache log files directly, you can simply extract the search terms from the log files. This requires a bit of processing too, and I have a tool that as well. But it will have to wait for another day.

Obviously it depends on the search volume for whatever term you’re interested in. Some terms have a lot of search traffic, and others not so much.

Whatever traffic you’re getting for a particular term now, I can tell you what you’ll get if it moves up or down in the SERPs. There’s a calculator after the jump…
If you’ve been working in SEO for a while, you’ll probably remember when AOL published 20 million not-entirely-anonymous web queries, including the rank of the search result clicked by each user. This led quickly to an analysis of the clickthrough rate for the top 10 positions in the SERPs:

Total Searches:9,038,794
Total Clicks: 4,926,623

Click Rank1: 2,075,765
Click Rank2: 586,100 = 3.5x less
Click Rank3: 418,643 = 4.9x less
Click Rank4: 298,532 = 6.9x less
Click Rank5: 242,169 = 8.5x less
Click Rank6: 199,541 = 10.4x less
Click Rank7: 168,080 = 12.3x less
Click Rank8: 148,489 = 14.0x less
Click Rank9: 140,356 = 14.8x less
Click Rank10: 147,551 = 14.1x less

Chances are this chart was simultaneously invented by numerous SEOs, but this particular version was the work of “Breakpoint,” a member of the EarnersForum site (which went offline earlier this year). Breakpoint’s analysis used a sample of roughly half the clickthrough data available in the full AOL data set, and included only the top 10 positions.

I’ve reworked the analysis using the full set of AOL data, and extended it to cover the top slots on page two of the SERPs. This makes it easy to calculate the value of any movement, up or down, within the first 12 results.

To use the chart, find the row matching the old (or current) rank of your page, then trace your finger to the right, to the column representing the new (or desired) rank. Multiply your current traffic (measured in clickthroughs per day/week/etc) by the number in the resulting cell to find out what your traffic would be, given the SERP rank change you’ve projected. The color is simply a visual hint about whether you’re going to gain or lose traffic.

(If the table looks like too much trouble, use the JavaScript SERP Rank Traffic Calculator below.)

It’s not always realistic to think you can move your site to #1, at least not in the short term, but it’s reasonable that some focused effort could bump it up 3-5 positions. What’s that worth? For example, if you’re getting 500 visits per day from your #8-ranked listing, you can project that you’d get 1415 visits per day if you could jump to position #3. (500 * 2.83 = 1415)

This is a great tool for justifying SEO investments — or for avoiding them.

If you can put a dollar value on each clickthrough, you can quickly calculate the relative value of each increase in ranking. By the same token, for some longer-tail searches you might find that there’s just not enough upside potential to justify any expense at all.

Let’s revisit our earlier example: if you make an average $5 cpm for clickthroughs from search, and you think it will cost you $1500 (for content development, linkbuilding, etc.) to realize that +5 jump in the SERPs, your net gain of (1415-500=915) 915 new visitors per day would take 328 days to pay for itself: $1500 / (915 visitors/day * $5/1000 visitors) = 328. That seems like a questionable investment — but at a higher CPM, maybe it would make more sense.

Not only can you use this grid to project the value of and therefore justify SEO investments, you can prevent yourself from sinking a ton of energy into elevating your rank for keywords that just aren’t driving enough revenue to matter.

SERP Rank Value Calculator

By the way, I’ll be at PubCon in Las Vegas next week. I’d be interested to hear your take on ROI for SEO. Please get in touch!

I know all the cool SEOs disparage PageRank, but relative measures ought to mean something — and wikipedia bests our page 6 to 4.

But as of today, the Google proxy Scroogle began showing Care2’s “13 Wonderful Ways to use Epsom Salts” article at #1. That means that at least some of Google’s datacenters are serving up our page in the top position.

Keep it coming, Google… we’ll put our foot soak / sedative bath / exfoliator up against wikipedia’s “anhydrous heptahydrate” any day. (Now that we’ve graduated from high school chemistry class, I mean.)

In some cases, the maintenance page can get indexed!

This is a screenshot of the top result for [jobfinder], our “green jobs” listing at jobs.care2.com:

The problem is that the site got indexed by Googlebot while we had it offline. The title and snippet in the SERP came from our maintenance page, which is hosted by a secondary webserver that responds to all requests when the production websites are offline. We retained our #1 ranking for [jobfinder], despite the not-relevant title and META description.

How can this be prevented?

robots.txt could be used to disallow all spider access during maintenance, but this is not a good answer. Google caches the file’s contents, which means, first, chances are googlebot wouldn’t check the file during the maintenance window, but would instead rely on the cached contents from a previous visit. Or, if robots.txt was read from the maintenance server, googlebot might not come back to the production site for a long time (until the cache expired).

On-page NOINDEX meta tags might work, but again, these might be cached, which we don’t want to happen.

I think the best answer is to use proper HTTP header codes. We had been serving 200 OK headers from the maintenance server, which is really just incorrect.

So, we have reconfigured the webserver that hosts our maintenance page to serve HTTP 503 Service Unavailable headers. According to w3.org, HTTP 503 means:

The server is currently unable to handle the request due to a temporary overloading or maintenance of the server. The implication is that this is a temporary condition which will be alleviated after some delay.

I have not seen confirmation from Google, but after making this change I did discover a relevant thread at WMW: Pages indexed with our temporary “site update” message

Unfortunately, it lies to me. Check out the top 10 items it finds in Google for the search [green living]; the Care2 site is listed at #7 and #8:

1. www.green-living.com/
2. www.thegreenguide.com/
3. www.thegreenguide.com/doc/122/collegetransport
4. www.nrdc.org/greenliving/
5. www.nrdc.org/greenliving/toolkit.asp
6. www.greenlivingtips.com/
7. www.care2.com/greenliving/
8. www.care2.com/greenliving/make-your-own-non-toxic-cleaning-kit.html
9. www.greenlivingjournal.com/
10. www.gliving.tv/

But look at what I get when I go to the actual Google search results — Care2 has only a single result, at #5:

1. www.green-living.com/
2. www.thegreenguide.com/
3. www.nrdc.org/greenliving/
4. www.greenlivingtips.com/
5. www.care2.com/greenliving/
6. www.greenlivingjournal.com/
7. www.gliving.tv/
8. greenliving.suite101.com/
9. greenlivingideas.com/
10. ecomall.com/greenshopping/

What’s going on?

My rank-tracker is apparently requesting 100 results at a time. Google serves different results if you request 100 of them. Not just more results — different results too.

I think that’s a problem, because the vast majority of Google users will only see 10 results. Which means my rank-tracking software is not telling me the truth.

Therefore, I’m really excited to announce the launch of a wonderful new blog…
The Adventures of Green Girl

Lily Berthold-Bond is the daughter of Annie Berthold-Bond, published author and well-known expert on household toxics and all things Healthy and Green. After growing up in an eco-friendly wonderland, Lily is now a college freshman, living on her own, and facing real-life decisions about whether to continue living a natural, chemical-free lifestyle.

She has a great story to tell, and a great voice with which to tell it. Tune in for more over the coming days.