Changing your content creation strategy
I'm working on a chapter on using search analytics to improve a site's content. It'll cover some of the stuff you'd expect, like using top null results queries to tune your content offerings. But I'm also hoping to dig a little deeper into some less-than-obvious uses of analyzing query data for improving content.
Here's one that I'm not entirely sure makes sense, and I'd love some input. Without a huge effort, it's possible to categorize queries. We've already posted an example where frequent queries from Michigan State University were categorized (note the color coding). This doesn't take that long to do, and by addressing the short head, you're actually categorizing a large portion of all of the site's queries.
If you tally the numbers associated with each category, you can start determining an overall profile of your searchers' needs. 21% of searchers might be looking for information on navigating the campus, 15% might want information on study abroad, 4% might want health information...
So here's the question: with such a profile in hand, might you revamp your content creation strategy?
If, for example, you found that your only 5% of your content addressed study abroad, would it make sense to try to increase the supply of such content to match the 15% demand?
I'm not sure this makes much sense, as I'm not sure how exactly one would determine the topicality of their content. I'm not certain the math makes all that much sense either.
But I wonder if there's something to this that I'm missing. If nothing else, could we use that breakdown of searcher's needs by percentage as some sort of loose guideline for content creation?
Comments
I'm not sure search volume should be an indicator how much content you have on that subject.
I think you are on target about using it for strategy but you should look at outcomes. What are people doing those searches trying to achieve? How well does the site support those outcomes? This goes beyond just getting them to the right content and gets at the ultimate value of the overall site.
You could conceivably have a single page that supports the desired outcomes of 20% of the searches. If it does, that is a great success.
Posted by: David Gammel | October 5, 2007 4:35 PM
David, you nailed my concerns with the numbers perfectly.
So what do we do with this data? I'm almost envisioning big pie chart of topical breakdowns that helps all sorts of people--content managers, information architects, and more--visualize the big picture of content demand. What this would provide beyond a potentially compelling visual is beyond me, but maybe that would be good enough...
Posted by: Lou Rosenfeld | October 5, 2007 4:41 PM
I am also skeptical of this, but would like to hear more.
First thoughts:
You may be assuming here that all content has the same requirements. That "content is content" and all content does its job equally. Some more complex content may require more explanation or may need to be broken across multiple pages.
On the "study abroad" question, it seems the bottom line is whether the 5% is working for people. Correlation doesn't show significance. Just because the 15% of searches equals the 15% of content doesn't tell you that people are finding what they are looking for.
David's point about the single page at the end of his comment seems right on. I might be concerned if the numbers got way out of correlation I suppose.
Posted by: Mike Steckel | October 5, 2007 4:47 PM
You guys have me pretty much convinced: give up the correlation. But aside from what I've already suggested, do you see a benefit from showing the breakdown of search queries by topic?
Posted by: Lou Rosenfeld | October 5, 2007 4:56 PM
Lou -
I use a couple of feedback loops of analytics to content creation.
There's one approach that says that you're looking for the pages that people view that they really have to dig for (e.g. page 2-n on Google search results), and do content creation to more specifically address those needs. 103bees reports directly on that, or you can export keywords out of your logs into a ranking tool and come at these indirectly.
If you filter the keyword list to find people typing in full blown questions (who what when where why how can ?) you'll often unearth explicit user intent that can help focus or refocus content that you have.
It's easy in Google Analytics to report on day-to-day or week-to-week variations on search term frequency. This will let you tell in advance when seasonal traffic is starting to kick in - last year's "when is trick or treating time" post generates traffic this year 30 days before Halloween, which gives you 30 days to get it right again for this year instead of scrambling the night before.
I wish I had the reporting on "people who viewed this page also viewed this page" a la Amazon right in my content management system, but I don't. In general, surfacing trends right to the reader forms a feedback loop that newspapers and news aggregators have discovered and paid attention to.
Posted by: Edward Vielmetti | October 5, 2007 5:10 PM
Definitely. Given the right circumstance. A university website's content landscape may change more slowly than an e-commerce site, I'm not sure. But for an e-commerce site, a breakdown could show a growing area or a changing area. Something people are expecting your site to have, that you may not carry. Maybe you could learn about a change in your audience's desires.
It may raise questions about whether people are having trouble navigating an area of the site (so they resort to searching). You may consider doing some user testing here if you are already nursing suspicions about that area.
I would see a breakdown as a sort of a "check up" document. In itself, it might not be that useful. But in connection with other data, it could come in handy.
Posted by: Mike Steckel | October 5, 2007 5:18 PM
Each of your top queries needs an answer on the site. Preferably, one answer, easy to find, on one page. The title of the that document needs to make it clear that the answer is in that document. If you are lucky, the actual answer will show up in the snippets or summary (maybe it is the customer support phone number).
I disagree that one page answering 20% of queries is a great success. If that makes it harder to find the answers, it is a big problem. If that was a good idea, we would all love the sites that put the entire FAQ on one page.
You can use traffic to adjust your editorial effort, but I'd do that with HTTP access logs. The search numbers are biased by all sorts of things. The HTTP access stats include visits through WWW search, bookmarks, links sent out in e-mails, even good 'ol "click here".
Checking the access logs is good, because you do need to eventually delete the "installing KERMIT on Windows 3.1" how-to.
Posted by: Walter Underwood | October 5, 2007 5:22 PM
I think some sort of at-a-glance reporting has to be useful. If nothing else, it helps the centralised web/intranet team to say: "well, I know you think you're section is super important, but only 2% of people are searching for information you currently provide"...
At the end of the day, as you well know, analytics tells you *what* people asked for, not *why* they were looking for it. Search analytics is a lot better than usage stats, but it still runs into this inherent limitation.
What I do think, however, is that the search solutions need to start thinking about the *business management* of search, which will require meaningful dashboards, summary stats and management tools (all integrated).
PS. perhaps there's some association with alignment diagrams (or whatever they're now called), in terms of matching up needs with available content?
Cheers,
James
Posted by: James Robertson | October 5, 2007 8:22 PM
Here's another nail in the coffin of trying to match search volume with content volume...
The volume of content needed to explain a given topic depends on the complexity of the topic, not to mention due consideration of the intended audience. To use an extrememe example, you'd probably need less content to explain single-digit addition and subraction than you would for say Einstein's Theory of Special Relativity.
So the amount of content you provide should be driven by communicating your message to your intended audience, rather than what percentage of search users are looking for it.
Cheers
Stewart
Posted by: Stewart Walker | October 7, 2007 7:43 PM
My experiences with online retailers suggest that that Search Analytics content creation is most effective when it demonstrably improves the relevancy of searches at the "term level." I use the following: subjective relevancy judgements, identifying component term elements to facilitate web anaytics and content optimization(e.g. "18v cordless drill" is an "attribute-modifier-noun" string), creating new synonyms (e.g. "fuzz buster" = "radar dector") and manufacturer part cross references, and classifying search terms in the existing product taxonomy to feed product addition processes.
Since conversion rates and user behavior can be tracked at the "keyword level," there is, in my experience, little need for rules of thumb for content-to-search-frequency ratios. For instance, I may offer 10 different products, but 90% of my sales and searches may be related to one of the ten products. Although I may naturally have more content than average related to the primary revenue generating item, the "content ratio" (how would this be calculated by the way, number of bytes?) would likely not be 9:1. Nor should it necessarily be.
Of course, my experience is with retail websites selling products or "SKUs" (stocking units). Websites skewed (no pun intended) towards information requests may find content ratio metrics more valuable.
Posted by: Eli Cooley | October 10, 2007 2:32 AM
As always, fantastic suggestions! Mike, I think you're really on to something with your suggestion of using this data as a longitudinal check-up tool.
Posted by: Lou Rosenfeld | November 6, 2007 11:38 PM