14 replies to this topic

[DanThies]

Some of you know that I have a "thing" about relevance. Targeting the right keywords isn't possible unless you have an understanding of what searchers are looking for.

Just for fun and grins, I've put our relevance calculation tool online for my High Rankings pals. It's another Excel spreadsheet. You can download it here:
http://www.seoresear...a/relevance.zip

Now here's how this works. We take a generic search term, like "interior design." In our example, we'll assume that the site we're working on is an interior design firm, and they want visitors who are hoping to hire an interior design firm.

To get numbers to work with, I used Wordtracker's "Compressed Multi Search" feature, to eliminate most of the goofiness due to capitalization, stemming, etc. Even with this I had to eliminate a couple plurals that snuck in.

I analyzed the top 10 terms that included "interior design" in my spreadsheet. I assigned relevance percentages generously, as follows:

interior design ideas - count of 389, 0% relevance
interior design software - count of 327, 0% relevance
interior design schools - count of 212, 0% relevance
interior design jobs - count of 201, 0% relevance
interior design magazine - count of 158, 0% relevance
interior design school - count of 145, 0% relevance
japanese interior design - count of 121, 0% relevance
interior design degree - count of 94, 0% relevance
interior design firms - count of 81, 100% relevance
home interior design - count of 79, 100% relevance (very generous)

I multiply the count for each term by its relevance, to get a weighted count. Then I take the total searches (for the sub-terms, not "interior design") divided by the total weighted count. In this case, a total of 1807 searches, with a weighted count of 160 based on relevance. 160 divided by 1807 yields a relevance score just under 9 percent.

We could use more search terms, and dig a little deeper (I have), but the number doesn't change much. If you still worry about these generic search terms, maybe this example will open your eyes. Competing for rankings may be fun, but competing for relevance is a lot more profitable for most sites.

[Jill]

I don't get it!

Where do the relevance % numbers come from? You have to make them up yourself? If so, how do you decide?

Jill

[Scottie]

I think it's just that people who are searching for these specific terms are not looking to hire an interior design firm, which is what the client is hoping to achieve as the goal, except for the last two search queries.

I'd take it a step futher and venture a guess that those last two still aren't terribly relevant to our design firm which generally works in or around say, Miami. It needs to go a step further and target a regional keyword as well, unless this a high-end design firm whose fees include travel.

[Jill]

I get what you're saying Scottie, but still not clear as to how you would put a percentage on the phrases.

From what you're saying, interior design firms should probably not have a relevance of 100% right?

[Scottie]

I wouldn't put 100% on it myself... if we are saying that 100% of the people who are searching for "interior design firms" would be looking to hire an interior design firm, then no.

[DanThies]

The idea of relevance is to ask "what percentage of searchers using this term are looking for what my site offers?" It's a judgement call. It's hard to make an accurate judgement with generic search terms, though.

It's much easier to do with specific search terms. If someone types "interior design school" then they definitely are not looking for a design firm to hire, so that gets 0% relevance. When someone types "interior design firm," it's not really 100% (I was being generous), but it's close.

In my experience, people vastly overestimate the relevance of generic search terms. Most interior designer firms would probably say that "interior design" was 50-100% relevant for their site. Digging deep, and giving the most optimistic/generous assessments of relevance, the best I can come up with under the rosiest scenario is still less than 15%.

The nice thing about this exercise is that it gives you a more accurate way to judge relevance for a generic search term. It's not perfect (perfect is impossible in this case), but it's proven very useful.

[cline]

To achieve the same objective, I like to run a robust Adwords campaign in advance of SEO. The Adwords gross clicks (CTR x impressions) is an excellent indicator of what is most relevant. It's even better if you have conversion data.

Like our esteemed Mr. Thies, I too find that usually the generic terms are not nearly as relevant as the site owner thinks they are, and usually the terms one rung up in order of specificity are the targeting sweet spots.

[DanThies]

We also recommend doing targeting studies with Adwords, they take time but the data is very valuable. The thing we run across is that they sometimes generate too small a sample, and that a lot of clicks happen just because the ad is "there."

One of the goals of this process is to provide a prediction of the outcome, without the time and expense. Ultimately you would want to validate the conclusions with results.

[Jill]

I dunno...guess I'm still confused.

Isn't that what we all do anyway when making recommendations for keywords? I just don't get all the extra math you have involved here. Is that just to make it look like something complicated so that it gets through to the client? Baffle them with BS, maybe?

I just don't get it! Someone gimme some more coffeeeeeeeeee!!!

Jill

[cline]

Jill,

Basically it's just a method for mathematically estimating which keyphrases have the largest volume of relevant traffic, which allows additional estimates to be built from there. One does, however, have to input judgments regarding the relevance of each term considered.

[DanThies]

Exactly - the idea is to move the "estimates" you have to make on relevance, to search terms that are less ambiguous. This is because the estimates people make on ambiguous/generic search terms are quite often far too optimistic.

How our clients use this, is when they are discussing keyword selection with their clients. If the client insists that they need a top ranking for a very generic search term, it's an easy way to demonstrate why that might not be the best approach.

[ghergich]

but the origional numbers are all a guess you could just say in my opinion interior design firm is a more relavant search than interior design. You don't need numbers since they are just you opinion to begin with. But I understand what you are trying to say...I say cut the numbers and just say it...then test your theory, only way to prove it for a fact

[DanThies]

A lot of folks like to just go with gut feel, seat of the pants, etc. What we're trying to put together is some better metrics than "guesswork." That means search volumes, competition, relevance, click-through rates, etc.

Inevitably, some folks will object to the methods and the metrics. We have experience applying these metrics profitably, so it's a bit more than a pet theory.

Testing will answer many questions, but testing costs. If you test with PPC, you're really testing something different than what you'll get with SEO. You can't really test SEO without doing SEO, link building, etc.

[RoadTrucker]

If I understand you correctly DanThies, your saying that since you estimate approximately 7% of your top ten search terms to be relevant to your website (i.e. Potential customers looking for your offer). You can extrapolate that same 7% to the generic search term, giving you approximately 300 of the total 4500 as clicks that probably are relevant. Depending on the results of individual terms you can get a good first estimate of whether that generic term can be profitable for you or not. As you collect real data, you can massage those numbers closer to reality. I am guessing you have decided (with or without evidence), that others NOT in the top 10 would NOT change that percentage by very much. (i.e. They would be irrelevant). I think this would useful with PPC too. Perhaps even more so. My sister used the term DVD the first time she used PPC and got hammered, since her DVDs were very specific. (i.e. They plugged into a cigarette lighter). Had she of done this and been way off, it probably still would have showed her, it was not worth the cost. I like it. You take a huge universe and give it some boundaries, even if you can't clearly see the boundaries, you still get some reasonable numbers to plug into your advertising budget.
One thing though, it would seem that since your top 10 only counted less than 1/2 of the generic search term, you might have missed some very relevant (or non-relevant) terms. Perhaps custom interior designs or interior design company. I would think this might possible change the numbers significantly. Is there some relationship between the total counts of the ten and the generic number of counts that could help you determine appropriateness? Perhaps a percentage.

[DanThies]

I am guessing you have decided (with or without evidence), that others NOT in the top 10 would NOT change that percentage by very much. (i.e. They would be irrelevant). I think this would useful with PPC too. Perhaps even more so.

To do the process right, you wouldn't just stop at 10 search terms. Basically, what you're looking for is all the ways that users have refined a generic search, because that tells you a lot about what they were looking for.

We did dig much deeper with the interior design terms, but posting an example that was going to confuse everyone anyway, I didn't want to put 88 search terms in there because they didn't change the number much anyway.

The assumption here, if you're looking for one, is that people are adding words to refine their search, and that the folks who try a completely different search wouldn't affect the outcome significantly. Reasonable assumption, in my opinion, since the results of this process have proved out well when applied to PPC advertising decisions.