The Real Relationship Between Search Data and Actual Clicks

When you’re doing keyword research, you need some way to predict the number of visits you’re likely to get for different search terms.

Usually, the method I’ve used has been to take certain percentages of the phrase-matched search stats (i.e. what Google AdWords Research Tool will give you). Why? Because I once heard somewhere that phrase-matching was the most reliable.

I’ve been wanting to test this assumption for a long time.

Here are the latest stats I’ve been using to give the proportion of clicks you’d get at each of the top 10 positions in Google:

1. 34.35%
2. 16.96%
3. 11.42%
4. 7.73%
5. 6.19%
6. 5.05%
7. 4.02%
8. 3.47%
9. 2.85%
10. 2.71%

So you could get about 1/3 of the searches in clicks. But what kind of searches? Broad or phrase? And is this real? Is it reliable?

These data are taken from large analyses of masses of search data, so they’ll represent the average over a very big sample.

Should you and I base our assumptions on this when we’re estimating traffic?

A Simple Test

• I exported about 15 of my top search results from Google Webmaster Tools.
• I then looked up the number of monthly worldwide searches, in all 3 match types, in Google AdWords Keyword Tool.
• Then I took the actual number of clicks I got from searches including each phrase from Google Analytics.

This allowed me to plot the percentage of clicks each term got, compared to the total broad/phrase/exact searches for that term per month.

The first thing I noticed was that the graphs were all over the place! (You can’t see that below.) I was using the “Avg. position” column from Webmaster Tools, but now I think you just can’t trust this data. It’s not the first time I’ve tried something like this, and frankly I don’t know where Google’s getting this from, but it seems to have little relation to what I see. Maybe it’s a worldwide average.

So I switched to RankTracker (part of SEO PowerSuite – I recommend you take the trial), and quickly got the ACTUAL positions on Google for the terms. That made the graph look a bit more reasonable.

See the results below. Most of the results were similar between RankTracker and GWT, but there are a few really oddball exceptions. We’ll stick with RankTracker.

What This Tells Us

The graph shows you the percentage of clicks this site got in a month, compared to broad-match (blue), phrase-match (red) and exact-match (yellow) searches.

What I’m looking for is the smoothest correlation between rank position and one of the match types.

1. First, we can see that the lines don’t cross over. That’s right, because there should always be fewer exact matches than phrase matches, and fewer phrase matches than broad matches.
2. Next, there is a general trend downwards as ranking position increases. That also makes sense. The higher up the search results a link is seen, the more likely it is to get clicked. So far, so obvious.
3. Is any one of the match types more regular / straight / predictable than the others? I would say that Exact (yellow) and Phrase (red) are fairly smooth, while Broad (blue) is not very useful. The blue line has similar results at positions 1, 4 and 6, before tailing off. By comparison, the yellow line has a large (50%) variation for the 3 results at #4. The red line has less variation. So my conclusion would be that Phrase-match (red line) is indeed the most reliable predictor of actual traffic.
4. The other thing I need to take from this is that, since I have been using Phrase-matched results with the average click-through prediction data (which I listed above), I have probably been under-estimating the likely traffic, as my click-through rates don’t go much above 20%, while the average for position #1 is over 34%. (However, see below!)
5. There are a lot of other factors that can come into play. The appeal of your “search snippet” is critical (as I examine here and here). Also, there are some terms that just have more than one meaning. Ranking well doesn’t necessarily mean that the meaning you give a term is the same as what most people are looking for.

I have to acknowledge that this is a very small sample. Next, I’ll try one with more data.

More Data

These 3 charts show the click share for 66 terms – again from the analytics for this website.

We’re really still looking for the tightest distribution – or at least the most predictable distribution at the rightmost edge of the curve (indicating the optimal performance, discounting under-performing results).

What this shows me is that Exact match actually seems to be the smoothest. It has more consistent results right through the range.

There are fewer serious underperformers. Look at the bottom left of each graph. Exact match has no results on the bottom  - indicating 1% clickthrough rates – until about position #5, whereas Broad and Phrase have very low CTRs in the top 5 of the results.

Plus, it’s only Exact which gets near the 33% clickthrough rates that the global average gives us, so maybe we should be using Exact match data for predicting traffic.

What this tells me is that I should use Exact numbers from this point on, which will actually give lower predictions than I have been getting for Phrase-matched results. However, the clicks will be more reliable, which is what’s most important.

• Tweet