I'm going to tell you a story. A story about a metric in AdWords that people trusted.

People grew to love this metric, they told their bosses how it was doing, they made changes to their campaigns based on it, and they judged their performance on whether it went up or down.

But those people didn't see below the surface. Lurking under the superficially obvious meaning of the metric was a hidden dark truth: the metric wasn't just pointless, it was lying to them. 

That metric is Average Position, and I'm sure quite a few of you are guilty (if unintentionally) of taking it at face value.

Impression share

The first thing to know about Average Position is that it's strongly affected by your impression share. If your impression share for a keyword is near 100% then that's great. But if it's at all off that figure then your position metric will be skewing up.

Remember that you can't see impression share per keyword...

Your ad enters the auction on every search for which your keyword is eligible (barring budget restrictions), but some of the time it doesn't reach the first page.

In those cases you have been rotated off in favour of other advertisers. Your ad's position is still calculated, but if you aren't on the first page you don't get an impression. No impression means no impact on average position. 

That's a problem. If you spend 10% of the time in position three and 90% of the time in position 23, your average position will be three. It's misleading to you as a campaign manager about how your keyword performs. 

So unless your impression share is 100% your average position is showing a higher value than your true average.

 AdWords impression share by position

Standard deviation

The next important point about average position is that it's a mean average. That means it's the simple kind of average you learn about at school: the total sum of positions divided by the number of impressions.

Unfortunately it's not very informative.

If you spent every search in position three then your average position will be three. Which is great. But if you spent half the time in position one and half the time in position five, you average position will still be three. Suddenly your average is telling you a position you were never really in.

You may have spent 20% of the time in each of the top five positions. You'll still have an average position of three, but it's incredibly misleading about what's really happening.

A mean average without a standard deviation tells you almost nothing about what your ad is really doing. AdWords won't give you that data (GA will give it to you for your clicks, but not your impressions.

Since you can expect a higher clickthrough rate (CTR) in a higher position, this figure will skew upwards too).

 In the chart below, the vertical scale represents impressions per actual position on the page. All three of these distributions will give you an average position of four.

 Impressions per position


Any frequency distribution (e.g. a chart of the number of times you appeared in certain positions) will either be symmetrical around the mean average, or asymmetrical.

This means that it could be showing a lot of times in position two, then a few times each in all the lower positions. The average might be three, but enough impressions in positions higher or lower could mean that your ad's distribution looks more like a long tail than a normal distribution.

Again, you're going to get no data about this, so you still don't really know what positions your ads were appearing in.

In the image below the average position is 4.4, although positions two and three have by far the highest occurrence of impressions.

 Average skew

Top vs other

Now we get down to what's important: it's crucial that you appear in the top three positions, the banner positions. You can expect your CTR to be 10x to 15x higher in these positions than on the right hand side. That's a 1,400% increase!

If your average position tells you that you are in position 2.5 and you take it at face value, you might believe that you are in the banner most of the time.

But as we've seen, this doesn't really tell you much. We already know that the average position is skewing up if we don't have 100% impression share, and we have no idea what our spread of impressions really looked like. 

The top vs other segment comes to the rescue. Top vs other is almost the metric that Average Position is a proxy for.

The top vs other segment can be applied at any level in your campaign, and lets you see the proportion of time you spent in the banner vs the right hand side. This is absolute gold for knowing how your keywords are doing.

If your ads for a specific keyword appear in the banner only 40% of the time, you know that 60% of your impressions could have happened with that higher CTR, if only you'd increased the bid.

So from now on, you're bidding for top vs other proportion, not average position.

Alistair Dent

Published 26 March, 2012 by Alistair Dent

Alistair Dent is Head of PPC at Periscopix at Periscopix and a contributor to Econsultancy. You can connect with Alistair via LinkedIn, or follow him on Twitter and Google+.

5 more posts from this author

You might be interested in

Comments (5)


Vibhanshu Abhishek

A very interesting article about the fallacies of average data, but your article just touches the tip of a much bigger issue. Daily Adwords reports do not only "lie" about position, but they also misrepresent other metrics like click-through rate (ctr), cost per-click (etc) which are important factors that affect the bidding decisions. You correctly point out that there is significant variation in the position of an during a day, but this variation is not observed by the advertisers. For example, if the ad showed up at position 1 for 50% of the impressions and position 5 for the other 50%, all the advertiser knows is that the average position of the ad was 3. In a paper (http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1490169) with Profs. Fader and Hosanagar from the Wharton School, we show that
this variation, coupled with the non-linear nature of the ctr-position curve (which usually follows the pattern of a power law or exponential curve), gives rise to what we term as aggregation bias. If the ctr at position 1 is 4% and the ctr at 5 is 0.4%, the daily summary would say that for an average position of 3, the avg. ctr is 2.2%. Not only was the ad never shown at position 3, the ctr at position 3 is grossly overestimated. The problem arises because of Jensen's inequality which predicts that mean of ctr at different positions is greater than the ctr at the mean positions. One would assume that the advertisers are worse off due to this miscalculation of ctr, but surprisingly, a careful game-theoretic analysis reveals that the search engines tend to lose a lot more due to this aggregation bias. Our analysis shows that search engines are losing as much as 17% of their revenues due to this issue. Better data standard in SEM are the need of the hour, and they would help not only the advertisers but more importantly make search engines like Google, Yahoo and Bing more profitable.

You can read the complete paper here: V. Abhishek, K. Hosanagar, P.S. Fader, "On Aggregation Bias in Sponsored Search Data: Existence and Implications" -

over 6 years ago


Company Rescue

Remember the first 3 rules of adwords

1. Google wants my dollar
2. Google wants my £
3. google wants my money

over 6 years ago



I wanted -just for today- to try to be a bit of the devil's advocate.
Even if everything said here makes sense, there could be a couple of exaggerations.
First I don't see that "Lost Impression Share due to Rank" is mantioned anywhere. Whith it you have a clear metric to know how many of your impressions went down below 2nd page.
2nd The Average position, is also no a lie. Is what it is, an average, and is great to have an article like this pointing out the dangers on taking easy, superficial decisions based on Average positions. The analysis is good, but not a prove of anybody lying. How would we be able to manage an account without an average position metric? Even if it "lies" I'd be willing to know a better alternative.
In the adwords interface you can read Avg Position on a search query level (something very, very few people get bothered to look into) This metric is more realistic, but if we should complaint, I agree it tells you only half of the story: only if clicks are involved metrics are showed (no clicks, no search query data). So we can never know which position all impressions had, and we can only trust the Keyword-level metric.
But we do can know on which position our Keywords were clicked, at leats -always-.
About a year ago I couldn't find any critic voice against the positioning control Adwords deprecated. This tool (even if slapy) would let you just bid higher while keeping a controlled position and control the positioning and money spend. Now they want you to use automated rules for this, but the rule acts in a reactive way: you can only adjust based on what happened yesterday, or last week, not on what happens on the fly.
Interesting subject, thanks!

over 6 years ago


Tom Heffernan

Awesome content! Just as a footnote, you can take an ad group down to a single keyword and then use the dimensions tab to view average position by hour, if you are really interested in that detail of information. Question: If beauty is in the eye of the beholder then is a lie on the tongue of the teller or in the ears of the listener?

almost 6 years ago



I find adwords works really well

over 5 years ago

Save or Cancel

Enjoying this article?

Get more just like this, delivered to your inbox.

Keep up to date with the latest analysis, inspiration and learning from the Econsultancy blog with our free Digital Pulse newsletter. You will receive a hand-picked digest of the latest and greatest articles, as well as snippets of new market data, best practice guides and trends research.