AdWords is now Google Ads. Our new name reflects the full range of advertising options we offer across Search, Display, YouTube, and more. Learn more

Ads
4.3K members online now
4.3K members online now
Improve your Google Ads performance and boost your ROI, CTR, and Quality Score
Guide Me
star_border
Reply

How to split test landing pages correctly?

Follower ✭ ✭ ✭
# 1
Follower ✭ ✭ ✭

I want to split test another version of my current landing page.

 

my current ad leads to mydomain.com/page/

 

Shall I simply duplicate my ad, and only change the destination url to mydomain.com/page2/,

and have the ads rotating evenly?

 

1. does it matter if the second page would be /page2/ or is it better to have it with a parameter such as /page/?version=2

 

2. since the current page and ad have a history in adwords (which might be also effecting the QS?), wouldn't adding a new ad with a new destination page, would cause the experiment not be "clean" enough, since a the new page doesn't have any "adwords history" and therefore might underperform?

 

3. anything else I should take note of?

 

Thanks

1 Expert replyverified_user
1 ACCEPTED SOLUTION

Accepted Solutions
Marked as Best Answer.
Solution
Accepted by MosheTLV (Top Contributor)
September 2015

Re: How to split test landing pages correctly?

Participant ✭ ✭ ✭
# 6
Participant ✭ ✭ ✭
Hi Jay s,

There is another way to do it, namely with campaign experiments.
learn more here:
https://support.google.com/adwords/answer/2375383?hl=en

That being said the 'problems' you envision are still applicable to the campaign experiments. But I will explain here why I believe this is not a problem at all.

Consider the situation in which you duplicate an ad and only change the destination URL, with rotation settings being set to evenly (the method I personally use). Your QS in this regard is calculated every time your ad enters an auction and is dependant on the formula including which ad is used. Now if the new landing page has a negative impact on QS this does not have a negative effect on your experiment, In fact it is relevant to your experiment.

Every change you make could have an impact on QS, so why are you trying to make an experiment that equalises QS between the 2 versions? It is an influencing factor on the performance of the test and is thus relevant to the test. If a landing page is less relevant or more relevant to the keyword this is something you need to consider when running the experiment.

You seem to be mostly concerned with the established QS that is achieved (the number you are shown) but this number is only a representation and not the actual QS in each auction. Remember that QS is based on account history as well, so any new additions to your account usually start with a QS in line with your overall account so even from this point of view the effect of a lower QS is minimised.

Furthermore with respect to an established QS, your experiment should be running for more than long enough that the new ad is able to normalise its QS. If it does not, then you are not running the ad for long enough. The first thousand or so impressions that it takes tio establish a base QS for the new ad should not have a statistically significant impact on the results of your experiment given an adequate sample size.

Consider the 2 options. Either your new dest URL performs worse or better. If the result is significant and you have statistical confidence then the apparent negative impact of initial QS only reinforces that. In the case of the new landing page being better, the fact that the first few results were worse than the average means that if you exclude that your result will be more significant. And the same applies to if the ad is worse as the 'worse' state of the new add will be more worse than the normalised QS initially assigned to the new ad.

With respect to the ad rank and 'junk' clicks. You can take that into consideration when looking at the data as there should be no significant difference in CTR between 2 identical ads. If there is an apparent significant difference then either, A there is a large difference in ad position that is having the effect or B your experiment, while showing a significant result has not taken into consideration population size and thus your results lack confidence and you need to run the experiment for longer.

Situation B is just to run the experiment longer. Situation A can be dealt with as you best decide, but if the experiment has been running long enough for significance than I would suggest that you decide what to do on a holistic approach. Decide which Landing page is best for ROI and go with that.

Hope this helps

View solution in original post

Warning: AdWords interface instructions or screenshots in this thread may be out of date as we move to the new AdWords experience.

LEARN MORE

Re: How to split test landing pages correctly?

Rising Star
# 2
Rising Star
Hi Jay,

There isn't a feature as such, but yes you can create identical ads within an ad group and with different landing pages and set the campaign to rotate evenly. You can use parameters (like version 2) for clear reporting, or simply go with page2 if you're able to identify the URL in reports when you pull them later.

Yes, landing page that you introduce wouldn't have a history and hence would have an effect on quality score.

Cheers,
Sumanth
Sumanth Sridhar

Re: How to split test landing pages correctly?

Follower ✭ ✭ ✭
# 3
Follower ✭ ✭ ✭
If the QS would be lower for the new page, all the results can be distorted...
for example, lets say my current ad at the top positions (#3 ), so I am getting a high CTR yet low conversions.
then, a new identical ad would be at least for the first days at #4, so I would get a lower CTR for it, yet conversion rate might be higher since it's legitimate to assume that if someone clicks on a #4 position ad, then he must be much more interested in my offer, as people who click on #1-#3 top ads might be doing this more by "default" so conversions can be lower.

In such case, I might wrongfully conclude that the new landing page increased conversions, while in fact it was a lower positioning which can be achieved be lowering my current bids.

Am I right? any way to deal with this?

Re: How to split test landing pages correctly?

Rising Star
# 4
Rising Star
I agree to that point Jay. If your primary objective is to see which version of the LP is performing well, you can create a duplicate of the original LP and test it against the new one.

- Sumanth
Sumanth Sridhar

Re: How to split test landing pages correctly?

Participant ✭ ☆ ☆
# 5
Participant ✭ ☆ ☆
Hi Jay S

To add to Sumanths excellent points and your own excellent points, I think it may be possible to use content experiments for this exercise. With Content Experiments, you can test which version of a landing page results in the greatest improvement in conversions (i.e. completed activities that you measure as goals) or metric value. You can test up to 10 variations of a landing page.

Take a look at content experiments within Google Analytics for more info on this.

https://support.google.com/analytics/answer/1745152?hl=en

All the best and Good luck!

James Edward
Marked as Best Answer.
Solution
Accepted by MosheTLV (Top Contributor)
September 2015

Re: How to split test landing pages correctly?

Participant ✭ ✭ ✭
# 6
Participant ✭ ✭ ✭
Hi Jay s,

There is another way to do it, namely with campaign experiments.
learn more here:
https://support.google.com/adwords/answer/2375383?hl=en

That being said the 'problems' you envision are still applicable to the campaign experiments. But I will explain here why I believe this is not a problem at all.

Consider the situation in which you duplicate an ad and only change the destination URL, with rotation settings being set to evenly (the method I personally use). Your QS in this regard is calculated every time your ad enters an auction and is dependant on the formula including which ad is used. Now if the new landing page has a negative impact on QS this does not have a negative effect on your experiment, In fact it is relevant to your experiment.

Every change you make could have an impact on QS, so why are you trying to make an experiment that equalises QS between the 2 versions? It is an influencing factor on the performance of the test and is thus relevant to the test. If a landing page is less relevant or more relevant to the keyword this is something you need to consider when running the experiment.

You seem to be mostly concerned with the established QS that is achieved (the number you are shown) but this number is only a representation and not the actual QS in each auction. Remember that QS is based on account history as well, so any new additions to your account usually start with a QS in line with your overall account so even from this point of view the effect of a lower QS is minimised.

Furthermore with respect to an established QS, your experiment should be running for more than long enough that the new ad is able to normalise its QS. If it does not, then you are not running the ad for long enough. The first thousand or so impressions that it takes tio establish a base QS for the new ad should not have a statistically significant impact on the results of your experiment given an adequate sample size.

Consider the 2 options. Either your new dest URL performs worse or better. If the result is significant and you have statistical confidence then the apparent negative impact of initial QS only reinforces that. In the case of the new landing page being better, the fact that the first few results were worse than the average means that if you exclude that your result will be more significant. And the same applies to if the ad is worse as the 'worse' state of the new add will be more worse than the normalised QS initially assigned to the new ad.

With respect to the ad rank and 'junk' clicks. You can take that into consideration when looking at the data as there should be no significant difference in CTR between 2 identical ads. If there is an apparent significant difference then either, A there is a large difference in ad position that is having the effect or B your experiment, while showing a significant result has not taken into consideration population size and thus your results lack confidence and you need to run the experiment for longer.

Situation B is just to run the experiment longer. Situation A can be dealt with as you best decide, but if the experiment has been running long enough for significance than I would suggest that you decide what to do on a holistic approach. Decide which Landing page is best for ROI and go with that.

Hope this helps

Re: How to split test landing pages correctly?

Participant ✭ ✭ ✭
# 7
Participant ✭ ✭ ✭
Reading @James Edward G's answer made me think of other options as well. You could perform the split test outside of AdWords using 302 Redirectsl.

302 not 301 unless you want the new landing page to be indexed.

You could then send half your traffic to one version and half to another.

This would eliminate the problem of new ad creation and limit QS changes to only changes brought about by landing page experience.

Hope this helps