For a test between two ads, switch the campaign's ad rotation from the default, which favours the ad AdWords expects to win, to rotate evenly or indefinitely, then let both run for several full weeks.

One ad ran, the other waited

A moving company wrote two versions of its AdWords ad in June: one leading with its hourly rates, the other with its years in business. The plan was to keep the better one.

Eight weeks later the owner opens the Ads tab. The rates ad has nearly all the impressions. The other has a few dozen, a single click and no calls at all.

Nothing was wrong with the second ad. AdWords decided early, on very little evidence, which of the two to show, and the test never really took place.

What the default rotation does

When an ad group holds more than one ad, AdWords has to choose which one enters each auction. The ad rotation setting, under each campaign's advanced settings, tells it how to choose.

New campaigns start on the setting that optimises for clicks: AdWords predicts which ad will earn the higher click-through rate and shows it more often, sometimes almost exclusively.

The prediction can come early, from a handful of clicks. Once one ad pulls ahead, it takes nearly all the impressions, so the other never gathers the data that might change the verdict.

The four rotation settings

The rotation setting currently offers four choices:

The first two are fine when you are not testing. The last two are the ones that let a test happen.

Even rotation does not mean identical numbers. Each ad still competes on its own quality in every auction, but both get a fair chance to show.

When to rotate evenly

Switch to even rotation whenever you are deliberately comparing ads. The setting covers the whole campaign, so every ad group in it rotates the same way.

Rotate evenly suits most tests, since few small advertisers need longer than 90 days to decide. Adding or editing an ad starts the 90 days again, so a new test gets a fresh period.

Rotate indefinitely suits advertisers who want to judge every result themselves and never hand the choice back. Someone then has to look, decide and pause the weaker ad.

How long a fair test needs

Small accounts produce small numbers, and small numbers jump about. Three clicks against two can reverse in a single afternoon, so an early lead tells you almost nothing.

Let each ad collect at least a few hundred impressions and several dozen clicks before judging. For many local businesses that means four to eight weeks, sometimes longer.

Run the test across whole weeks, and for a mover across at least one month end, when many people move. An ad that wins only on quiet Tuesdays has not really won.

Judge by enquiries, not only clicks

Click-through rate shows which ad people tap. It does not show which ad brings jobs. The ad promising the lowest rate may win clicks from people who never book.

Count what happens next: calls, quote requests and bookings. Conversion tracking can record website forms, and staff can note which callers came from a search ad.

Change one thing at a time. If the two ads differ in headline, offer and display address all at once, the result cannot tell you which change mattered.

A worked example

A cargo bike delivery service in a dense city neighbourhood carries parcels, flowers and catering trays for local shops and offices, and runs a small AdWords campaign on delivery searches.

In the spring its owner wrote two ads, one promising delivery within the hour and the other a flat neighbourhood price. A month later the second had shown a few dozen times.

The campaign was on the default setting. AdWords had favoured the speed ad within days, after a handful of clicks, and the comparison the owner wanted had never started.

The campaign was switched to rotate indefinitely, both ads were left exactly as written, and the owner set an end date six weeks out, long enough to include busy and quiet weeks.

Calls and website quote requests were counted alongside clicks. At the end, one ad led on enquiries as well as taps, so it stayed, and a new challenger replaced the other.

Reading the results in the Ads tab

In the Ads tab, set the date range to the test period and compare impressions, clicks, click-through rate and conversions for each ad, side by side.

If one ad still has far fewer impressions than the other, check the rotation setting again, and check that the ad was not edited or disapproved partway through.

Testing in a small account

With a modest budget, test two ads at a time, not four. Every extra version divides the same few clicks and stretches the time each needs to show anything.

Run tests in the ad groups that get the most searches. A test in an ad group with ten impressions a week will still be running next spring.

And write the result down: the dates, the two texts and what you counted. Six months later, nobody remembers why the current ad won.

The counter-case

When you are not testing, the default is reasonable: you want the stronger ad shown more often, and AdWords usually judges well which one people click.

In a large account with thousands of clicks a week, its predictions rest on plenty of data, and many advertisers leave the default on between tests.

Even rotation also has a cost: the weaker ad keeps its share while the test runs. That is the price of an answer, and short, focused tests keep it small.

Setting up a fair test

  1. Write two ads that differ in one thing only.
  2. Open the campaign settings and find ad rotation.
  3. Choose rotate evenly or rotate indefinitely.
  4. Set up conversion tracking for quote requests and calls.
  5. Pick an end date several full weeks away.
  6. Compare enquiries, not just clicks, at the end.
  7. Pause the weaker ad and write the next challenger.

Step three decides whether a test happens at all. Without it, AdWords picks the winner before you have collected anything worth comparing, and the second ad waits unseen.

An ad that never shows at all may simply have been disapproved, and ad text that AdWords will not approve shows how to read the reasons.

Frequently asked questions

Why does one of my test ads get almost no impressions?

The campaign is probably on the default rotation, which favours the ad AdWords expects to get more clicks. It often decides early, so the other ad barely runs.

Where is the ad rotation setting?

In each campaign's settings, under the advanced settings for ad delivery. It applies to every ad group in that campaign, so check it before a test.

How do rotate evenly and rotate indefinitely differ?

Rotate evenly shows ads more evenly for 90 days, then starts optimising. Rotate indefinitely keeps rotating with no time limit until you change it.

How long should a test run in a small account?

Until each ad has a few hundred impressions and several dozen clicks, across whole weeks. For many local businesses that takes four to eight weeks.

Should the winner be chosen by click-through rate?

Not by that alone. Count the calls, quote requests and bookings each ad brings, because the ad people tap most is not always the one that brings work.

Is the default setting ever the right choice?

Yes, when you are not testing and simply want the stronger ad shown more often, or in large accounts where its predictions rest on plenty of data.

West Coast Media Solutions Inc. provides web design, web development, hosting, digital marketing, and business consulting to organisations across Canada, drawing on more than twenty-five years in the field.

Running two ads and seeing only one?

Send us the campaign name, and we will check its rotation setting and suggest how to set up a test that actually runs.

Start a Conversation