The most dangerous analytical mistakes in Google Ads tend to have something in common. The data looks straightforward at a glance, but behind that surface number, there’s some unseen variable, subtly altering its meaning.
Here are five that easily catch us out:
1. Hidden sub-groups
A comparison in Google Ads can look perfectly clear at the top level, but completely reverse once you break the data down.
Take ad CTR. One ad looks like it’s underperforming another. But apply the top vs. other segment and you sometimes find the ‘weaker’ ad actually has a higher CTR in both top and other positions. It’s not underperforming – it simply has a higher share of impressions in other positions, where CTR is naturally much lower, and that drags the aggregate down.
📹 Watch Microtip #25 on YouTube
The same thing can happen with device mix. Two ads look meaningfully different on conversion rate, but segment by device and both are performing similarly on desktop and similarly on mobile. One is just attracting a disproportionate share of mobile traffic, where conversion rates are weaker across the board. The apparent difference between the ads is being driven by traffic mix, and the relevant disparity in performance sits with device, not in the ads themselves.
📹 Watch Microtip #1 on YouTube
It happens at the campaign level too. Search Partners traffic almost always behaves very differently from Google Search. If you’re comparing campaigns or time periods without segmenting by network, the blended view can hide the true story.
In statistics, this pitfall is called Simpson’s Paradox – where a trend or comparison reverses once you break the data into its constituent parts. It crops up more often in PPC than most of us realise. Whenever a comparison looks clean at the top level, it’s worth asking whether there’s a sub-group you haven’t checked…
2. Be careful with date ranges
Many a mistake in analysis is made through carelessness with dates.
The most familiar version is comparing months of different lengths. February is nearly 10% shorter than January – so a drop of less than 10% in conversions or revenue isn’t a decline at all. In relative terms, it’s actually an increase. Nobody should need to explain a modest month-on-month dip in volume when the explanation is simply that it’s February!
But there’s a subtler version that’s just as important. Weekends often see very different behaviour from weekdays. When comparing two date ranges, if one contains more weekend days than the other, or happens to end on a weekend, the differing proportion can create the misleading appearance of a rise or decline that isn’t really there.
(I often go back to this neat example from 2020…)
This is especially relevant for B2B accounts where weekday traffic carries most of the conversion volume. A month that sneaks in a fifth weekend can look weaker than its predecessor for no strategic reason at all.
Before attributing a period-on-period change to anything you or the market has done, check whether the calendar explains it first.
3. Matched locations vs actual locations
Google Ads has two location dimensions: targeted locations and matched locations. Given those names, it would be reasonable to assume that ‘matched locations’ tells you where users actually were when they clicked. It doesn’t.
Matched locations show which of your targeted locations Google matched the user to – either by physical presence or by interest. So someone sitting in another country can show up under one of your UK locations if Google ‘matched’ their search intent to your target area.
If you want to know whether people outside your intended areas are triggering your ads, you need the ‘user location’ dimensions in the report editor instead, which show where Google believes the user physically was at the time.
📹 Watch Microtip #8 on YouTube
The problem of off-target users has always been a common one, and it used to be easy to check within standard reports. While the diagnosis is now fiddlier, the solution is the same: make sure the location setting is switched to ‘presence only’.
4. The denominator problem
Any metric with ‘divided by’ in its formula is vulnerable to this, and it’s easy to be caught out by it.
I wrote about this recently when impression share tanked on one of my accounts after enabling AI Max. In the auction insights report, every competitor’s share dropped at the same time – which was the clue.
AI Max had expanded the pool of eligible auctions such that the total pie grew, and even though our slice hadn’t shrunk in absolute terms, impression share went down while actual impressions went up.
The same logic applies across Google Ads. Expanding match types or enabling Search Partners is likely to gain clicks but lose CTR. Lower ROAS often goes hand in hand with higher revenue. A lower conversion rate can coincide with more conversions if you’ve adopted one of Google’s many expansive features.
It may look like these metrics measure performance – but more accurately, they measure performance divided by opportunity, where opportunity is the ghost in the machine…
📹 Read the full blog post on the denominator problem
5. The conversion blind spot
Google Ads normally assigns conversions to the time of the click/s that led to them, not the time when the conversion actually happened. For longer-term analysis that’s the right approach, but it creates a blind spot in the short term that can look alarming if you’re not expecting it.
If you’re checking how many conversions came in yesterday, the standard conversions column can often look almost empty – not because performance has fallen off a cliff, but because the conversions from recent clicks haven’t happened yet. When they do, they’ll be backdated to whenever those clicks occurred.
📹 Watch Microtip #13 on YouTube
The answer is to use conversions by conversion time, which shows when conversions actually happened regardless of the time of those preceding clicks.
There’s a second application worth knowing too. If you’re ever worried that tracking might have broken because recent conversions look suspiciously thin, check conversions by conversion time for the last few days. If conversions are appearing there, tracking was working when they came through – the standard column just hasn’t caught up yet. A simple column switch that can save that unnecessary panic.
None of these are hard to spot once you know to look for them. But they’re all easy to miss if you don’t… There’s often an important layer underneath the surface-level number – and a quick segmentation or sanity check can be the difference between a good decision and a confident mistake.
