Click fraud detection in FouAnalytics

Can humans complete an online lead form without moving the mouse to the first text field, without clicking the text field to place the cursor, and without typing their responses in those text fields? Of course not.

Should you count accidental clicks or clicks that are faked by bots? See the following thread on Reddit, where they are still debating this. Shout out to Outrageous-Month-355 for asking the question and posting their observations: https://www.reddit.com/r/programmatic/comments/1eoazfr/rejecting_the_notion_that_clicks_are_accidental/

So many clicks are outright faked, but you've never seen that because legacy fraud verification vendors don't show you any details. Here's what you can "see Fou yourself" if you use FouAnalytics to measure your ads and your sites.

FouAnalytics in-ad measurement

Can humans click an ad without touching the screen on their smartphone? Of course not. The click charts below show 3 screen resolutions with clicks and corresponding touch events. This helps you visually corroborate.

In the following, you can see "click -touch-exists:1" (click, minus touch-exists:1). It is obvious and makes common sense that clicks without touch events on mobile cannot be from real humans. You can also see clicks on desktop/laptop -- 1366x768 and 1220x1280. Those are color coded orange (declared bots). So in FouAnalytics, there are multiple ways to confirm those clicks are NOT from humans.

You all have already seen the following. In the first four 300x250 ads, the green clicks concentrated in the upper right corner show you that users were trying to CLOSE the ad, not click the ad. For contrast, the blue clicks in the five 300x250 ads below that were spread out across the face of the ad. These users were trying to click the ad, and we can see what was clicked as well -- adclick.g.doubleclick .net. That allows you to confirm that these are clicks that should be counted towards the success of the campaign.

FouAnalytics on-site measurement

Similarly, when on-site FouAnalytics tags are added to the websites and landing pages, we can look at clicks, mouse move, touch events, and page scroll. Similarly, humans have to touch the screen on their smart phones in order to scroll or click something. We can see this in the data below for mobile:1. Between 79 - 81% of the users touched the screen, and between 66 - 69% of the users clicked something. The percentage of users that clicked must be lower than the percentage that touched the screen. And it is.

For funsies, can you tell from the touch charts above, which hand the user was holding the phone in? If you have the answer, write it in comments.

In the click charts below, you will see two desktop screen resolutions -- 1920x1080 and 1366x768. And you will see two mobile screen resolutions -- 360x640 and 375x667. You will see that clicks are spread out across the page, but there are some natural clustering of clicks because humans usually click on site navigation, buttons, links, etc. These clusters appear in roughly the same locations when comparing the 2 desktop screen resolutions and the 2 mobile screen resolutions.

The fifth click chart in red, lower right, shows entirely different looking clicks -- large circles. Large circles in FouAnalytics means repeated clicks on the exact same pixel (x,y coordinate) on screen. You can't get a bunch of humans to repeatedly click the same pixel. But it's trivial for bots to repeatedly click on 0x0 or any other pixel for that matter. This is what it looks like in FouAnalytics -- 1) data table on the left, 2) click chart on the right. Larger circles reflect the larger number of clicks on the same x,y coordinates.

The way you use this data for your digital campaigns is to see which paid digital channels send more dark blue clicks (humans) and of those, which users are more attentive. In the trio of charts below, you will see 1) users coming from utm_source=google (google search) are very dark blue (humans) and very attentive (79 - 83%) clicked something on the landing page. 2) users coming from TikTok are 43% dark red (bots) and only 1 - 3% clicked anything on the landing page (very low attentiveness). And 3) users coming from PMax (Google's "performance max") were 84% dark red (bots) AND these fake users also faked the clicks on the landing page to simulate engagement. Note the strange patterns of click locations.

So what?

Given the above, let me answer this poor guy's question about clicks being accidental.

1. "site retargeting consistently drives higher CTRs." That's because most of those clicks are from bots. The bots deliberately visit the advertisers' websites first, to collect retargeting cookies. Then they go to cash-out sites to cause the ads to load (those sites make ad revenue, hence the name "cash-out" sites). The bots also click on those retargeted ads to make ad buyers think retargeting is working, so the invest more money and more money goes to said bad guys. This is why "site retargeting consistently drives higher CTRs." See the color coding under the retargeting and remarketing rows in the 2 data grids below.

2. "high net worth and high income audiences display higher interaction rates" That's because bots visit magazine sites targeted to high net worth individuals and they look at expensive products. This tricks the data brokers into adding those cookies (bots) into audience segments they think are high net worth individuals. The higher interaction rates are from bots, not high net worth individuals interacting with the ads or sites. High net worth individuals have their assistants buy stuff and they are likely to be using ad blockers so they are not bothered by so many micro-targeted ads.

3. "lower funnel creative messaging gets more clicks than brand awareness messages" makes sense, nothing to add.

So I reject the notion that clicks are accidental, not because the clicks are real, but because the clicks are definitely and deliberately faked, not accidental.

So "Why do people constantly say that clicks are not a valuable source of data?" Because SO many folks have seen FouAnalytics data over the last 13 years or have actually used FouAnalytics to know that clicks are easily faked by bots AND they have the visual evidence like the above to know that it's rampant click fraud and not accidental clicks.

Clicks ARE a valuable source of data if you can tell apart which clicks are real and which are faked by bots. In other words ...

Clicks are a valuable source of data only if you can "see Fou yourself" with FouAnalytics.

Note that you get none of the above analytics and details from legacy fraud verification vendors. Time to upgrade.

Further reading: https://www.dhirubhai.net/in/augustinefou/recent-activity/newsletter/