How to reduce false positive link clicks in email statistics?

Ask a Question
1

A lot of email clients click automatically on the links on any email for security purposes.
They do usually click all links so it should be easy to filter then out from the email statistics.
Is aweber doing anything about this?
Can you detect bots for us?

I really want to send an email to everyone who has sent a specific link and did not click any other link in an email. Is that possible?

Here’s GPTs take on this topic:

Yes, there are a few ways to filter out automated clicks and focus on real user interactions, though it’s challenging to be 100% accurate. Here are some strategies you can use:

Identify Click Patterns by Timing: Automated clicks often occur within seconds or a few minutes after the email is sent, whereas real users usually take longer to open and click on links. You can filter out clicks that occur very quickly after sending.

Check User Agent Data: Some automated link checkers have unique user agents or IP addresses. By analyzing the user agent string or IP address of each click, you can filter out known bot identifiers. This requires maintaining a list of user agents associated with email security tools.

Analyze Click Frequency per User: Automation tools often “click” every link in an email simultaneously. If you notice multiple link clicks from a single user within a very short time frame, these are likely automated clicks. You can ignore these when calculating engagement.

Link Obfuscation for Sensitive Links: In cases where specific links shouldn’t be checked by bots, consider using a method to obfuscate or slightly alter the URL, making it less recognizable as a clickable link to automated scanners. This could be done by requiring a form submission or by using link shortening tools with user verification steps. However, this may reduce ease of access for legitimate users.

Request for Verification Clicks: If you’re sending links that require user actions (e.g., confirmations), use a secondary page that users need to click through after the first link. Bots generally won’t complete this secondary action.

Email Tracking Solutions with Bot Filtering: Some advanced email tracking services offer bot-filtering features, leveraging data patterns and known bot behavior to filter out automated clicks. Look for services that specify anti-bot capabilities in their tracking.

Each method has its limitations, but combining them can give you a clearer picture of genuine engagement.

2

Hi Matteo,

The market is filled with a lot of fearful misinformation on the subject of bots. Bot clicks in email content, particularly clicks that happen on ALL links are an extremely small part of overall engagement activity. Most of that activity happens on tiny corporate domains that don’t have many subscribers in the typical users list. When your lists consist of 70%+ of Gmail, Yahoo, etc domains, you’re not going to see enough bot clicks for it to have any meaningful impact on your stats.

  • click patterns by timing: Many automated clicks do occur within seconds or minutes of message receipt. However, many real users also click within seconds or minutes of message receipt.
  • user agent data: Many privacy focused mail box providers also proxy this information these days making it impossible to distinguish between real users and automated bots.
  • click frequency per user: This is the most clear cut one at detecting bots, but the number of instances where this occurs is vanishingly small in the scheme of overall system wide engagement. It’s noise and not relevant to most senders.
  • link obfuscation: Making links harder for bots to recognize also makes it harder for humans to see and use. Not something I would ever recommend.

At the core of it here, I would recommend thinking about why specifically you want to detect bots? What core problem are you looking to solve? The follow up question you had to “send an email to everyone who has sent a specific link and did not click any other link in an email” is not impacted in any way by bot engagement. Sending to that segment would not send to bots. Even if it did send to bots, the bots are typically security scanners that are validating that an email is okay before passing it off to a human. You still want the human that owns that mailbox to receive and be able to engage with your message.

I really want to send an email to everyone who has sent a specific link and did not click any other link in an email. Is that possible?

Yes, you’ll want to use the subscriber search to create a segment matching the relevant links from the message in question.

2024-10-28_10-21-37
2024-10-28_10-21-37444×152 10.2 KB

KB: Create a segment.
KB: Send to a segment.

2 Likes
3

Hi Tom,

nice to meet you.

This is a wonderful answer and it does solve the problem.

I would love it if my marketing automation software didn’t require me to create queries on my own and just told me clearly:

Matteo, from your last campaign you will see a total of 103 people clicked on your main CTA link. We think that only 6 of these clicks are unique clicks.
We strongly recommend you follow up with these 6 subs. We have create a segment to make it easy for you.

These are actual numbers by the way.

And again, for emphasis: marketing automation, not database queries. Thanks for your kind attention.

Best Regards,
Matteo

4

The quickstats report that’s available after you send has a two click segment creation where all you have to do is give it a name and the segment is made. No queries are required for that, just point and click.

2024-10-28_12-39-53
2024-10-28_12-39-532362×1026 226 KB

5

Hi Tom,

many thanks for the reply.

When I do that I am back at 103 people + bots who clicked my main CTA link. I want the 6 that actually did.

Do you see the problem?

On one tab I have a conversation with Socrates and Marcus Aurelius via a LLM. On the other I am inputting queries to exclude bots via Aweber. There’s a bit of a disconnect there.

It’s your job to figure out who TRULY clicked on that link and tell me: Hey Matteo, you gotta follow up with these 6 smoking hot leads, they’re ready to buy. Hurry up.
Instead of creating auto-segments where 95% of the people NEVER clicked the link.

I am sorry if this message is more colorful and less factual than it needs to be. I like aweber and I am writing this to you to suggest how you can do better in the future.

You know better than me how to run your business. But I feel you’re falling short and underserving me.

Cheers,
Matteo

6

Hi Matteo,

There’s a bunch of layers to this, certainly…

Automation to automagically pick out:

  • your primary call to action link.
  • filter out any possible bot clicks.
  • make a segment automatically of just those users.
  • then notify you in some way that they should be followed up with.

Does every message you send require that kind of analysis and follow up? It doesn’t seem like you’re doing this sort of follow up on past sent messages. How would an automated system like you describe know that you want this analysis versus when you don’t?

LLMs are pretty amazing tech, we’ve embedded them in a number of places inside AWeber for both message content creation, message refinement, and subject line suggestions. We’re continuing to explore their usage where they are appropriate given their known propensity to hallucinate and give clearly invalid responses. It’s important to remember that LLMs are not reasoning or logic models. They are putting words together that show up in proximity to one another in their training data. Thee tech is advancing quickly and you’ll definitely be seeing more of their use inside AWeber where appropriate.

I would suggest submitting a feature request so that yourself and other community members can vote on ideas that are important to be added.

1 Like
7

Hi Tom,

I am sorry I won’t really be able create a feature request. I would hope your team would pick up on this scenario and just be extra interested in it.

Right now all the reporting on email clicks is off because of bots. It should be a priority to fix it.

The fix I see is pretty simple.

Let’s say that my email contains 2 links A and B.
If people click both A and B I would assume they are bot. If they only click A or B separately I would assume they are human. And would create a segment of Clicked A not B, Click B not A.

Now when you add 3 or more clicks the situation is similar, only you add more options. People can click A, B, and C separately, ABC together, AB, AC, BC. I would just make sure that for each link this work is done so to approximate the number of actual clicks for each link.

I assume you would want to do something like this quickly: it gives everyone who has an aweber account real data on what is actually going on.

For me: in the coming months I might shop for a different service that has this built in. Actually I will send them this thread and see what competitive email services think about it.

Cheers,
Matteo

P.S.
GPT has a reasoning model :wink:

8

Let’s say that my email contains 2 links A and B.
If people click both A and B I would assume they are bot. If they only click A or B separately I would assume they are human. And would create a segment of Clicked A not B, Click B not A

In your particular account right at this moment that might be the case, but it’s not the case across all users and even in your particular account it’s not the case. If you get clicks from an interested user on both links, your model would now classify them as a bot.

Most bot detection is far from perfect or downright misleading. Many places will claim to have it, I know how many of them work, and many are blatantly wrong. The way that many providers are proxying user engagements to enhance privacy makes it impossible to accurately identify human vs. non human interactions.

Language learning models do not reason like humans do. They identify patterns they’ve seen before and apply them to the information they are given. They frequently hallucinate and give blatantly bad information to users. When we’re building a product for hundreds of thousands of users, we have to take into consideration both the pros and cons of technology and the responsibility we have towards users.

9

I would much rather have those 6 hot leads (+ or - the margin of error), than the illusion that 103 people clicked.

I value a partial truth more than a blatant lie.

You don’t need LLMs to implement something like that.

I will shop around for alternatives, in the meantime I will have to resort to “clicked this hasn’t clicked that” by hand (which you suggested) and that is cumbersome but effective in figuring out manually who is really engaged.