Bot Traffic Is Poisoning Your Search Console Data, and Google Does Not Say It Filters It Out

How non-human traffic reaches your reports at all

Google defines an impression as a link being seen or potentially seen, and CTR as clicks divided by impressions. The documentation setting out those definitions does not mention filtering bot or invalid traffic. So you cannot assume the impressions you are looking at are all people.

Start with the definitions, because the mechanics fall straight out of them. Google states that an impression means a user has seen, or potentially seen, a link to your site in Search, Discover or News, that a click is counted when it sends a user to a page outside those surfaces, and that CTR is the calculation of clicks divided by impressions¹. Nothing exotic is required to distort that ratio. Anything that generates the appearance of a result being seen, without a corresponding click, pushes it down.

Now the part worth stating carefully. That same documentation makes no mention of filtering bot traffic or invalid traffic from these metrics¹. We are not claiming Google filters nothing, because we cannot see inside the system and Google has not said. What we are saying is narrower and still useful: the documentation you would expect to describe such filtering does not describe it, so treating your reported impressions as a clean human count is an assumption rather than a documented fact.

The practical shape of the problem is a sudden divergence. Impressions climb steeply, clicks stay flat, and the ratio between them falls off a cliff. Because the two numbers move independently, the collapse looks like a content failure even when nothing about your content changed on the day it started.

What it actually looks like in a real account

On our own sister site, roughly 121,000 desktop impressions returned a 0.02 percent click-through rate. A separate client account showed 852 fast-bounce sessions inside 30 days. Neither figure describes human behavior, and both were sitting inside reports somebody could easily have acted on.

Take the first number seriously for a moment. A 0.02 percent click-through rate across 121,000 impressions means that for every five thousand times a link was recorded as seen, roughly one person clicked. Since CTR is defined simply as clicks divided by impressions¹, a ratio that low means the denominator was filled by something that never had any intention of clicking. Human audiences do not behave that way at any position in the results, for any query, in any industry. When you see a number that far outside the possible, the correct first conclusion is that the measurement is wrong, not that the page is.

The second pattern is quieter and arguably more expensive. A Southwest Florida professional services client showed 852 sessions inside thirty days that arrived and left in under three seconds. Individually each looks like a real visit. Collectively they drag every engagement metric downward and make a healthy page look like it is failing to hold attention.

Both of these sat inside ordinary reports that an owner or an agency would review in a normal month. That is the real exposure. Not a dramatic outage, but a slow corruption of the numbers you use to decide what to fix, presented in the same interface and the same font as the numbers you can trust.

If your own reports look like this and you are not sure, the free SEO audit is a reasonable starting point for the technical side, and you are welcome to send us a screenshot of the pattern and we will tell you honestly whether it looks like noise or like something worth acting on. Sometimes it is genuinely just a seasonal swing, and we would rather say so.

The claim we are not going to make

The standard pitch is that this traffic poisons your click signals and Google demotes you in response. Google says clicks are used for evaluation and experimentation, not for ranking, and explains that click data is too noisy and too easily spammed to serve as a ranking input. We are not going to sell you a mechanism Google rejects.

Google has addressed this directly. Its position, as reported when the question was put to it, is that interactions are used in a variety of ways such as personalization, evaluation and training data². Gary Illyes of Google has put it more bluntly, saying clicks are used for evaluation and for experimentation but not for ranking, and explaining that clicks are extremely noisy and very hard to clean up².

Read that reasoning again, because it is the strongest argument against the sales pitch. The reason Google gives for not using click data as a ranking input is precisely that it is easy to spam. A system that ranked on clicks would be trivially manipulable by exactly the traffic this article is about, which is why it is not a plausible design.

So when a vendor tells you that bot traffic is dragging your rankings down and their product will stop it, they are describing a causal chain that Google denies. It might still be true, since nobody outside Google can see the system, and there are longstanding arguments in the industry about it. What it is not is established fact, and it should not be the reason you spend money.

We would rather sell the honest version. This traffic is real, it is measurable in your own account, and it wrecks your ability to make decisions. That is enough of a problem to be worth solving without borrowing a scarier one.

The cost that is actually yours to pay

Corrupted measurement leads to corrected pages that were never broken. The expensive outcome is not a ranking change you cannot see; it is the month you spend rewriting titles, or the agency you fire, because a poisoned number told you something was wrong.

Clicks are already scarce: SparkToro and Similarweb found 68.01 percent of Google searches in early 2026 ended without a click at all³, so the clicks you do earn carry more weight in every judgement you make. Distorting the denominator underneath them is therefore worse now than it would have been a few years ago. Watch how the error propagates. Your CTR looks disastrous, so the obvious response is to rewrite titles and descriptions. You change pages that were performing well for humans, and the number does not improve, because the number was never about humans. Now you have spent a month, degraded pages that were fine, and learned nothing.

The same corruption hides real problems. If your baseline engagement metrics are already polluted, a genuine drop disappears into the noise. You lose the ability to detect the thing measurement exists to detect, which is the more serious failure even though it is the harder one to notice.

There is a third cost that lands on whoever you have hired. If you judge an agency on reported traffic and engagement, and those numbers are being moved by non-human activity, you are evaluating their work through a distorted lens in both directions. We wrote separately about how to tell whether your agency is actually doing anything, and the same principle applies here: judge on booked work and calls, which are much harder to fake than impressions.

• Do not rewrite titles in response to a CTR collapse you have not investigated.
• Check whether impressions are concentrated on odd URLs, devices or countries.
• Note the date the pattern started; genuine content changes rarely produce cliff edges.
• Judge outcomes on calls and booked work, which are far harder to distort.
• Keep a record of what your numbers looked like before, so you have a baseline to compare against.

What to do about it, in order

Diagnose before you filter, and filter before you conclude anything about performance. The aim is not to make a number look better. It is to get back to a set of measurements you can actually make decisions with, which is a different and more modest goal.

Segment first. Google reports impressions and clicks broken down by page, country and device¹, so split your data along those lines and look for concentration. Non-human traffic tends to cluster in ways real audiences do not, arriving on an unusual set of URLs, from an unusual place, on an unusual device split, and starting abruptly rather than building. That pattern is usually visible within ten minutes of looking properly.

Then decide whether it is worth acting on at all. Some of this traffic is harmless background noise that every site on the internet receives, and filtering it changes nothing that matters to you. It becomes worth addressing when the volume is large enough to distort your reporting, or when it is heavy enough to affect the responsiveness of your site for real visitors.

When filtering is warranted, the rule that matters more than any other is that a real customer must never be blocked. A filter that silently turns away a genuine buyer costs more than every bot it stops. That constraint is the whole design problem, and any protection worth having is built around it rather than around how aggressively it can block.

Finally, keep the goal modest and honest. Filtering this traffic gives you back trustworthy measurement. We are not going to tell you it will lift your rankings, because the mechanism that would make that true is one Google denies. Clean data is worth having on its own terms.

1. Segment before you act. Split by device, country and page in Search Console. Look for concentration and for an abrupt start date.
2. Compare against your own history. A cliff edge in impressions with flat clicks is the signature. Gradual change usually is not this.
3. Check whether it is affecting real visitors. Volume heavy enough to slow the site for customers is a different and more urgent problem than distorted reporting.
4. Decide whether it is worth filtering at all. Background noise that does not distort your decisions can reasonably be left alone.
5. If you filter, protect the customer path first. A filter that blocks a genuine buyer costs more than the bots it stops. That is the binding constraint.
6. Re-baseline afterwards. Record what clean numbers look like, so the next anomaly is obvious rather than arguable.