The short version
- An analytics audit found roughly 92% of one client's raw sessions were not human.
- Nothing was visibly broken. Filtering had never been applied, so every ratio quoted from those reports was built on a fictional denominator.
- Bot traffic concentrates on particular paths, so the top-pages report sends content teams to produce more of the wrong thing.
- Check the ratio before optimising anything. You cannot improve a conversion rate whose denominator is fiction.
We ran an analytics audit on a client account expecting to find a tracking gap. What we found was that roughly 92% of the raw sessions were not human.
Nothing was obviously broken. No misconfigured tag, no rogue script. Filtering had simply never been applied, and every ratio anybody had ever quoted from those reports was built on a denominator that was mostly crawlers.
Why does this matter if bots do not convert?
Because they sit in the denominator.
| Metric | With unfiltered traffic | What it causes |
|---|---|---|
| Conversion rate | Looks far worse than reality | Redesigning pages that were fine |
| Top pages by sessions | Skewed to bot-favoured paths | Commissioning the wrong content |
| Engagement rate | Suppressed | Concluding the audience is disengaged |
| Cost per visit | Understated | Overvaluing a paid channel |
| Traffic trend | Noisy | Chasing changes that never happened |
The worst effect is on the top-pages report. Bot traffic does not distribute evenly. It concentrates on particular paths, so those pages look like winners, and a content team dutifully produces more of them.
How do you spot it?
Five signals, none conclusive alone.
Sessions with zero engagement time. One pageview, immediate exit, no scroll. A real reader who bounces still takes a few seconds.
Concentration. A handful of URLs taking a wildly disproportionate share, especially ones with no internal links pointing at them.
Data-centre networks. Real buyers arrive on consumer and corporate connections.
No human rhythm. Human traffic has a shape: weekday peaks, evening dips, holiday troughs. Automated traffic is flat.
Impossible geography. Volume from markets where you have no presence, no content and no ads.
You cannot improve a conversion rate whose denominator is fiction.
What do you do about it?
Turn on the platform’s known-bot filtering, which is off or partial more often than people assume. Exclude internal and agency IPs. Segment out data-centre traffic. Then build the reporting view on engaged sessions rather than raw sessions.
Then rebuild the baseline, or at least annotate the date filtering started. Otherwise the reported drop looks like a collapse and somebody will spend a quarter fixing a problem that does not exist.
How much traffic should you expect to lose?
Enough that somebody senior needs warning before you switch filtering on.
We have seen filtered numbers land anywhere from a modest trim to a fraction of the original. The direction of the change is never the problem. The problem is a leadership team seeing a chart fall off a cliff in the same month you took over the account, with no note explaining why.
Send the explanation before the report, not with it.
The general lesson
Nobody had checked. The account had run for years, with monthly reports going to a leadership team, and the ratio had never been examined.
That is not unusual. Analytics gets configured once, at the start, usually by whoever was free that week, and is then treated as ground truth forever. Every audit we run now begins by asking what proportion of this traffic is real, before anyone is allowed to draw a conclusion from it. It takes twenty minutes and it has changed the recommendation more than once. The same discipline applied to rankings is in ranking without clicks.
Frequently asked questions
How much of website traffic is bots?
It varies enormously by site and by analytics configuration. In one audit we ran, roughly 92% of raw sessions were non-human. Nothing was misconfigured in an obvious way. Filtering had simply never been applied.
How can you tell if your traffic is bot traffic?
Look for sessions with zero engagement time, heavy concentration on a handful of URLs with no internal links pointing at them, traffic from data-centre networks rather than consumer ones, volume with no human daily or weekly rhythm, and geography where you have no presence or content.
Why does bot traffic matter if it does not convert?
Because it sits in the denominator. Conversion rate, engagement rate and cost per visit are all ratios, so an inflated session count makes each look worse than reality and ranks your real pages incorrectly against each other.
Does bot traffic hurt SEO?
Not directly. It corrupts your decisions, which is worse. If bots cluster on certain paths you will read those pages as winners and commission more like them, while underrating the pages real buyers use.
How do you filter bot traffic?
Enable the platform's known-bot filtering, which is often off or partial. Exclude internal and agency IPs. Segment out data-centre traffic. Then build reporting on engaged sessions rather than raw sessions.
Should you rebuild historical reports after filtering?
Yes, or at minimum annotate the date filtering began. Otherwise the reported drop reads as a performance collapse and somebody spends a quarter fixing a problem that does not exist.
Sources
- Chua Network delivery data across 8 client accounts (internal fact bank)
- Chua Network engagement records, anonymized (internal experience bank)
