Bot Sessions Are Sitting in Your Conversion Rate Denominator
Two small sites measured 82% and 73% non-human sessions. Crawlers never convert, but they land in the denominator of every conversion rate, so the arithmetic flatters the traffic and damns the product. Here is the ten-minute recount.
Ready to start?
Try ContentMation Pro free for 14 days, then $15.99/mo. Cancel anytime.
Start your 14-day free trialShown as “SeekerPro” at checkout · same plan, one subscription
Secure checkout by Stripe. To cancel later: Settings › Cancel my plan.
14-day free trial · Cancel anytime · 30-day money-back guarantee.
Another option: Lock in founding $9.99/mo · Start on the free plan, no card
Two numbers off two small sites, read from their session recordings, with humans and bots bucketed separately by the recording tool rather than by hand: Site A logged 22 human sessions against 99 bot sessions. Site B logged 18 human against 48 bot. That is 82% and 73% non-human. Both sites are small, and the size is the whole point, because the crawler floor is roughly constant while the human number is not. The smaller the audience, the larger the share of measured traffic that was never a person, and the more badly every rate computed from that traffic misleads.
What Actually Pulls Those Pages
Crawlers, uptime monitors, link-preview fetchers and AI scrapers all request pages. They fire no interaction events, they never reach a checkout, and unless they have been excluded they sit in the denominator of every conversion rate computed from session counts. The numerator stays honest, because machines do not convert. The denominator inflates. The rate collapses.
The arithmetic then reads as lots of people come and almost nobody buys, when the accurate sentence is most of that was never a person. Those two sentences point at completely different work. The first sends a team to rewrite a landing page or cut a price. The second sends them to fix a measurement filter and then look again.
The Error Never Points Randomly
This is the part worth internalising, because it separates this problem from ordinary measurement noise. Random error averages out and can be tolerated. This error has a fixed direction: it always flatters the traffic and always damns the product.
Traffic looks larger than it is, because machine sessions are counted as arrivals. Conversion looks worse than it is, because those arrivals never convert. So the conclusion reached is reliably the wrong one, and it is reached with confidence, because the numbers are real numbers from a real dashboard. Teams conclude the landing page is broken, or the pricing is wrong, or the market does not want the thing, on the strength of a figure whose denominator was mostly machines.
Want to see these results for your brand? Start a free campaign in 60 seconds.
Get started freeThe Ten-Minute Recount, Step by Step
Open the analytics for a window long enough to hold a reasonable number of conversions, normally 30 days on a small site. Record the raw session count and the raw conversion count, and compute the rate the ordinary way, so there is a before number to compare against.
Now apply one filter and nothing else: keep only sessions that fired at least one interaction event. A scroll, a click, a keypress, a form focus, any evidence that something moved. Recompute the same rate on that filtered set of sessions, with the same conversion numerator.
Compare the two rates. If the number moves by less than roughly 20%, the original measurement was already close enough to people, and the product verdict drawn from it stands. If it moves by 2x or more, the original number was not describing humans at all, and any verdict reached on it was arithmetic rather than evidence.
A Worked Example on the Site A Figures
Take Site A exactly as measured: 22 human sessions and 99 bot sessions, 121 sessions in total. Now suppose, purely as an illustration of the arithmetic rather than as a reported result, that three of those sessions converted. All three were necessarily human, because machines do not fill forms.
Computed the ordinary way, the rate is three conversions over 121 sessions, roughly 2.5%. Computed over sessions that showed evidence of a person, it is three over 22, roughly 13.6%. Same site, same day, same three conversions, and a figure that moves by more than five times depending only on which denominator was used.
A 2.5% rate on a landing page invites a redesign. A 13.6% rate invites buying more of whatever sent those 22 people. One of those decisions is expensive and wrong, and nothing in the dashboard signals which one is which.
Two Traps That Quietly Void the Check
The first trap is reaching for unique visitors instead of sessions and assuming the problem is handled. It is not. Well-behaved crawlers are issued their own session identifiers exactly like people, so a unique-visitor count carries the same inflation as a session count and simply hides it behind a more trustworthy-sounding noun.
The second trap is filtering by user-agent string alone. That catches only the machines that are honest about being machines. A headless fetcher that reports itself as a normal browser passes straight through a user-agent filter. The interaction-event filter catches both kinds in a single pass, for one plain reason: a headless fetcher that lies about its user agent still does not scroll.
The Honest Limit on These Numbers
The bot and human split quoted above is the recording tool's own classifier, not ground truth, and classifiers are wrong in both directions. Some real people using strict privacy settings get bucketed as machines, and some well-built scrapers get bucketed as people.
So this is a report of what two dashboards said, not a proof of a law, and certainly not a claim about what any other site would find. The useful move is not to accept 82% as a general figure. It is to run the recount on your own numbers this week and find out whether the product verdict you are carrying around survives a denominator made only of people.
Frequently Asked Questions
What is this article about?
Two small sites measured 82% and 73% non-human sessions. Crawlers never convert, but they land in the denominator of every conversion rate, so the arithmetic flatters the traffic and damns the product. Here is the ten-minute recount.
How does "Bot Sessions Are Sitting in Your Conversion Rate Denominator" apply to AI-powered marketing automation with ContentMation?
Based on campaigns distributed into startup communities of 150K+ entrepreneurs, AI-generated platform-specific content outperforms generic content by 3-5x in engagement and 2-3x in conversion rates. ContentMation delivers marketing results that are 3x cheaper than Google Ads with 5x faster results.
Try ContentMation Pro free for 14 days
$15.99/mo after the trial · Cancel anytime · Or choose Founding $9.99/mo locked for life.
Start 14-day free trialShown as “SeekerPro” at checkout · same plan, one subscription
Secure checkout by Stripe. To cancel later: Settings › Cancel my plan.
14-day free trial · Cancel anytime. 30-day money-back guarantee.
Launch your marketing on autopilot
Free to get started. Upgrade when you're ready.
Start Free CampaignTrusted by 828 marketers worldwide
Paid plans include a 30-day money-back guarantee, full refund if you are not 100% satisfied, no questions asked.
Reader comments
No account needed. Share what worked, what did not, and what you would add.
Add your take below. A concrete tip, a result or a question helps the next reader most.