Reporting and bot filtering

You should be able to tell
a human from a machine.

Security scanners and inbox previews often open and click an email before a person sees it. Counted as engagement, that activity inflates your opens, fakes your clicks, can crown the wrong A/B winner, and can start follow-ups nobody asked for.

Take a tour See pricing
A broadcast report metrics row: open rate, click rate and click-to-open each shown twice, once including bots and previews and once filtered, over a green trust line counting the machine interactions caught and removed, with a link that opens filtered activity.

Every open, click, and visit is classified, and known machine activity is filtered out before it counts.

Filtered reporting

What stays honest without the bots

Your open rate.

It counts people. When it goes up, something actually worked.

Your segments.

"Clicked twice this week" means a person clicked twice this week, so the list you're about to email is the list you think it is.

Your scores.

Machines earn nothing. A contact near the top is someone who's actually paying attention.

Your A/B tests.

Bots click too. If they're in the count, they get a vote on which version looks like the winner.

Your automations.

A follow-up goes out when a person acts, not when a scanner trips a link. Nobody gets a "thanks for your interest" because their firewall opened your email.

How it works

How we tell them apart.

Plenty of real people read your email from behind corporate mail systems. Plenty of bots don't come from anywhere that looks suspicious. A domain lookup can't tell them apart.

So RadarSend analyzes more than twenty signals on every open, click, and visit: how fast it came, whether every link was hit at once, whether it came from someone's phone or a data center, whether a browser ever actually drew the page, and whether anything human happened afterward.

It isn't a yes-or-no answer. Machine activity usually gives itself away. People are harder to prove, so RadarSend says how sure it is: likely a person when everything lines up, maybe when the signals are mixed. If the evidence isn't in yet, it waits.

Reasons to doubt it was a person

SpeedIt happened faster than a person could moveThree clicks in two seconds. Five links in half a minute. A click recorded before the message even reached the inbox.
Self-declaredIt told us what it wasSome scanners announce themselves in the request. Others show up with no browser identity at all.
OriginIt came from machine infrastructureThe click came from a data center, not someone's desk. Or one address sweeping every link in the entire send.
PatternOne "person" appeared in too many places at onceOne address across ten recipients. The same device on unrelated contacts. Intervals too even to be people.
AbsenceIt only did the parts a scanner has to doA click with no page view behind it. No scrolling at all. No reading behavior afterward.
MimicryIt tried to look human, but the details didn't add upPage view times that just aren't random. Scrolling too consistent to be human. A phone with a display no phone has.
See the evidence

Proof it was a person.

When someone really does read your email, they leave traces a scanner rarely produces. Those count for more than anything on the other side.

Real readingGenuine interaction, real scrolling, and enough time to have read it.
CuriosityOne page leading to another, and another. Machines don't read like people do.
Something afterwardA purchase, a form, a booked meeting, traced back to the click.
They came backActivity across separate sessions, which a single-pass scanner doesn't produce.
How post-click tracking works →
Show the work

See what was removed and why.

Open the filtered-activity drawer on any broadcast and RadarSend breaks down exactly what was caught. How many automatic inbox previews. How many security scanner clicks. How many gateway pre-scans. How many suspicious patterns. Which domains they came from.

Not a black box. You can see what was removed, why, and where it originated. You can inspect the same thing on an individual contact.

app.sonarsend.com/t/kestrel-supply/broadcasts/fieldnotes-47
A broadcast report with the Filtered activity drawer open: the machine interactions caught and removed as a share of all recorded opens and clicks, counted as machine opens, machine clicks and real clicks kept, then split into automatic inbox previews, security scanners, gateway pre-scans and suspicious patterns, with the domains the activity came from.
A/B testing

Don't let the machines pick the winner.

RadarSend compares A/B variants using filtered clicks and apparent clicks separately.

When both versions point to the same winner, the result is straightforward. When they disagree, the report says so plainly. The email that attracted more scanner clicks doesn't get mistaken for the email that worked better with people.

A/B test results on a broadcast report: a flagged message reading Filtering changed the winner, Variant B leads after bot filtering, with each variant showing its open, click and bounce rates and the apparent numbers beneath them.
More on A/B testing →
Where it matters

Filter once. Use the same answer everywhere.

Bot filtering isn't something applied only to a campaign report. The same classification follows the activity into segments, contact scores, and automations.

A machine click doesn't increase a score, add someone to an engaged segment, send them down an interested branch, or complete a sequence goal.

Reports, segments, contact records, and automation all read the same classification. If the foundation is noisy, everything built on it is noisy too.

Reporting should be useful, not flattering.

See pricing Take a tour