Agent · Track

Know who opened. Honestly.

One pixel per recipient — so a five-person thread tells you which of the five opened, not just that “someone” did. Every fetch passes six bot layers and earns a verdict, so Apple’s Mail Privacy pre-fetch and Gmail’s proxy can’t masquerade as a real read. We count only signals that actually fire.

Pipeline

pixel + click → 6 bot layers → verdict → score

signalfilterverdict

Recent verdicts

Score

0 / 100

The loop

4 steps

Send is instrumented

A unique 1×1 pixel is embedded per recipient, links are wrapped, attachments optionally get their own pixel — all at send time.

Signal fires

A pixel load, a link click, an attachment open, or a honeypot hit lands at the pixel-worker with its raw headers.

Bot filter runs

Six layers: UA blocklist, Cloudflare bot score, header heuristics, IP class (Apple MPP / Gmail proxy / datacenter), timing, honeypot correlation.

Verdict + score

Each event is tagged HUMAN / MPP_PROXY / GMAIL_PROXY / BOT and rolls up into a per-recipient engagement score.

Spec

what you configure

What we track

Per-recipient pixel (opens), wrapped links (clicks), attachment pixel (best-effort open), honeypot (bot trap). Pixel-only for opens — no CSS/font/DNS beacon theatrics.

Per recipient

Every recipient gets their own pixel token, so on a multi-recipient send you see opens by person — not a single anonymous "opened".

Attachments

Body pixel is per-recipient. The in-file pixel is per-email (same file for everyone) — so we can confirm the attachment was opened, honestly not by whom.

Bot layers

UA blocklist, Cloudflare bot score, header heuristics, IP class, timing, honeypot correlation. Six independent checks per event.

Verdicts

HUMAN (counts). MPP_PROXY (Apple pre-fetch — excluded, never a read). GMAIL_PROXY (real open via Google's proxy, partial credit). BOT (0).

Score weights

Direct pixel 20 · Gmail-proxy open 15 · first click 45 · document open/engaged/revisit bonuses. No phantom "API-confirmed" tier — the Gmail API can't see a recipient's open, so we don't claim it.

Honeypot

A trap only a scanner fetches. It flags a gateway scan (Proofpoint, Mimecast) and discounts that burst — it does NOT zero the person, so a real open later still counts.

Labeling

The dashboard never says "opened" for a proxy fetch without a qualifier — e.g. "Loaded by Apple — may not be a real read." Honest by default.

In practice

Real human read

→

Pixel, then a click from a different IP minutes later. HUMAN on both. Counts, per that recipient.

Apple MPP pre-fetch

→

Pixel from an Apple proxy seconds after send, nothing after. MPP_PROXY — excluded. We do not call it an open.

Enterprise gateway

→

Honeypot + pixel fire together at delivery (a security scan). That burst is discounted; the prospect's real open 3 hours later still scores. Not floored.

Document deep-read

→

Attachment pixel opens at +20m, then a multi-page read with revisits. Scored as genuine engagement.

Composes with

followup→

Followup's skip-on-engagement reads the per-recipient Track score at every step fire-time.

watcher→

Engagement on a watched thread updates the row's last-activity timestamp for sorting.

remind→

Engagement events surface on reminder rows so you can see if a stuck thread is at least being read.

Try Track free

100 tracked emails + 100 actions per month. Free forever.

Get started