What employers actually send back

"I never hear back" is the most common complaint about applying for jobs. The record shows something slightly different, and worth knowing: you hear back constantly, almost never from a person.

The mix

Three numbers from that table are the whole story.

83.5% of employer mail is machine-generated. 6,836 receipts plus 1,696 other automated messages. A receipt is not a response — it is a database insert with a friendly sentence attached — but it looks exactly like one in an inbox, which is why the experience of applying feels busier than it is.

524 rejection-class messages. Employers who tell you no are a small minority of the mail, and they are doing something the rest are not. 364 were explicitly worded; 160 arrived worded as "important information about your application", which is a rejection that does not use the word.

53 emails, total, were about an interview. Across 10,217 inbound messages and 700 companies: 26 confirmations, 22 scheduling requests, 5 reminders. That is 0.5%.

So does anyone actually say no?

Mostly, no — they say nothing. Counted from the event record instead of the inbox: across the whole history of this search, interview-related events (confirmations, scheduling requests, reminders, interviews that happened) have been recorded for 19 companies — 54 events in total, against 6,565 applications submitted on this codebase alone. The rest sit in the state they were submitted in.

We want to be careful about what that means, because it is easy to over-read. An application with no recorded outcome is not proof of employer silence. It can equally be a reply that arrived and was never matched back to the right application — the matching is done by company name and subject line, which fails on generic senders and name mismatches. In this data, silence and an unmatched reply look identical, and no honest version of this page can separate them. What can be said is the thing the inbox itself shows: explicit outcomes of either kind are rare, and automated receipts are not.

How the classification works, and how it gets things wrong

Not by a human reading 10,217 emails, and not by an AI model. By about eight regular expressions over the subject line plus a check on the sender address — roughly: thank you for your interest and will not be proceeding mean rejection; thank you for applying and application received mean receipt; interview confirmation means what it says; a sender like no-reply@, careers@, notifications@ or @myworkday marks the message as machine-generated.

This is crude, and the failure modes are specific and worth stating:

• A rejection worded unusually is not counted as a rejection. It falls through to "other machine-generated mail" or "written by a person". So 524 is a floor on rejections, not a total.
• One rule had to be added after it was caught being wrong: a real recruiter writing "regarding your application…" was being filed as an automated receipt and disappearing from view. Machine-shaped classifications now only stick to machine-shaped senders. That one bug was silently deleting the most valuable mail in the inbox, which is a good argument for keeping the classifier readable.
• "Written by a person" is a residual category — anything from a human-shaped sender that matched no rule. It contains real recruiter outreach, but also newsletters and one-off notifications that happen not to come from a no-reply address. 1,108 is an over-count of genuine human contact.

Reproducing this

Run 2026-11-07 (machine-readable at /stats.json):

SELECT kind, count(*) FROM job_events WHERE direction='in' GROUP BY 1 ORDER BY 2 DESC, 1; — the mix table, which must sum to the headline

SELECT count(*), count(DISTINCT company), min(at)::date, max(at)::date FROM job_events WHERE direction='in'; → 10,217 · 700 · 2026-06-01 · 2026-11-07

SELECT count(*), count(DISTINCT company) FROM job_events WHERE kind IN ('interview_confirmed','scheduling','interview_reminder','interview_happened'); → 54 events · 19 companies