Bait and Switch, or Just a Bug

The invite promised a screening call and the funnel delivered an application. By design or by accident, the candidate walks.

The invite that over promised

On October 12 an email arrived inviting me to a "brief, 5-minute intro call" about a Sr. Product Manager, Employer ATS role at ZipRecruiter. It had a named sender and a time-slot, Wednesday at 2:00 PM PDT. It named the call a live event: I would join a virtual queue, and the sender would speak with me as soon as they were available.

I clicked RSVP. The landing page said "No sessions scheduled" with two calls to action beneath: Apply Now and No Thanks. Reader, I did not apply.

An application wearing a calendar invite as a costume

The costume is convincing. The email is real, and the link goes exactly where it says: ZipRecruiter's own RSVP page for the listed job. Its "Next Available" time is hard-coded in plain text, so it was fixed the moment the email went out. The link itself carries a job ID and no time slot, passing no data to it’s landing page.

When the landing page opens, it asks ZipRecruiter's server about the job, and the request can name only the job ID, not the time in the email. When I checked on October 17, after the slot had passed, the answer was an empty, inactive session with RSVP switched off. I cannot see the server's answer on October 13, only the page it produced: "No sessions scheduled." Either way, the email and the page were not comparing notes, and the invitation promised something the page could not deliver.

The fine print does say "automated message" and "not a guaranteed meeting", in 12px. The nonexistent time slot dwarfs it at 16px bold.

Two doors, and one of them works

I track my job search like a data pipeline, which is the subject of my earlier post. Between 1 June and 12 October I submitted 92 more applications. How each one started looks like the strongest signal in my data.

In that previous cohort (Jul 2025–May 2026) my cold-apply screen rate was 6.7%. Counting only cold routes (alerts, boards, direct applies and untagged), it is now 2 in 78, or 2.6%. Add the four referrals and every non-recruiter route comes to 3 in 82, or 3.7%. These are small samples and different cohorts, and a recruiter who writes first has already chosen me, so the rows are not like for like. But the pattern is loud: a human choosing me first works, and the jobs portal submissions almost never does.

That is why the invite landed the way it did, k’snag! An email saying a recruiter wants a call is the one message my data says to answer.

It also shows what the email swapped. A screening call sits at the bottom of the funnel. An application sits at the top. The invite promised the first and delivered the second.

I cannot see ZipRecruiter's metrics. If this funnel is judged on conversion to application, the swap is a win for the sender and a loss for me. That is a theory, not a finding.

Three funnel basics the page broke

"Automated" and "not guaranteed" belong in the first line, not the fine print.

Hi, Bex, hold my beer! Hugs, Phil (Your Career Advisor)

A second email landed on Oct 14. It said "I think you're a really good fit" for a job called testjob_apr01 at a company called WFNQAFR44D. That looks like a test listing that reached a real inbox. It's the same missing gate as the RSVP, earlier in the pipeline: nothing checked that the job was real before the send. One QA leak says nothing about the whole company. It does show where the checkpoint belongs. A bait and switch would have checked the bait. This one shipped a job called testjob_apr01. My read is a missing guardrail, not an evil scheme. Either way, I clicked, I did not apply.

What I cannot say

I have no evidence that ZipRecruiter's applicant tracking system, or anyone's, treated an application unfairly. I never applied through either of these emails. An invitation email and an ATS are different products. I am describing a funnel, not accusing a vendor.

What the follow-up will cover

My full follow-up to "I Treated My Job Search Like a Data Pipeline" lands at the end of October, after my next read of the data. Its angle: I fixed the resume, and the rate went down. The thing I controlled improved. The thing I did not control decided.

It will sort what is on record from what is alleged from what is only theory, because the public evidence on AI in hiring contains all three. One example of the first kind: University of Washington researchers tested embedding models for resume screening on more than 500 resumes. The models favored White-associated names in 85.1% of cases and female-associated names in only 11.1%. That is a controlled study, not a live hiring system. Whether anything like it sits behind a portal I applied through, I cannot know.