Autopilot AI
Sorts the pile, hopes for the best.
Black-box AI rankers compress the candidate to a single number and rely on the recruiter to trust it. When the verdict is wrong, there is no productive conversation — only deference.
AI recruiting copilot
The AI recruiting copilot that lets the recruiter make the call.
Autopilot belongs in airplanes, not in hiring. ShortlistTable is the AI copilot for recruiters: the AI reads the pile, surfaces evidence per criterion, and queues the ambiguous cases for human review. You keep the judgement; the AI removes the keystrokes.
Copilot, not autopilot · Recruiter override on every cell · No hidden screen-outs
shortlisttable.app / candidate / alice-chen
Candidate
Alice Chen
Backend platform engineer · 7y
Criterion
Has the candidate owned an on-call rotation for ≥ 12 months?
Yes · AI verdictverdict · source-backed
Evidence in resume
“Owned primary on-call for the payments event bus across 14 services from 2022 to present.”
Resume p.1 · Experience · alice-chen.pdf · ShortlistTable AI ⇄ Recruiter
Every verdict in ShortlistTable links to the exact sentence in the resume that supports it. If we picked the wrong sentence, you adjust the column. If the sentence is right but the verdict is wrong, you override the cell.
Copilot
AI reads, recruiter decides
Per-cell
Evidence + override
0
Hidden screen-outs
Audit
Trail for every override
Why autopilot tools backfire
Hiring is the wrong place to be fully automated.
The autopilot pitch — AI that picks the best candidate, schedules the call, sends the offer — is the wrong target. Hiring has long-run consequences, ambiguous signals, and legal exposure. The places automation pays back are the keystrokes between judgement, not the judgement itself.
A copilot framing is the better fit. The AI does the work that doesn’t need a human: extracting fields, checking criteria, citing evidence, flagging ambiguous cases. The recruiter does the work that does need a human: reading the candidate as a person, deciding who to call, owning the rejection. That division of labour is the entire product.
Manual screening
Drowns the recruiter in the read step.
Reading 200 resumes against the same 5 criteria, 200 times, is exactly the kind of work AI should be doing. The recruiter’s hour is better spent on the 30 candidates that actually need a careful read.
Copilot screening
AI reads. You decide.
The AI gathers per-criterion evidence and presents it to the recruiter cell-by-cell. The recruiter agrees, overrides, or queues for review. The shortlist is human-authored; the keystrokes are not.
How the copilot works
What changes when AI is a copilot, not the pilot.
01 · Feature
AI reads every resume
Per-criterion AI verdicts on every candidate, surfaced in a single sheet. No PDF-by-PDF reading; the recruiter starts the review with all the evidence already gathered.
Yes · 5 must-havesReview · 1 ambiguousNo · 3 missing
02 · Feature
Evidence per cell, not per candidate
The recruiter sees the source sentence behind each verdict — not a summary of the resume, not a generic reasoning blurb. The actual line from the actual page.
Resume p.1 · Experience“Owned primary on-call for the payments event bus across 14 services.”
03 · Feature
Recruiter override is first-class
Disagree with the AI on a single cell? Override it. The audit log captures both values. The shortlist always reflects the recruiter’s most recent call.
AI: review · 0.62
Recruiter: yes (3y at Stripe ledger)
04 · Feature
Review queue for the ambiguous cases
Below your confidence threshold, the AI queues candidates for human review instead of deciding. You spend your time on the cases that actually need judgement.
31
in review queue
(of 200)
05 · Feature
Never auto-reject
The bottom of the list is a hold queue, not a delete queue. Low-fit candidates stay accessible with evidence intact, so any rejection is reviewable.
Hold queueAudit trailRe-runnableNo silent screen-outs
06 · Feature
Audit trail per cell
Every AI verdict, recruiter override, re-run, and column edit is timestamped. The copilot pattern produces a defensible decision log by design.
AI · 09:14 · 0.91 confidence
Recruiter override · 11:02 · note
Re-run · next day · column edited
How it compares
Copilot vs autopilot AI.
| Property | Autopilot AI screener | Conversational AI (Paradox-style) | Manual screening | ShortlistTable |
|---|---|---|---|---|
| Recruiter as final authority | ✕Model decides | –Limited override | ✓Yes | ✓Every cell overridable |
| Per-cell evidence visible | ✕Score only | –Conversation log | ✓Manual | ✓Sentence + page |
| No auto-reject | ✕Common | ✕Auto-stage | ✓Manual | ✓Hold queue only |
| Speed at batch scale | ✓Fast | ✓Fast | ✕Slow | ✓Fast — batched |
| Audit trail per decision | ✕Score only | –Conv. log | ✕Manual notes | ✓Per-cell timestamps |
✓ Supported– Partial / manual✕ Not supported
A copilot session
What working with a copilot actually feels like.
Monday, 9:15am. Your weekend inbound: 47 resumes for the senior backend role. Pre-copilot, the next four hours would be PDF-by-PDF reading and copy-pasting fields into a sheet. With a copilot, the next four hours look very different.
Open the workspace. The AI has already done a per-criterion pass on every resume — 5 columns × 47 candidates = 235 cells, each with a verdict and a source sentence. 18 candidates pass all 5 must-haves; they’re in the “call first” bucket. 19 land in the review queue with at least one ambiguous cell. The remaining 10 are in the hold queue, evidence intact.
You open the review queue. The first candidate has 4/5 green, 1 amber on “on-call ownership” because the resume says “rotated through on-call” without specifying primary vs secondary. The AI flagged this honestly instead of guessing. You override the cell to “Review fit” with a note: “ask in the call”.
Twenty-three minutes later you’ve reviewed all 19 queue candidates. Now you spend the rest of the morning on the 18 “call first” candidates — same workflow, but you’re reviewing for nuance rather than presence. By lunch you’ve sent screening-call invites to 12 candidates with per-criterion fit summaries inline.
Notice what didn’t happen: no auto-reject, no model-vs-recruiter argument, no opaque score. The AI did its job, you did yours, and the audit trail is complete by default. That’s the copilot pattern.
FAQ
Common questions on the copilot pattern.
What is a recruiting copilot, exactly?+
A workflow tool where AI handles the mechanical parts of screening — reading every resume against every criterion, citing evidence, queuing ambiguous cases — while the recruiter handles the judgement parts: agreeing or overriding each verdict, deciding who to call, owning the rejection. The AI never makes a final decision on its own.
How is this different from autopilot AI screeners?+
Autopilot AI screeners output a sorted list and treat the model as the decision-maker. Copilot AI surfaces evidence per criterion and treats the recruiter as the decision-maker. The architectural difference is that every verdict in a copilot is overridable and the bottom of the list is never silently dropped.
Is this slower than full autopilot?+
For the part the recruiter spends time on — reviewing the borderline cases — yes, marginally. For the total time-to-shortlist, no: the mechanical reading is automated, the recruiter only spends time on the cells that actually need judgement. Net throughput is higher than either autopilot or manual.
How does this comply with the EU AI Act and NYC AEDT?+
These laws specifically target “automated employment decision tools” — systems that make or substantially influence the decision. ShortlistTable is a copilot by design: every verdict is overridable, no candidate is removed without a recruiter action, and the per-cell audit trail produces the documentation those laws require. See the explainable AI screening page for the detailed compliance posture.
Can the copilot be tuned to be more or less aggressive?+
Yes — per-column confidence thresholds let you decide where the AI’s verdict counts as authoritative versus needing recruiter review. A lower threshold means more cells land in the review queue. A higher threshold means more cells are auto-categorised.
Does the recruiter have to review every cell?+
No. Cells above the confidence threshold are auto-categorised; the review queue contains only the ambiguous cases. The whole point is to give the recruiter their attention back for the cases that actually need it.
More from us
Keep exploring.
Related
The AI resume screener
Auditable AI screening, end-to-end.
Related
Explainable AI screening
Compliance with EU AI Act / NYC AEDT.
Related
Source-backed screening
Citations on every cell.
Free tool
Try the outreach generator
AI drafts, recruiter sends.
Read
Keyword vs semantic screening
What semantic AI changes.
Copilot, not autopilot
Try the AI recruiting copilot on your next pile.
AI does the reading. You make the call. Try free on 25 resumes — no credit card.