Interviewing
Interview Scorecard With an AI-Use Section
People ops and interviewers. Copy it into your own form, fill one row per claim during the round.
A blank interview scorecard for a round where the candidate worked with an AI assistant. One row per claim the round was assigned to test, the evidence quoted in their own words, a line naming where each observation came from, and a per-claim judgment of yes, not yet or no. Nothing on it adds up, and no line asks whether an answer sounded machine-written. Take it when a loop has its claims written down and the form keeps losing the evidence behind them.
Setup
- Write four to six claims before the round, taken from the work the person will actually do. Write each one as something they will do: "can scope an ambiguous request before starting" rather than "Problem Solving".
- Check the agenda reaches every claim. Hand any claim it cannot reach to another round instead of leaving it on this form.
- Agree in writing what an acceptable answer contains, before anyone interviews 1.
- Fill the evidence column during the answer. The reconstruction at the end of the day is where the specifics disappear and the impression takes over.
- File the finished form with the rest of the hiring record for that requisition, under the retention rule already covering your interview records.
The claims table
One row per claim, written before the round. Copy the row for each extra claim and delete the rest.
| Claim | The question or task behind it |
|---|---|
| — | — |
| — | — |
| — | — |
| — | — |
Evidence rows
One row per claim, filled during the round.
| Claim | What the candidate said or did | Where it came from | Judgment |
|---|---|---|---|
| — | — | — | — |
| — | — | — | — |
| — | — | — | — |
| — | — | — | — |
What the candidate said or did. Six to fifteen words in their own words, or the name of a specific artifact. "Dropped the third paragraph because the case it cited does not exist" still means the same thing in a debrief next quarter. "Good verification instincts" means whatever the reader brings to it, and the two cost the same thirty seconds to write. What an interviewer writes down about the capabilities the job called for carries something a number does not 2.
Where it came from. One word per row: watched, read, relayed, or tool-produced. A claim you watched a candidate make and a claim a summary reported to you are not the same evidence, and once both are typed into one box they stop being distinguishable. Loops now mix rounds carrying very different amounts of human observation, and the researchers behind the current selection-validity estimates flag automated administration and scoring as a setting where validity still needs to be evaluated 3.
Judgment. Against the standard from setup step 3, does the evidence for this claim clear what the work requires: yes, not yet, or no. Three values, no scale, and nothing that adds up.
The AI-use section
Four rows, filled inside the competency the AI question was attached to rather than as a block of its own. A block of its own invites one line standing for the whole topic, which is the thing being removed.
| Line | What the candidate said or did | Where it came from | Judgment |
|---|---|---|---|
| Delegation and the reason. What went to the assistant, what stayed with the person, and the reason in their own words | — | — | — |
| The check. The specific thing they did to test the part that mattered, and what they compared it against | — | — | — |
| The catch. What they found, or the moment they moved past something they should have found | — | — | — |
| The revision. What they would do differently, and whether that names a change to the process or only to the effort | — | — | — |
If the candidate used no assistant, record what they did instead and how they checked it under the same four lines, and write down the constraint they were working under. A tool ban, a regulated setting or an air-gapped system can produce strong verification habits and no model anywhere in the story.
Coverage gaps
| Claim the round did not reach | Reason | Round that picks it up |
|---|---|---|
| — | — | — |
| — | — | — |
Write "not observed" and the reason rather than leaving a cell blank or filling one in anyway. A blank cell reads as an oversight, a filled one manufactures evidence, and by next quarter nothing separates either from the rows that were really observed.
The call
| The call | The claims it rests on |
|---|---|
| — | — |
Hire or no hire, plus the one or two claims it rests on. Somebody who was not in the room should be able to reconstruct it from the rows above.
Limits of this form
- Nothing on it adds up. There is no weighted column and no single number standing for the candidate. A number here would summarise a judgment that never got written down, and the summary would be all that was left.
- No line asks whether an answer sounded machine-written. Non-expert readers asked to separate machine-written from human-written text performed at chance, and brief training barely helped 4; the tools sold for that job misfire hardest on people writing in a second language 5. Write down what was actually missing instead: no artifact named, no consequence described, no source given. Then ask the follow-up.
- Nothing leaves your own file. The employer fills this in, on the employer's own system. Olive computes, stores and transmits nothing per person from it.
- It applies no law to your facts. Consent and notice duties for a recorded or tool-analysed round are separate, they vary by state, and this form states none of them.
- It is a starting document for your own attorney to edit, and not a substitute for the advice of an attorney.
Take it
The file and the credit
The publishing entity legal name and postal address are not filled in yet, and both sit inside the disclaimer every packaged format renders. No file is emitted until they are.
Credit line, to paste beside anything you quote from this document.
<!-- Olive template. Licence: https://olive.is/licence/template/1-0 -->
<p><a href="https://olive.is/answers/tools/interview-scorecard-with-ai-section/" rel="nofollow">Olive</a>, Interview Scorecard With an AI-Use Section, version 1.0.0, checked 2026-08-26.</p>Important notes
Not a substitute for the advice of an attorney. This is a starting document for your own attorney to edit. It applies no law to your facts. No attorney has reviewed it for your state or your facts, and Olive is not your lawyer.
Published by Olive Independent Study, Inc., [DELAWARE INCORPORATING ADDRESS], United States. Contact hello@olive.is. A person reads every complaint and answers within ten working days. All concerns that Olive has engaged in the unauthorized practice of law are referred to the North Carolina State Bar, wherever the complaint came from.
Checked 2026-08-26 against the sources listed in this file. Version 1.0.0.
This text disclaims no warranty, caps no liability, waives no remedy, and names no court or state for a dispute. Those absences are deliberate.
Olive assesses how a person works with AI. It does not detect AI-written documents, and it never produces a score, a ranking, or a match percentage for a person. Candidates read the same report the employer reads.
Packaged files
- The publishing entity legal name and postal address are not filled in yet, and both sit inside the disclaimer every packaged format renders. No file is emitted until they are.
Checks
- Sources last re-opened August 26, 2026.
- Next review due February 26, 2027.
Where it came from
Read the argument behind it
- Four Lines Worth Scoring on AI Use, and Two to Delete Four lines to put on the interview scorecard for AI use, each anchored to a quote or an artifact, and the two lines to take off the form entirely.
- A Scorecard Holds Evidence, a Call, and Nothing Else A hiring scorecard needs five fields: the claims the round tests, the evidence for each, where the observation came from, the claims it missed, and one call.
References
- Structured Interviews: A Practical Guide U.S. Office of Personnel Management, 2008. opm.gov Supports setup step 3: a structured interview requires agreeing in advance what an acceptable answer contains.
- Predictive Validity of Interviewer Post-interview Notes on Candidates' Job Outcomes: Evidence Using Text Data From a Leading Chinese IT Company Frontiers in Psychology, Volume 11, Sec. Organizational Psychology (Shanshi Liu, Yuanzheng Chang, Jianwu Jiang, Haigang Ma and Huaikang Zhou), 2021. frontiersin.org Supports the evidence column: post-interview notes on 7,650 hired candidates at one company, where the job-related capabilities an interviewer named tracked later outcomes.
- Revisiting Meta-Analytic Estimates of Validity in Personnel Selection: Addressing Systematic Overcorrection for Restriction of Range Journal of Applied Psychology, 107(11), 2040-2068 (American Psychological Association); accepted manuscript hosted by co-author Filip Lievens, 2022. static1.squarespace.com Supports the provenance column: the authors of the revised selection-validity estimates flag automated administration and scoring as a setting where validity still needs to be evaluated.
- All That's 'Human' Is Not Gold: Evaluating Human Evaluation of Generated Text Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics (ACL-IJCNLP 2021), 2021. aclanthology.org Supports the removal of any line judging whether an answer sounded machine-written: non-expert readers performed at chance and brief training barely helped.
- GPT detectors are biased against non-native English writers Patterns (Cell Press), via PubMed Central, 2023. pmc.ncbi.nlm.nih.gov Supports the same removal from the tooling side: the error in judging text authorship lands unevenly on second-language writers.
5 sources, numbered by first appearance. How Olive sources claims