# Our scorecard gave her 5/5 for reliability. She'd told us she can't start before 7:45. > A car wash scorecard gave a candidate 5/5 across the board, while the interviewer's own notes showed she couldn't make the 6:30 AM opens. We tested a How I AI workflow that grades the interview notes instead of the candidate. By Jordan Menlo · Published October 26, 2026 · Updated October 27, 2026 Section: Markets & Main Street › Small Business · Series: AI Prompts, Reviewed (https://mcafax.com/notebook/prompts) Canonical: https://mcafax.com/notebook/ai-prompt-review-interview-feedback Source: MCAFax, The Broker's Notebook (https://mcafax.com/notebook). MCAFax is not a lender or a law firm. --- "Great energy, young and hungry, perfect culture fit. 5/5 across the board. Hire." That's the whole scorecard **Cole** wrote after interviewing a candidate for an opening attendant job. **Cole is the assistant manager at Bluegate Car Wash (a made-up business)**, and the job is mostly opens. At the bottom of his own notes, in small type: "Can't start before 7:45 most days." The opens start at 6:30 AM, four days a week. **AI Prompts, Reviewed** takes one prompt or workflow from How I AI, Claire Vo's show where guests share their screens and walk through how they really use AI. We run it on a small-business job, score it, and tell you what we'd change. **Credit:** Based on "How to Improve Interview Feedback Consistency with a Custom GPT," a workflow from How I AI (hosted by Claire Vo, written up by the ChatPRD team), shown by Zach Davis, engineering director at LaunchDarkly, in the episode "How I AI: Zach Davis's 3 Workflows for Enterprise Engineering with AI." [Read the original workflow on How I AI](https://www.chatprd.ai/how-i-ai/workflows/how-to-improve-interview-feedback-consistency-with-a-custom-gpt). [Watch the full episode](https://www.chatprd.ai/how-i-ai/enterprise-engineering-with-ai). Guest site: none listed on the How I AI workflow page. ## What the original does In the episode, Zach pastes an interviewer's scorecard into a custom GPT that holds his company's hiring rubric plus strong and weak example scorecards. The GPT rates the write-up Excellent, Good, Fair or Poor, points to vague or unsupported claims, and drafts a short Slack note he can edit before sending. The written workflow has 4 steps: - Write a role rubric that keeps "is this a good candidate" apart from "is this a good scorecard." - Build the GPT with the rubric, contrasting examples and a fixed output format. - Paste in a scorecard, with private data removed, and ask for evidence behind every claim. - Edit the coaching note, then use repeat problems to improve training and the rubric. The line we'd tape to the monitor: **"Do not make a hire or no hire recommendation."** ## The test Everything here is made up. Bluegate is hiring one opening attendant. - Pay: **$17** an hour, about **30** hours a week - Opens: four a week, starting at 6:30 AM - Must-haves on Renata's (the owner's) rubric: early-shift reliability, cash and card handling, calming an upset customer, and safe use of the wash equipment - Training: **24** hours, with Cole paid **$22** an hour to train **Run 1, the shortcut:** "Is this a good scorecard? Should we hire her?" The AI called the scorecard "clear and enthusiastic" and said she sounded like a strong hire. It never looked for evidence. It skipped right past the 7:45 note, and it repeated "young and hungry" back as a strength. That's the trap. Asked about the candidate, the AI grades the candidate, using notes that hold almost nothing. ## The rerun **Run 2:** same tool, our prompt below. It grades only the notes, checks each must-have, and flags wording that isn't about the job. It rated the scorecard **Poor**. Each of the four must-haves had a 5/5 score, and none had a single quote or example behind it. Evidence for 0 of 4. It also caught the clash. Reliability was scored 5/5, but the only fact on the card about time says she can't start before 7:45. Opens start at 6:30, so that's a 1.25-hour gap on each of four opens, or **5** hours a week nobody has covered. It flagged "young" as an age comment, not job evidence, and "culture fit" as a feeling with nothing behind it. Both should come off the card. Then the stakes. Renata's last opener had the same schedule problem and quit after 3 weeks. Training cost **24 x ($17 + $22) = 24 x $39 = $936**, and all of it walked out the door. Over a year, the gap adds up too: **5 x 52 = 260** hours of opens someone else has to cover. Renata called the candidate back. She was great with customers and happy to work closes, so Bluegate hired her for evenings and kept looking for an opener. Cole rewrote his scorecard with quotes, and his next one scored Good. ## Verdict: 7/10 **Strengths:** It grades the interviewer's notes, which is the step most owners never check. The fixed format makes two interviewers' cards comparable. And it refuses to make the hiring call. **Weakness:** It assumes you already have a polished rubric and a library of good and bad scorecards. A shop with one opening and two interviewers has neither, and the original doesn't tell the AI to check hard limits like schedule and pay first. **Founder tweak:** Turn the job's must-haves into the rubric, write one good and one weak example card yourself, and make the AI mark each must-have as asked, answered or not asked. ## Copy the MCAFax version This is our own rewrite, not the original. **[Copy the original from Zach Davis on How I AI](https://www.chatprd.ai/how-i-ai/workflows/how-to-improve-interview-feedback-consistency-with-a-custom-gpt).** ``` I own [business type] and I'm hiring for [job title]. You are checking the quality of an interview scorecard. Do not judge the candidate. Never say whether to hire, reject or advance this person. MUST-HAVES for this job (my rubric): Schedule: [days and start times]. Pay: [$ per hour]. Key tasks: [list]. Skills I score: [list each, with what good evidence looks like]. GOOD EXAMPLE of a scorecard: [paste one you wrote] WEAK EXAMPLE of a scorecard: [paste one you wrote] Step 1. For each must-have and skill, mark it Asked, Answered or Not asked. Quote the exact words from the scorecard that support each mark. Step 2. Find any score with no quote or example behind it. Step 3. Find any fact on the card that clashes with a score or a must-have. Show the math when it involves hours, dates or pay. Step 4. Flag wording about age, family, health, looks or "fit" that isn't job evidence. Step 5. Rate the scorecard Excellent, Good, Fair or Poor and explain why in 3 lines. Step 6. Draft a short, kind note to the interviewer with the 2 most important fixes. I will read and edit the note before sending it. SCORECARD: [Paste the scorecard, with the candidate's name and contact details removed] ``` Paste into any AI chat tool, then edit the note before anyone sees it. ## Before you hire 1. List the schedule, pay and must-have tasks before the first interview. 2. Ask every candidate the same questions about those must-haves. 3. Have each interviewer write quotes and examples, not just numbers. 4. Run each scorecard through the prompt, and make the call yourself. 5. Once you have notes worth comparing, sort new applicants faster with our [job application screening prompt](https://mcafax.com/notebook/prompts/job-application-screening). More owner prompts: [browse the MCAFax prompt library](https://mcafax.com/notebook/prompts). Want your AI to do this better? [Get the MCAFax Finance skill](https://mcafax.com/skills?utm_source=notebook&utm_medium=article&utm_campaign=skill-install&utm_content=ai-prompt-review-interview-feedback) and add it to your AI. ## Questions readers ask **What tool does the original workflow use?** ChatGPT, set up as a custom GPT that holds the hiring rubric and example scorecards. The guest pastes in each scorecard by hand. **Why not just ask the AI whether to hire someone?** Because it will answer from whatever notes you give it, even thin ones. Grading the notes first shows you what the interviewer actually learned. **What makes a scorecard good?** Each score has a quote or example from the interview behind it, and every must-have for the job was asked about and answered. **Why flag words like "young" or "culture fit"?** They aren't evidence about the job, and comments about age or family don't belong on a hiring record. Ask for what the candidate said or did instead. **Do I need a custom GPT for this?** No. Paste the prompt into any AI chat and add your rubric and examples each time, or save them in a project if your tool offers one. **Is it safe to paste interview notes into an AI tool?** Remove the candidate's name, contact details and anything private first. The check works just as well with "the candidate" in place of a name.