--- name: getting-hired-checks description: 10 rules from the Noesa course "Getting hired, with AI in the room". For anyone applying for work who now has AI on the desk and is not sure which parts of the application are safe to hand over. --- # Getting hired, with AI in the room — the rules Use with: Claude Code or Claude (save as a skill), Cursor (save under .cursor/rules as .mdc), ChatGPT or any other assistant (paste the text below into custom instructions or a project's instructions). 10 rules, taken from the course at https://noesa.leafsoft.online/c/getting-hired Each heading is one thing the course teaches. Most are checks to run on your own output before presenting it as done; a few are background you are expected to have. 8 also name a mistake models make by default, under "Watch for". Apply these to the thing you are producing — the type, the schema, the query, the copy — not only to how you explain it. Where a rule names a field, a format or an identifier, that name belongs in the output. ## Read a job ad for what is screened Separate a job ad's real screening criteria from its wish list, and name the evidence you have for each. **Watch for:** If you asked an AI to tailor your CV to a specific job ad, what would it likely get wrong? It will mirror the ad's most distinctive language back at you — "passionate", "fast-paced", "end to end" — because matching vocabulary is the pattern it was trained to complete. That is the half a reviewer discounts. It cannot supply the dates, tools and outcomes that decide the screen, because those were never in the prompt, so the tailored version often reads *more* generic than what you started with. Check every returned sentence against one question: could a stranger verify this? ## Split the CV into draftable and yours Mark every line of your CV as AI-draftable, AI-editable, or yours alone — and say why. **Watch for:** If you asked an AI to make your CV bullets sound more impressive, what would it likely get wrong? It reaches for intensity rather than evidence, because "impressive" in its training data looks like stronger verbs and bigger abstractions — *spearheaded*, *drove significant improvement*. That trades checkable detail for unfalsifiable claim, which is the opposite of what a reviewer needs, and it will occasionally invent a scale or a result that sounds plausible for your role. Read every rewritten line and delete any fact you could not defend in an interview. ## Make an achievement checkable Rewrite any achievement so it names a quantity, a scope, and the part you personally did. **Watch for:** If you asked an AI to quantify the achievements on your CV, what would it likely get wrong? It will supply plausible-looking numbers — a 30% improvement, a team of five, a 40% reduction — because achievement bullets in its training data almost always carry a figure, and the shape of the sentence demands one. Those numbers are generated, not recalled, and you will be asked to source them. Treat any figure you did not personally look up as a placeholder to replace or delete. ## Spot the generated cover letter Name the four tells of a generated cover letter and find them in your own draft. **Watch for:** If you asked an AI to write a cover letter for a job you want, what would it likely get wrong? It will echo the advert's own vocabulary back at the company, because the ad is the strongest signal in the prompt and mirroring it looks like relevance. To a reviewer who has read the ad a hundred times, that reads as an applicant with nothing of their own to say. It also cannot know anything about the company beyond what you pasted, so it substitutes intensity — *perfectly*, *extensive*, *deeply* — for the specifics that would make the letter worth reading. Supply the facts, or expect a letter that could be sent anywhere. ## Use AI on a take-home honestly Say which parts of a take-home exercise AI may touch, and defend that line if asked. **Watch for:** If you asked an AI to do a take-home data exercise for you, what would it likely get wrong? It will pick the most common grouping in its training data rather than the one this dataset warrants, and present the result with no sign that a choice was made at all. That confidence is the problem: the exercise usually turns on exactly that choice, and the write-up will not tell you it was a decision, so you cannot defend it later. Make the structural decision yourself, then let it help with the parts nobody is grading. ## Rehearse decisions, not answers Prepare for an interview by listing the decisions you have made, rather than memorising answers. **Watch for:** If you asked an AI for answers to common behavioural interview questions, what would it likely get wrong? It produces the median answer — a tidy situation, a proportionate action, a positive result — because that is what the genre looks like across thousands of examples. Interviewers have heard that shape all week, and it collapses at the first follow-up because there is no real memory underneath it to draw on. Use it to generate the *questions* you will be asked, not the answers you will give. ## Tell a story that survives follow-ups Structure a work story so each follow-up question has somewhere to go. ## Say what you actually used AI for Answer "how do you use AI?" with a specific, defensible account of your own working method. **Watch for:** If you asked an AI how to answer interview questions about your use of AI, what would it likely get wrong? It tends to produce a diplomatic, balanced-sounding statement about AI as a productivity tool used responsibly — language that signals nothing, because every candidate can generate the same paragraph. What makes the answer land is a specific instance with a verification step, and that has to come from your own week. Ask it to critique your example instead of writing one. ## Justify a salary number out loud State a salary number and give the two-sentence reasoning behind it without hedging. ## Run one application end to end Take one real application through every stage of this course and name the three lines a reviewer would challenge. **Watch for:** If you asked an AI to review your job application before you sent it, what would it likely get wrong? It reviews for quality, not for rejection. Asked to improve an application it will polish sentences and suggest stronger phrasing, because that is what "review this" means in most of its training data. But the thing that gets applications discarded is an unanswered screening requirement, and nothing in your document says what the screen was. Give it the job ad, ask it which requirements go unanswered, and do the judging yourself.