AI for non-technical roles — 25 jobs, the task that compounds, the trap, the prompt

Compounding task: draft listing copy from a property checklist and pull a neighborhood brief from public data. Daily prompt: 'You are drafting MLS-compliant listing copy. Inputs: [bullet list of features, square footage, year built]. Constraints: no fair-housing-protected adjectives (steered language, religion, family-status hints), under 250 words, two-paragraph hook + features. Return three variants.' Trap: the model invents school ratings or comp prices. Verify both against your MLS and your state Realtor association's Fair Housing guidance before posting.

Compounding task: first-pass summarization of long discovery documents and depositions. Daily prompt: 'Summarize this [deposition / contract / motion]. Return: 1) one-paragraph TL;DR, 2) chronology of dates and events, 3) five quoted lines with page:line citations, 4) three open questions for the attorney. Do not interpret legal strategy.' Trap: hallucinated citations. Multiple sanctions orders in 2023–2024 against attorneys who filed AI-generated briefs with fake case law (Mata v. Avianca being the canonical one). Use a model with grounded retrieval or hand-verify every cite.

Compounding task: differentiating one lesson plan into reading-level variants and IEP-aligned versions. Daily prompt: 'I have a 7th-grade lesson on [topic]. Output three variants: (a) one for students reading at 4th-grade level, (b) one for advanced learners with an extension question, (c) one with sensory-load accommodations per the IEP guidance I'll paste below. Keep learning objectives identical.' Trap: feeding student names or IEP details into a public consumer model violates FERPA in most US districts. Use only a tool your district has cleared, or anonymize first.

Compounding task: drafting shift-handoff SBAR notes and patient-facing education leaflets from your charting shorthand. Daily prompt: 'Convert these shift notes into an SBAR handoff: Situation, Background, Assessment, Recommendation. Keep all clinical numbers exact. Flag anything ambiguous as [VERIFY] rather than guessing.' Trap: PHI leakage. HIPAA-cleared tools (the major EHR vendors all have ambient-documentation partnerships as of 2026 — check your hospital's IT) exist for this. Do not paste patient identifiers into a consumer chat product.

Compounding task: campaign-brief to first-draft assets — email, social caption, landing-page hero. Daily prompt: 'Here is the campaign brief: [audience, problem, value prop, CTA, channel mix, voice rules]. Output: 1) three email subject lines, 2) the email body in plain HTML, 3) three LinkedIn captions in our voice, 4) a 60-word landing-page hook. Match the voice rules exactly; flag anything that needs a claim review.' Trap: model defaults to generic SaaS-speak. Build a 'voice rules' document once (banned words, sentence-length targets, register) and paste it into every brief.

Compounding task: writing the narrative around the numbers — reconciliation explanations, variance commentary, audit-prep memos. Daily prompt: 'Here is the variance: budget $X, actual $Y, delta $Z, prior-year actual $W. Drivers I know: [list]. Draft a 150-word management-discussion paragraph in plain English, no jargon, suitable for a non-finance reader on the leadership team.' Trap: the model will make up reasons for variances if you don't supply the drivers. Never let it speculate on cause — supply the cause, ask it to write the prose.

Compounding task: drafting case notes from voice memos and explaining benefits-program rules at a 6th-grade reading level. Daily prompt: 'Convert this voice memo into a case note: [transcript]. Use BIRP format (Behavior, Intervention, Response, Plan). Keep all client quotes verbatim; do not paraphrase quoted speech.' Trap: same as nursing — client identifiers are confidential. Strip names and DOB before pasting into anything not on your agency's approved list. SAMHSA 42 CFR Part 2 rules apply for substance-use cases.

Compounding task: reader's-advisory and research-guide drafting. Daily prompt: 'A patron asked for books like [title] but with [softer ending / female protagonist / under 300 pages]. Suggest five published titles with author, year, and a one-sentence why-this-matches. Mark any title you are not certain exists.' Trap: invented book titles. The library science community has been documenting this since 2023; ask the model to flag any title it isn't sure about, then verify in your ILS.

Compounding task: research compression and outline scaffolding for assigned pieces. Daily prompt: 'I'm writing a 1,500-word feature on [topic] for [outlet]. Outlet voice rules: [paste]. Give me: 1) a five-section outline, 2) three open questions I should answer with reporting, 3) five candidate sources (real people or publications I should reach out to). Do not draft the piece itself.' Trap: shipping AI drafts as your own undisclosed work — most outlets now have AI policies. Read your contract.

Compounding task: weekly status reports from raw stand-up notes and risk-register updates. Daily prompt: 'Here are the stand-up notes from the last five days: [paste]. Produce: 1) RAG status (Red/Amber/Green) per workstream with one-sentence justification, 2) top three risks with owner and next action, 3) a four-bullet exec summary for an audience of one VP.' Trap: PMs over-format. Force the model into your shop's existing template; don't let it invent a new one every week.

Compounding task: first-draft replies to inbound tickets, grounded in your knowledge base. Daily prompt: 'Customer wrote: [ticket]. Our policy on this is: [paste relevant KB article]. Draft a reply in our brand voice (friendly, concise, plain English, no exclamation points). Include the steps the customer needs to take. If the KB doesn't cover this, say so and don't invent.' Trap: hallucinated policy. Always paste the actual KB article into context; never let the model rely on its training-data memory of your product.

Compounding task: pre-call account research and post-call summary + next-step email. Daily prompt: 'Here is the prospect: [LinkedIn paste + company URL]. Pull: 1) three plausible business reasons they'd care about our [product], 2) two questions to ask on the discovery call to test those reasons, 3) one disqualifying signal to watch for.' Trap: AI-written cold emails are now pattern-matched by buyers and ignored. Use the model for research and post-call follow-up; write the cold outreach yourself.

Compounding task: turning legal-pad policy questions into plain-English explanations for employees. Daily prompt: 'An employee asked: [question]. Our policy says: [paste handbook section]. Draft a reply in plain English, 6th-grade reading level, that answers the question, points them to the exact handbook section, and tells them who to email for an exception.' Trap: employee data into public models. Use an enterprise tier or a tool with a signed BAA/DPA per your jurisdiction.

Compounding task: job-description writing and intake-call summarization. Daily prompt: 'I just had an intake call with a hiring manager. Notes: [paste]. Produce: 1) a JD draft following our template (paste template), 2) a one-screen 'manager debrief' I can send back to them, 3) five Boolean searches I can run on LinkedIn for this role.' Trap: bias in JDs (gendered language, ageist requirements). Run a tool like Textio or ask the model explicitly to flag biased terms.

Compounding task: pre-interview prep and skim summaries of long source documents. Daily prompt: 'I'm interviewing [name, role, org] tomorrow about [topic]. Their background: [paste bio + recent quotes]. Suggest: 1) five questions that would be hard for them to dodge, 2) two follow-ups I should be ready with, 3) a one-paragraph briefing on context I should know before the call.' Trap: AI-generated quotes are a career-ender. Models cannot interview people. Don't ask them to.

Compounding task: voice-locked variation generation. Daily prompt: 'Voice rules: [paste detailed voice doc with banned words, sentence-length targets, register, example paragraphs of yes/no]. Brief: [paste]. Output six variations of the headline + sub. Score each on a 1–10 against the voice rules and explain the score.' Trap: drift toward generic. Reseed the voice document every quarter; the model trains you to write like it as much as the other way around.

Compounding task: client briefs and post-shoot caption packs. Daily prompt: 'Wedding clients sent this questionnaire: [paste]. Draft: 1) a 200-word shot list grouped by location and time-block, 2) three caption-pack drafts (Instagram, blog, gallery) once the shoot is done — leave [SHOT DESCRIPTION] placeholders, 3) three follow-up email beats.' Trap: AI-rewriting client questionnaires loses their actual words. Quote the client verbatim where it matters.

Compounding task: moodboard-prose translation and critique passes. Daily prompt: 'Client said: [paste rambling brief]. Translate into a one-page creative brief with: audience, mood (three adjectives), reference styles (named designers/eras only — no invented references), constraints, deliverables, success criteria. Flag anything the client hasn't answered.' Trap: model invents reference designers. Ask explicitly for named, verifiable references and check them.

Compounding task: transcript-driven rough cuts and chapter markers. Daily prompt: 'Here is the transcript of a 60-minute interview: [paste]. Produce: 1) 8–12 chapter markers with timecodes and titles, 2) the three strongest 30-second clips with start/end timecodes and why, 3) a 120-word YouTube description.' Trap: timecode drift. Always verify the model's timecodes against the source; some models silently round or hallucinate.

Compounding task: show notes, episode summaries, and pull-quote extraction. Daily prompt: 'Episode transcript: [paste]. Produce: 1) a 100-word episode summary, 2) five pull-quotes with timestamps, 3) ten chapter markers, 4) three social caption variants in our voice [paste voice rules], 5) five SEO-relevant tags. Do not invent guest credentials — leave [VERIFY] placeholders.' Trap: misattributed quotes. Verify timestamps before publishing.

Compounding task: turning everything-in-your-head into written standard ops. Daily prompt: 'I do [process]. Here's how I do it, in messy form: [voice-memo transcript]. Convert into a clean SOP with: trigger, inputs, step-by-step (numbered), outputs, who can run this, common failure modes. Use plain English.' Trap: the SOP becomes outdated within a month and no one updates it. Schedule a quarterly re-prompt with current reality.

Compounding task: drafting plan summaries and client-meeting prep. Daily prompt: 'Client meeting tomorrow. Here is their plan-of-record: [paste]. Recent market context: [paste your house view]. Draft: 1) a one-page meeting agenda, 2) three questions to ask them, 3) two scenario walkthroughs in plain English (no investment advice, just narrative).' Trap: anything that looks like personalized investment advice generated by AI is a compliance issue under SEC Marketing Rule and FINRA supervision rules. Keep AI to logistics and narrative; the recommendation is yours.

Compounding task: pro forma narratives and permit-package narrative sections. Daily prompt: 'Here is the pro forma summary: [paste numbers]. Draft a 300-word investment-memo narrative aimed at LPs: project, market, sponsor edge, sources/uses, risk, exit. Do not invent comps or rents — use what I supply.' Trap: invented comparables. Models pattern-match to typical CRE deals and fabricate numbers. Lock them to your supplied data.

Compounding task: bid-letter drafting and RFI documentation. Daily prompt: 'Owner asked for a bid on [scope]. My estimator gave me these numbers: [paste]. Draft a clean bid letter: cover paragraph, scope inclusions, scope exclusions, clarifications, payment terms, validity period. Use my company's standard exclusions [paste].' Trap: model adds exclusions you don't actually want, or softens contract language. Read every line; this is a legal document.

Compounding tasks: menu-copy drafts, supplier emails, schedule-change narratives, incident reports. Daily prompt: 'Draft a polite email to my supplier: [situation]. Goal: [outcome]. Tone: firm but not adversarial; we want to keep the relationship.' Trap: AI politeness is the same flavor across every restaurant on the block. Edit until it sounds like you, not like a SaaS company.