Applying Smart, Not Wide: Three Prompts for the Live Search
WEEK 104 :: POST 3 :: CLAUDE
Directions Given To The A.I. This Week+
Instructions Given to each A.I. — Please provide 3 prompt variations that share this objective:
Each A.I. also received two static attachments: the blog post template (structure) and the authoring instructions (voice and standards). The text below is the week-specific assignment as sent — reflowed for the web; wording unchanged.
I'd like you to write this week's Ketelsen.ai post. Two files are attached: the blog post template (the structure to follow) and the authoring instructions (context, voice, and standards). Please read both before you begin, then produce the complete post in a single response.
This week's theme: "The Application Engine — Applying Smart, Not Wide."
This is Week 5 of an eight-week series on running a job search with AI. The reader now has assets (Week 3) and a ranked company list (Week 4). This week they go live — and the trap waiting for them is spray-and-pray: hundreds of applications, a trickle of responses, and morale falling with every silent rejection.
There is a delicious irony at the center of this week, and the post should lean into it — plainly, and without gloating, because some readers will have already sent the two hundred generic letters and this post is their way out, not their scolding: AI-generated application slop is now something recruiters actively watch for. An AI prompt blog is exactly the place to say this plainly: the losing move is using AI to mass-produce two hundred generic cover letters; the winning move is using AI for tailoring intelligence — reading one posting deeply, matching it against the reader's real record, and producing an application that could not have been written for any other job. Same tool, opposite outcomes, and the difference is the prompting — which is the thesis of this whole series, and this is the week to argue it in full rather than just assert it. This is the series' thesis week. The post should say, in as many words, what the series contract promises readers: use AI like an analyst, not a ghostwriter — because employers restrict AI in their own hiring for real legal and compliance reasons, because recruiters discard what reads machine-made, and because an application the reader cannot stand behind in the interview was never an asset to begin with.
By this point the reader is also juggling artifacts from four earlier weeks — a spec sheet, a master résumé, a dossier — and keeping that context straight across sessions is itself becoming work. The strongest posts can acknowledge that honestly: structured, reusable prompting is what keeps a multi-week project coherent.
The deliverable the reader should walk away holding: a working application pipeline — a tracked set of live, tailored applications aimed at the top of the Week 4 ranking, with outreach running alongside.
THE SERIES CONTRACT — identical every week; it binds every prompt you design. This series' tagline is its editorial contract: "Use AI like an analyst, not a ghostwriter." It is written for a reader in a market unsettled by AI itself — some readers are searching precisely because AI eliminated their last role. Write with that reader at the table: no AI-efficiency cheerleading, no automation jokes, no promises that AI will "do it for you" anywhere a human hiring decision is involved. And hold one line in every prompt: the AI is the reader's private analyst, coach, and sparring partner — it structures, researches, rehearses, and questions. It does not ghostwrite. Anything a hiring human will read or hear — résumé lines, cover letters, outreach messages, interview answers, negotiation emails, resignation letters — must end in the reader's own words and be true. Prompts should drive toward drafts the reader rewrites and owns, and should say so explicitly. Employers increasingly restrict how AI may be used in their own hiring decisions for legal and compliance reasons, and recruiters increasingly recognize — and discard — material that reads machine-written. A prompt that makes a reader look AI-generated hurts them twice. Posts that ignore this contract should expect to lose the week. Two practical notes. First: a standing “About this series” notice covering these same points is added to every published post automatically at publication — acknowledge the frame in your own voice where your week's prompt calls for it, but do not write a formal disclaimer block of your own, and do not open every post with the same acknowledgment paragraph: outside the weeks whose prompts explicitly carry the series frame, this contract lives in your tone and your prompt design. Second, the framing is POSITIVE: used this way — analyst backstage, reader on the page — AI is an advantage no hiring human will ever hold against your reader. Write like that is true, because it is.
The three prompts should help a reader:
- Decode the job description. What the posting actually says versus what it means: the must-haves versus the wish list, the responsibilities hiding behind buzzwords, the signals about level and team — so the reader decides in minutes whether this posting deserves a tailored application at all.
- Build one tailored application package. A résumé variant cut from the Week 3 master against this specific posting, and a cover letter grounded in the reader's actual specifics — concrete, short, and impossible to mistake for a template. The reader supplies the posting and their real materials; nothing gets invented.
- Run the full pipeline system. The operating rhythm: an application tracker with stages and dates, referral-outreach messages that respect the person being asked, informational-interview requests that are not stealth applications, and a follow-up cadence that is persistent without being unhinged.
At the advanced tier, the strongest version of this week is the pipeline as an operating system — decode, tailor, send, track, follow up, learn — where the tracker feeds back into targeting (which kinds of postings respond?) and outreach runs as a parallel lane rather than an afterthought. That loop is worth reaching for.
A constraint for this week. Response rates, referral-conversion multipliers, and "applications per hire" figures circulate everywhere and verify almost nowhere. No prompt may ask the AI to state response or conversion statistics as fact, and no outreach prompt may have the AI fabricate a relationship, a mutual contact, or credentials the reader does not have. Honest outreach written from the reader's real situation is both more effective and the only kind this blog will teach.
Design the prompts so the AI does what it is genuinely good at: deep reading of one document, precise matching against another, drafting from supplied specifics, and imposing system on a chaotic process. The reader supplies the postings, the master assets, and the truth; the AI supplies the decode, the tailoring, and the discipline.
Series dependency chain, for the Metadata block: Week 5 consumes Week 3's master résumé and positioning narrative (variants are cut from them) and Week 4's ranked dossier (effort goes to the top of the list). Week 5 produces the live application pipeline — consumed by Week 6 (interviews come out of the pipeline) and Week 7 (multiple live processes become negotiation leverage).
Because readers may arrive at this post without having read the earlier weeks, the prompts should work for someone with a résumé and a posting in hand, while making clear the engine runs far better fed by a real Week 3 master and a Week 4 ranking.
Three difficulty tiers as always — Beginner, Intermediate, Advanced — each a genuinely different approach to the same problem, not the same prompt at three lengths.
On examples: this is a career topic that touches every industry. The template lists tech startup / retail / freelance as suggested industry examples — those are marked MAY, and adapting them is expected here. A reader tailoring one application to a dream posting, another asking a former colleague for a referral without making it weird, and a third rescuing a pipeline that is all silence are the right kinds of contexts. Choosing them over the suggested business examples is correct behaviour and will not be scored against you.
A note on supplied figures. Anything marked `[SUPPLIED — use as given]` above came from Ketelsen.ai's own research brief. Use it freely — you are not fabricating by repeating it, and you will not be marked down for leaving it uncited. Do not attach an invented source to it. (No supplied figures this week — response-rate and referral-conversion numbers especially. If the sentence needs a statistic to land, rewrite the sentence.)
## BEFORE YOU SUBMIT — STRUCTURAL CHECK
(This block is identical every week. It exists because these specific items are the ones posts drop, and a dropped structural item costs compliance points for something that takes one minute to add.)
Your post is parsed by a script before any human reads it. Confirm all seven:
1. ☐ Response begins with `PLATFORM: <your name>` and `WEEK: 5` 2. ☐ `## Lead` present once, at the very top, before Variation 1 3. ☐ `## In one line` present in all three variations 4. ☐ `## What this prompt gives you` present in all three variations 5. ☐ `## The Prompt` present in all three variations, with the prompt in double quotes beneath it 6. ☐ `## Introductory Hook` and `## Current Use` present in all three variations (three of each — not one) 7. ☐ Every template heading written as `##`, none bolded instead; prompt breakdown is running text split on ` : `, with no `###` headings inside it
A complete post has 57 `##` headings. If your count is well short, a section is missing or was bolded instead of hashed.
One extra check this week: confirm no prompt asks the AI to state response or conversion statistics as fact, and no outreach prompt invites the AI to invent a relationship, mutual contact, or credential the reader does not actually have. Also confirm every prompt that produces a cover letter, outreach message, or any other text a hiring human will read ends with the reader rewriting and owning the final words (the series contract's final-pass rule) — this is the week that produces exactly the material recruiters screen for AI slop.
About this series. AI is reshaping work — for some readers, it's part of why you're searching at all. We don't pretend otherwise. And the hiring world is wary of AI-written material: many employers restrict how AI may be used in their own hiring decisions, and recruiters increasingly recognize — and discard — machine-written applications. So this series teaches a different approach: use AI as your private research analyst, interview coach, and thinking partner, while every word an employer sees or hears from you stays genuinely, verifiably yours. AI behind the scenes. You on the page.
You have the assets and you have the target list, so this is the week you go live — and the week most job searches quietly go wrong, because firing off two hundred near-identical applications feels like progress right up until the silence sets in. This post gives you three prompts at three depths: one that reads a single posting closely enough to tell you in five minutes whether it deserves your afternoon, one that matches that posting against your real record and shows you exactly where your evidence is thin, and one that runs the whole pipeline — decode, tailor, send, track, follow up, learn — as a system that gets sharper every week. There is an irony worth naming before we start. AI is what is currently burying recruiters in generic applications, and careful AI use is also the clearest way out of that pile. Same tool, opposite outcomes, and the entire difference is in how you prompt it. ---
The Posting Decoder
Know in five minutes whether a posting deserves your effort.
Most people read a job posting the way they read a menu at a restaurant they have already decided to eat at. They skim for the parts that sound like them, feel a small flare of hope or dread, and then either apply or close the tab. Almost nobody reads a posting the way the person who wrote it did — as a document assembled from a hiring manager's actual problem, a recruiter's shorthand, and a legal team's boilerplate, with the genuinely non-negotiable requirements sitting in the same paragraph as the ones somebody added because it seemed nice to have. Learning to separate those two things is the single highest-leverage skill in a job search, because it decides where your finite hours go. This prompt teaches your AI to do that reading for you, in about the time it takes to make coffee.
Why this matters now
Right now you are standing at the top of a funnel with more openings in front of you than you have weeks to pursue, and every hour spent tailoring an application to a posting you were never going to clear is an hour stolen from one you could have won. The decode step is what makes the rest of this week affordable. It also protects your morale, which is a real resource in a search: a candidate who has skipped forty postings on purpose feels in control, while a candidate who has applied to forty and heard nothing feels invisible. Same forty postings, opposite psychology. Do the reading first.
You are helping me read one job posting carefully before I decide whether to apply. I will paste the full text of the posting below. Do not write anything for me to send to anyone — I only want a close reading.
Give me these five sections, in this order:
1. Must-haves. The requirements that read as genuinely non-negotiable. Quote the exact wording from the posting for each one.
2. Wish list. The requirements that read as preferences, soft asks, or nice-to-haves. Quote the exact wording for each one.
3. Plain English. Rewrite each listed responsibility as one short sentence with no buzzwords, so I can see what the person in this job actually does all day.
4. Signals. What the posting suggests about seniority, team size, who I would report to, whether this role is new or a backfill, and what problem the team is trying to solve by hiring. Label every item here as inferred.
5. Unanswered. Three things this posting does not tell me that I would want to know before spending an afternoon on it.
Rules. Work only from the text I paste. Do not guess at salary, company size, culture, or funding if the posting does not state them. If a requirement is ambiguous about whether it is a must-have, put it in the wish list and say why you were unsure.
Finish with one line answering this question: based only on this posting, what kind of candidate is this written for?
Here is the posting:
[PASTE THE FULL JOB POSTING HERE]
How the AI reads this prompt
Practical examples from different industries
A warehouse operations supervisor eyeing a logistics coordinator role. The posting runs to nine hundred words and reads, on the surface, like it wants a degree and five years of WMS administration. Run through the decoder, the must-haves collapse to three: experience with a specific warehouse management system, availability for a rotating early shift, and forklift certification. The degree sits in the wish list next to "familiarity with continuous improvement methodologies." The plain-English pass reveals that four of the seven responsibilities are really about reconciling inventory counts between two systems that do not talk to each other. That is a job this reader has effectively been doing for two years without the title — and they were about to skip the posting over the degree line.
A high school teacher moving into instructional design. Corporate postings in this field are written in a vocabulary the reader has never used, which makes every one of them feel like a wall. The decode pass translates "stakeholder-driven curriculum development in an agile environment" into "you will build training in short cycles while a subject-matter expert keeps changing their mind," which is a recognisable description of teaching in a district that reshuffles standards every spring. The signals section flags a small team and a new role, inferred from the responsibility mix. The reader stops treating the posting as foreign and starts seeing which of their real experiences map onto it.
A freelance graphic designer weighing a return to salaried work. Six postings are open and all six sound similar. Running each through the decoder in sequence produces a comparison the reader could not have made by skimming: two postings list production output as the first three responsibilities and are effectively pixel-pushing jobs at a senior title, one is genuinely a brand-strategy role wearing a design job's clothes, and three are unclear enough that the unanswered-questions section fills with the same query — who owns the creative direction here? That question becomes the first thing the reader asks on any screening call, and it costs them nothing to have found it.
Creative use case ideas
Decoding a grant RFP.
Foundation funding calls are job postings with a different ribbon on them, full of must-haves buried in wish lists and language designed to sound open while being quite narrow. A small nonprofit can run an RFP through this prompt and find out in five minutes whether the funder actually funds organisations at their size and stage before writing eleven pages of application.
Reading a graduate program's admissions page.
The stated requirements and the actual expectations of an admissions committee are rarely the same document. The decode pass separates the hard gates from the preferences and surfaces the questions worth emailing a program coordinator about — which is a better use of a first contact than "am I a good fit?"
Sizing up a volunteer or board role.
Community organisations write role descriptions that badly understate the time commitment, usually not on purpose. Running the description through the plain-English pass turns "provide strategic oversight and support fundraising initiatives" into a realistic picture of monthly meetings plus donor calls, which is how you avoid saying yes to something you will resent by March.
Comparing two offers from the same company.
If you are being considered for two roles internally, decoding both postings side by side shows you which one is written for a specialist and which one is written for someone who will be asked to do a bit of everything. That distinction shapes the next three years of your career and is nowhere in either job title.
Reverse-engineering your own next posting.
Run the decoder on the job you want two steps from now, then read the must-haves as a curriculum. It is a considerably more honest skills plan than any generic career-path article, because it comes from what employers are actually asking for in your market this month.
Adaptability tips
Change the number of postings, not the depth. This prompt is built for one document at a time on purpose — paste three and the model starts blending them, and the quotes stop being reliable. If you want to compare postings, run the prompt separately and compare the outputs.
If you are changing fields, add a sixth section: "Translation for an outsider — which of these requirements have a different name in [YOUR CURRENT FIELD]?" This is the single highest-value modification for career changers, because it converts unfamiliar vocabulary into experience you already have.
If you are early in your career, swap the closing question for "what would make an applicant with less experience than this posting asks for still worth interviewing?" The answer tells you what to lead with.
If a company posts many similar roles, add "and tell me how this posting differs from a standard [ROLE TITLE] posting" to surface what is genuinely specific to this team rather than generic to the function.
To make the output shorter, ask for the must-haves and the closing question only. It runs in under a minute and is still enough for a first-pass filter on a long list.
Pro tips
Paste the posting from the company's own careers page rather than a job board. Aggregators frequently truncate the requirements section or strip the "about the team" paragraph, and those are exactly the parts this prompt reads hardest.
When the model lists a must-have you do not have, resist the instinct to close the tab. Ask a follow-up: "for each must-have, what would a hiring manager accept as a near-substitute?" You will learn quickly which gates are real and which are proxies for something you can demonstrate another way.
Save the decode output. In three weeks, when a recruiter finally calls about a posting you barely remember, that five-section summary is the fastest possible re-brief, and it is written in your own filing rather than the company's marketing voice.
Run the decoder on a posting for a job you already have. Seeing your current role described as a stranger would read it is unsettling in a useful way, and it tends to produce two or three résumé bullets you had never thought to write.
Prerequisites
The full text of one job posting, copied and pasted rather than linked — many AI tools cannot open a careers-page URL, and the ones that can may retrieve a stripped version. Nothing else. You do not need your résumé for this variation, which is deliberate: the decision this prompt supports is whether the posting is worth your time at all, and mixing your own materials in at this stage tends to make the reading optimistic. If you have been following this series, the Week 4 ranked list is a good source of which postings to run first, but the prompt works fine on a posting you stumbled across this morning.
Required tools
Any general-purpose AI assistant with a chat interface — Claude, ChatGPT, Gemini, or a comparable tool. The free tier of any of them handles this prompt without difficulty, since a job posting is a small document and the analysis is a single pass. No file uploads, plugins, or web access are required, because you are pasting the text directly. A notes app or document where you can keep the outputs is genuinely useful once you are running several postings a week.
Frequently asked questions
What if the AI puts something in the wish list that the posting clearly required?
Push back in the same conversation and ask it to justify the placement with the quote. Sorting a twelve-item requirements block into two buckets involves judgment, and judgment is sometimes wrong. The value of the quote requirement is precisely that you can audit each call in about two seconds, which is faster than doing the sort yourself and considerably faster than applying to the wrong job.
Is it cheating to have AI read a posting for me?
No, and it is worth being clear about why. Nothing produced here goes to an employer — this is analysis for your own decision-making, which is the kind of work AI is genuinely good at and the kind no hiring process has any interest in restricting. The line this series holds is about text a hiring human reads, and there is none of that here.
The posting is vague and the output was vague. Now what?
That is a finding, not a failure. A posting that will not say what the job involves is often a role the team has not defined yet, which can mean unusual freedom or unusual chaos. Use the unanswered-questions section as the agenda for a recruiter screen, and decide after you have answers rather than guessing from the text.
Should I run this on every posting I find?
Run it on anything you are seriously considering and on anything you are about to dismiss for a single requirement. Those two categories are where the prompt pays for itself. Postings you are certain about in either direction do not need a decode, and building a ritual out of it will slow you down.
Can I ask it to score the posting or give me a match percentage?
You can, but treat any number it produces as decoration rather than data. A match score sounds precise and is generated from nothing more than the same reading you just got in prose. The closing question — what kind of candidate is this written for — carries the same information and is harder to fool yourself with.
Recommended follow-up prompts
The Tailoring Pass
(Variation 2 below). The natural next step once a posting clears the decode: take the must-haves you just extracted and test them against your actual record before you write a word of an application.
A screening-call question builder.
Feed the unanswered-questions section back in and ask for the three questions ranked by how much they would change your decision, plus how to ask each one without sounding like you are interrogating a recruiter on a first call.
A market-pattern read.
After decoding ten postings in a field, paste the ten must-have lists together and ask which requirements appear in nearly all of them and which appear in only one or two. The common set is your real skills gap; the rare set is noise you can safely ignore.
Tags and categories
Tags:
job search, job postings, application strategy, career change, reading comprehension, prompt basics, filtering, hiring signals
Categories:
Career & Job Search, Beginner Prompts
Citations
NOT APPLICABLE
---
The Tailoring Pass
Match one posting to your real record, evidence first.
Here is the uncomfortable thing about tailoring. Everyone knows they should do it, most people believe they are doing it, and what they are usually doing is swapping the company name and moving one bullet up. Real tailoring is not a cosmetic operation on a document — it is an argument, built requirement by requirement, that connects what this specific employer said they need to something you have actually done. The reason people skip it is not laziness. It is that the argument is hard to build from the inside, because you cannot see your own record clearly enough to know which of your forty accomplishments happens to be the one that matters here. That is a matching problem across two documents, and matching across documents is something an AI is genuinely, unglamorously good at. This prompt puts it to work on the only application that counts today: this one.
Why this matters now
The application you send this week is competing against a pile that is measurably more machine-made than it was two years ago, which has changed what recruiters look for. Fluent, confident, oddly weightless prose is now a signal to move on, and specificity is the thing that survives the skim. That is genuinely good news for you, because specifics are the one thing an AI cannot manufacture and you can — you were there, you know what actually happened, you remember the number. What this prompt does is find where your specifics line up with their stated needs, and hand you the raw material rather than the finished text. You still write the letter. That is not a limitation of the method; it is the method.
Act as a hiring-side reviewer helping me tailor one application to one posting. I will give you two things: the job posting, and my real career record. Work only from the material I give you. If I have not given you evidence for something, say so — never fill a gap with a plausible-sounding claim.
Work through four steps in order.
Step 1 — Evidence map. For each must-have in the posting, find the strongest supporting evidence in my record. For each one give me three things on one line: the requirement, the evidence quoted directly from my material, and a strength rating of strong, partial, or none. Leave the none rows empty of evidence. Do not soften them.
Step 2 — Résumé cut. Working only from bullets that already exist in my record, tell me which to keep, which to cut, and which to rewrite for this posting, and in what order they should appear. For each rewrite, propose new wording that uses only facts already present in my material. Target length: [ONE PAGE / TWO PAGES]. Explain each cut in six words or fewer.
Step 3 — Cover letter raw material. Do not write a cover letter. Instead give me: the single strongest specific from my record for this job and why it is the strongest; the one thing in the posting that suggests what this team is actually worried about; and three concrete details from my record I could build a letter around. No adjectives about me, no enthusiasm language, no praise of the company.
Step 4 — Gaps. List every none row from Step 1. For each, tell me whether to address it head-on, work around it, or ignore it, and give me the reasoning in one sentence.
Rules that apply throughout. Never invent an employer, title, date, metric, tool, or outcome. If a number would strengthen a bullet and I have not supplied one, write [NEED NUMBER] rather than estimating. If you are unsure whether something in my record supports a requirement, say you are unsure instead of resolving it in my favour. Keep your tone plain — I will be writing the final words myself and I need material, not polish.
Here is the posting:
[PASTE THE FULL JOB POSTING HERE]
Here is my record — résumé, project notes, anything factual:
[PASTE YOUR MASTER RÉSUMÉ AND SUPPORTING MATERIAL HERE]
How the AI reads this prompt
Practical examples from different industries
A registered nurse applying to a clinical informatics role at a regional health system. The posting's must-haves include experience with a named electronic health record platform, a background in workflow analysis, and a clinical licence. The evidence map returns strong on the licence and the platform, quoting seven years of daily documentation in it, but only partial on workflow analysis — the only supporting item is a line about serving on a unit committee. Step 4 rules that gap as address head-on, reasoning that the committee work is real and specific but currently invisible on the résumé. The reader spends the afternoon writing two honest bullets about a discharge-documentation change that committee actually shipped, which is a far better use of the time than another round of polish.
A small-agency account manager targeting a nonprofit development role. The record is full of client-retention numbers; the posting is full of language about donor stewardship and board reporting. The map finds three strongs the reader had not connected — a renewal process, a quarterly reporting rhythm, and a difficult account they saved — and one none, on grant writing, which the reader has never done. Step 4 rules that one as work around, because the posting lists it fourth among five responsibilities and the top three are all relationship management. The reader stops apologising for the grant gap and builds the letter around the account they saved.
A laid-off software engineer applying after nine months out of work. The record is strong and the gap in the timeline is the anxiety. Running the pass produces something useful and unsentimental: the evidence map comes back almost entirely strong, and the only none is on a framework introduced eighteen months ago. Seeing the map makes the shape of the real problem clear — this is not a credibility problem, it is a recency problem attached to exactly one item. Step 4 rules it address head-on, and the reader spends a weekend building something small and real in that framework rather than another week rewriting a summary paragraph nobody reads.
Creative use case ideas
Preparing for an internal promotion conversation.
Run the pass with your company's own job architecture document for the next level up as the posting. The evidence map becomes the agenda for a conversation with your manager, and the none rows become a development plan written in your employer's own vocabulary rather than yours.
Auditing a résumé you have not touched in three years.
Pick a posting for the job you actually want and run the pass purely diagnostically, with no intention of applying. The gaps you find are your next twelve months, and they arrive as a short list rather than a vague feeling of falling behind.
Applying to a competitive volunteer or fellowship position.
Community foundation board seats, coaching roles, artist residencies, and public library programs all publish requirements and all attract applications that read identically. The same evidence discipline applies, and almost nobody brings it to a volunteer application.
Helping someone else apply.
The pass is easier to run on a friend or a family member than on yourself, because you can read their record without the emotional weight. Sitting with a partner or a recently-graduated relative and running one posting together is a genuinely good hour, and the none rows are much easier to hear from a screen than from a person.
Testing a career pivot before committing.
Run the pass against three postings in the field you are considering. If the evidence maps come back with the same nones every time, you have found the actual barrier to entry, and you now know whether it is a course, a project, or a whole credential — before you have spent six months applying blind.
Adaptability tips
Feed it more than a résumé. The evidence map gets substantially better when the record you paste includes project notes, old performance reviews, and the positioning narrative from Week 3, because a résumé is already compressed and compression is where the specifics went. Anything factual about your work is fair material.
If you are changing fields, add a line to Step 1: "for any none row, tell me whether something in my record is a functional equivalent under a different name, and say plainly if nothing is." The escape hatch at the end matters — without it, the model gets creative about equivalence.
Adjust the strictness by changing the strength vocabulary. Replacing strong, partial, none with a stricter set — direct evidence, adjacent evidence, no evidence — produces a harsher and often more accurate map when you are applying above your current level.
For a posting you are borderline on, add a fifth step: "given this map, tell me whether this application is worth two hours, and what would have to be true for the answer to change." It is a good corrective when you are emotionally attached to a company.
If your material is long, run Step 1 alone first and read the map before letting the model proceed. The map is the decision; the rest is execution, and there is no point executing on a posting the map just talked you out of.
Pro tips
Run the pass twice with different postings for similar roles at different companies, then compare the two evidence maps. The requirements that come back strong both times are your actual market position, and that phrase belongs at the top of your résumé.
When Step 3 names what the team is worried about, check it against the decode from Variation 1 before you build a letter on it. Two independent reads agreeing is a much stronger signal than one read asserting.
Keep every [NEED NUMBER] bracket in a running list. Most of them can be answered in one sitting from old email, an old dashboard, or a former manager's LinkedIn recommendation, and a résumé with real numbers in it is a different document from one without.
Before you write the letter, close the AI window. Write the first draft from the three concrete details in Step 3 and nothing else, in your own voice, badly if necessary. Then bring it back for a read against the posting. Drafting first and editing with help produces writing that sounds like you; the reverse order almost never does.
Prerequisites
You need one decoded posting and your real career record in text form. The record should be as complete as you can make it — the Week 3 master résumé and positioning narrative are the ideal input, since the entire point of building a master document is that variants get cut from it rather than written fresh each time. If you have not built one, a current résumé plus any notes about projects, numbers, and outcomes you can assemble in twenty minutes will work. Be prepared for the map to come back with nones. That is the prompt functioning correctly, and it is considerably better to find them now than in a screening call.
Required tools
A general-purpose AI assistant with a context window large enough to hold a full job posting and your complete career record at once — every current mainstream tool clears this comfortably, but note that free tiers occasionally truncate long pastes, and a silently truncated résumé produces a confidently wrong evidence map. Paid tiers with file upload make the paste step easier if your record lives in a document. A word processor for the résumé variant, and somewhere to keep your [NEED NUMBER] list.
Frequently asked questions
Why won't it just write the cover letter? I would edit it.
Because the editing almost never happens at the depth it would need to. A fluent draft anchors you to its structure and its claims, and what comes out the other side is a lightly personalised version of a generic letter — which is the exact artifact recruiters have learned to spot. Building from three concrete details takes longer the first time and produces something that could not have been written for any other posting, which is the whole objective.
Is it dishonest to use AI to reorder my own résumé?
No. Nothing in Step 2 creates a claim about you; it decides which true things to show and in what order, which is what tailoring has always meant. The line worth holding is between selection and invention. Selecting from your real record is judgment, and it is yours regardless of what helped you see the options.
The evidence map says partial on almost everything. Should I apply?
Probably yes, and this is worth internalising. Postings describe an ideal that frequently does not exist, and a candidate with strong evidence on the top three must-haves and partials below is a normal, competitive applicant. Use Step 4's rulings to decide what to address rather than reading a wall of partials as a verdict.
Can I paste my record once and reuse it across many postings?
Within a single conversation, yes, and it is the efficient way to work through a shortlist — paste the record once and then feed postings one at a time. Across sessions you will have to paste it again, which is one of those small frictions that quietly makes a multi-week search harder than it should be. Keeping your master materials in one file you can drop in cleanly is worth the twenty minutes it takes to set up.
What if I disagree with a strength rating?
Say so and ask it to re-examine that row against your material. Disagreement is useful data — if you believe something is strong evidence and a hiring-side reader does not see it, the more likely explanation is that your résumé is under-describing it, not that the reader is wrong. That is a fixable problem, and finding it is worth the argument.
Recommended follow-up prompts
The Pipeline Operating System
(Variation 3 below). Once you can build one tailored application, the constraint becomes running ten of them without losing track, and that is a systems problem rather than a writing problem.
A gap-closing plan.
Feed the none rows back in and ask for the fastest credible way to convert each one from none to partial, ranked by time required — with the explicit instruction that credentials you do not hold are never an acceptable answer.
An interview pre-mortem.
Take the finished tailored résumé and ask which three lines a skeptical interviewer would probe hardest and what evidence you would need ready for each. It is the natural bridge into Week 6, and it is best done the day you send the application, while the material is fresh.
Tags and categories
Tags:
job search, résumé tailoring, cover letters, evidence mapping, application strategy, career change, document matching, honest AI use
Categories:
Career & Job Search, Intermediate Prompts
Citations
U.S. Equal Employment Opportunity Commission, technical assistance document on assessing adverse impact in software, algorithms, and artificial intelligence used in employment selection procedures under Title VII of the Civil Rights Act (May 2023) — background on why employers increasingly constrain AI in their own hiring processes.
New York City Local Law 144 of 2021, governing automated employment decision tools, in effect since July 2023 — an example of a jurisdiction-specific rule requiring bias auditing and candidate notice for automated hiring tools.
---
The Pipeline Operating System
Run your search as a system that learns each week.
By week three of a serious search, the failure mode stops being the writing and starts being the bookkeeping. You have eleven applications out, four of which you cannot remember the exact version you sent; two former colleagues you meant to message and did not; a company you followed up with twice and another you followed up with never; and a growing suspicion that you are working hard without being able to say what is working. This is not a discipline problem. It is a systems problem, and it is the specific kind of systems problem AI handles well: high-volume, rule-governed, requiring perfect recall of state and zero creativity. The prompt below sets up a pipeline you operate — decode, tailor, send, contact, track, follow up, learn — where each week's results change the next week's targeting. It is the difference between applying to jobs and running a search.
Why this matters now
The reason this matters now rather than later is that the learning loop only works if it starts collecting data before you have sent everything. A pipeline you set up in week one tells you by week four that postings from a certain company size respond and postings from another do not, and you redirect while you still have runway. A pipeline you set up after eighty applications tells you the same thing far too late to act on. There is also a quieter benefit worth naming: an operating rhythm gives structure to weeks that otherwise dissolve, and a search where you can see the machine working is far easier to sustain than one that feels like shouting into a void.
Act as the operations lead for my job search. You run the system behind the search. You do not write anything a hiring human will read as finished text — every draft you produce is raw material I rewrite in my own words before it goes anywhere. Hold that line even if I ask you to relax it.
Begin by asking me for a state block containing: my target roles, my top-ranked target companies, my current live applications with their stage and last-contact date, my available hours per week, and any deadlines. If I paste an existing state block, read it and continue rather than starting over.
Then build and run the following system.
1. Pipeline stages. Use these six: Decoded, Tailored, Sent, Contacted, In process, Closed. For each stage define the entry condition, the one action that moves an item forward, and the maximum number of days an item may sit there before I either act or close it. Tell me what a healthy distribution across stages looks like given my weekly hours, and treat any stage holding more than its share as a bottleneck to name.
2. Tracker schema. Specify the exact fields I should keep for each application, including target company, role, source, date sent, tailoring notes, referral status, next action, and next action date. Keep it to fields I will actually maintain. Tell me which single field is most likely to be the one I stop filling in, and how to make it cheap enough that I do not.
3. Weekly rhythm. Design an operating week around my available hours, split into decoding, tailoring, outreach, and follow-up blocks, with an explicit rule for what gets dropped first in a bad week. Assume some weeks will go badly.
4. Outreach lane. This runs in parallel with applications, not after them. Give me three message types as raw material I will rewrite: a referral request to someone I actually know, a request for a short informational conversation, and a post-application follow-up. Rules for all three. Never invent a relationship, a mutual contact, a shared employer, or a credential I have not told you about. If a message needs a connection I have not supplied, leave a bracket for me to fill or tell me the message should not be sent. An informational request must be genuinely informational — if I want the job, the message should not pretend otherwise. Every draft ends with the line REWRITE IN YOUR OWN WORDS BEFORE SENDING.
5. Follow-up cadence. Define timing rules for follow-up after applying, after a referral request, and after an interview, with a hard stop after which I move on. Tell me what each follow-up should contain so that it adds something rather than only asking again.
6. Weekly review. I will paste my tracker to you each Friday. Read only my own data. Do not tell me typical response rates, benchmark conversion figures, or how many applications a search usually takes — those numbers are unreliable and irrelevant to my situation. Instead tell me: which parts of my pipeline moved this week and which did not; what the items that moved have in common that the stalled ones do not; which of my targeting assumptions the week's evidence contradicts; and the one change to next week's targeting or approach that follows from the evidence. If the sample is too small to support a conclusion, say so plainly rather than manufacturing a pattern.
Output format. Give me the system as six clearly labelled sections. End with a compact state block I can copy and paste back at the start of the next session so we do not lose context.
Here is my starting information:
[PASTE YOUR STATE BLOCK, OR ASK ME AND I WILL FILL IT IN]
How the AI reads this prompt
Practical examples from different industries
A restaurant general manager moving into food-service operations for a hospital system. The target list is short — nine employers in one metro area, and most of them post rarely. The pipeline that comes back is built for that constraint: the tracker gains a field for posting-watch dates, the weekly rhythm shifts most of the hours from decoding to outreach, and the review question that matters becomes which of the nine have anyone in them who will talk. By week three the review flags that four applications sent through a general portal have all gone silent while the two sent after a conversation both moved, and the targeting change writes itself.
A mid-career project manager with fifteen hours a week around an existing job. The scarce resource is time and the failure mode is a bad week wiping out a month of momentum. The system's degradation rule turns out to be the whole value — decoding gets dropped first, outreach never does — and the day limits on each stage surface three applications sitting in Sent past their follow-up window that had simply fallen out of memory. The Friday review, running on the reader's own thirteen rows, notices that every response so far has come from postings under two weeks old, which changes what gets decoded on Monday.
A recent graduate with no network and a lot of hours. Here the outreach lane is the uncomfortable part, and the honesty rules in the prompt are what make it usable — with no mutual contacts to invent, the drafts have to be built from real, small connections like a shared program, a professor's course, or a public talk the reader actually attended. The tracker's referral-status field stays mostly empty for two weeks, which the review names as the bottleneck rather than letting a rising application count disguise it. The correction is fewer applications and four real conversations a week.
Creative use case ideas
Running a freelance or consulting pipeline.
The structure is nearly identical — decoded becomes qualified lead, tailored becomes proposal, and the weekly review answers which kinds of prospects convert. Anyone who has ever lost a client because a follow-up fell through a crack has a use for this.
Managing a graduate school or fellowship application season.
Six programs, six sets of requirements, three recommenders who need reminding at different intervals, and deadlines that do not move. The stage limits and follow-up cadence transfer without modification, and the outreach lane becomes contact with current students and faculty.
Coordinating a community organisation's volunteer recruitment.
A neighbourhood group chasing twenty volunteers for a spring program has the same shape of problem — a list of people at different stages of yes, a follow-up cadence that has to be persistent without being irritating, and no memory beyond whoever is holding the clipboard. The tracker schema and the parallel-lane thinking apply directly.
Running a house hunt or a rental search in a tight market.
Listings decoded, applications tailored to what a given landlord is worried about, follow-ups cadenced, and a weekly review that notices which neighbourhoods actually respond. It is an unromantic use of a job-search system and it works remarkably well.
Managing a creative submission calendar.
Writers, illustrators, and musicians submit to markets with response windows measured in months and rules about simultaneous submissions. The pipeline handles the state-tracking problem that spreadsheets usually handle badly, and the weekly review looks at which kinds of markets respond to your kind of work.
Adaptability tips
Scale the whole system by hours rather than by ambition. Tell it eight hours a week and it should return a smaller pipeline with tighter stage limits, not the same pipeline with an unrealistic schedule attached. If the design it gives you assumes more time than you have, say so and make it rebuild — a system you cannot run is worse than no system.
If you already track applications in a spreadsheet, paste the header row into the state block and ask it to work within your existing fields, adding at most two. Migration friction is the most common reason a better system loses to a worse one already in use.
For a search concentrated in one industry, add a seventh section asking for the two or three signals specific to that field that should be tracked as fields — clearance status, licence reciprocity, portfolio version sent, whatever your world runs on.
If you are searching while employed, add a confidentiality rule to the outreach lane specifying which of your contacts may be approached and which may not, and require the model to check any draft against it. It is much better as a stated constraint than as something you remember at the moment of sending.
To run the review more often than weekly, shorten the window and lower the bar for calling something a pattern to zero — daily data will never support a conclusion, and the honest answer most days is that nothing has changed.
Pro tips
Keep your state block in a plain text file rather than in chat history. It is the memory of the entire project, it takes ten seconds to update after each session, and it makes the system portable if you switch AI tools halfway through the search.
When the Friday review names a targeting change, write it into the state block as a dated line rather than only acting on it. Six weeks in, that list of dated changes is the most accurate account of your search that exists, and it is genuinely useful in an interview when someone asks what you have been doing.
Run the review before you look at the week's rejections rather than after. The order matters more than it should — reading the data first produces analysis, reading the rejections first produces a mood, and the mood then writes the analysis.
Give the outreach lane a weekly minimum and no maximum. Applications have a natural pull because they feel productive; conversations do not, and they are consistently the lane that moves things. A floor is the only reliable defence.
Prerequisites
You need something to run the system on: a set of target companies, ideally the ranked dossier from Week 4, and at least a few live or imminent applications. You also need a genuine estimate of your weekly hours, which is worth calculating rather than guessing, since the whole design scales from it. Somewhere to keep the tracker — a spreadsheet, a notes file, a simple board — and somewhere to keep the state block. Familiarity with the Variation 2 tailoring pass helps, because Tailored is a real stage in this pipeline and it needs to mean something specific rather than "I changed the company name."
Required tools
A general-purpose AI assistant capable of holding a long structured conversation and returning consistent formatted output across sessions. A spreadsheet or lightweight project board for the tracker itself — the AI designs the schema and reads the data back, but it should not be your system of record, because you will need the tracker on a phone in a parking lot before an interview. A plain text file for the state block. Tools that support saved projects or persistent instructions make the paste-back step lighter, though the state block is what actually carries the continuity and works everywhere.
Frequently asked questions
Isn't this a lot of overhead for a job search?
The setup runs about an hour once, and the weekly review runs about fifteen minutes. Against that, the cost of not having it is the applications that go stale in Sent, the referral you meant to ask for in March, and eight weeks of effort you cannot analyse because you never recorded what you did. The overhead is real and it is small compared to what it recovers.
Why can't the AI just keep the tracker for me?
Because it has no memory between sessions unless you give it one, and a tracker that lives in a chat window is a tracker you will lose. Keeping the data in a file you own and pasting it in for the weekly read is the arrangement that actually survives eight weeks. It also means the record is yours if you decide to change tools, which people frequently do mid-search.
The weekly review keeps telling me the sample is too small. Is it broken?
It is working correctly, and this is the clause doing its job. With nine applications and one response there is no reliable pattern to find, and a model that invented one would send you re-targeting on noise. Expect real signal somewhere around the third or fourth week of consistent volume, and until then use the review to check whether you are actually doing the things the rhythm calls for.
How do I keep outreach from feeling like begging?
Mostly by asking for something specific and small, and by not disguising your intent. The prompt's rule about informational requests is the important one here — a genuine question about how a team works is a reasonable thing to ask a stranger, and a job request wearing that costume is not, and everyone can tell the difference. Ask a real question, keep it short, and make it easy to say no.
Can I use this if I have not done Weeks 3 and 4?
Yes. Give it your target companies as best you can name them and your current applications, and the system will run. It runs considerably better on a real ranked list, because the pipeline's entire premise is that finite effort goes to the top of a considered list rather than to whatever appeared in a feed this morning. If you are working without one, the weekly review becomes more valuable, since it is the only thing correcting your targeting.
Recommended follow-up prompts
An interview preparation pass built from the pipeline.
Take the tailored materials for a specific live application and generate the questions a skeptical panel would ask, along with which of your own claims need evidence ready. This is where Week 6 picks up, and the pipeline is what feeds it.
A pipeline post-mortem.
After four weeks, paste the full tracker and ask which of your original targeting assumptions the evidence contradicted, restricted strictly to your own data. It is a harder read than the weekly review and it tends to be the one that changes the search.
A parallel-process planning prompt.
When two or three processes go live at once, ask for a sequencing plan that keeps timelines aligned without misleading anyone about where you are — which is the honest foundation for the negotiation leverage Week 7 builds on.
Tags and categories
Tags:
job search, pipeline management, application tracking, networking, referral outreach, follow-up cadence, systems thinking, reusable prompts, multi-session context
Categories:
Career & Job Search, Advanced Prompts
Citations
Illinois Artificial Intelligence Video Interview Act, 820 ILCS 42, in effect since January 2020 — an early state-level example of statutory constraints on AI use within employer hiring processes.
European Union Artificial Intelligence Act (Regulation 2024/1689), which classifies AI systems used in employment, worker management, and access to self-employment as high-risk and subject to additional obligations — relevant context for why employers are cautious about AI in hiring workflows.
Anthropic, Claude prompt engineering documentation (docs.claude.com), on role assignment, explicit constraints, and structured output — general technique references underlying the design of this variation.
---
Which of the three should you use?
The three prompts here are not one idea at three sizes; they are three different operations that happen to sit next to each other in a week. The Posting Decoder reads one document and returns a decision. The Tailoring Pass reads two documents against each other and returns an argument. The Pipeline Operating System reads no documents at all and returns a process. If you are wondering which to start with, the honest answer is the one matching the problem you actually have this week, not the one matching how experienced you feel.
Start with the decoder if your list is longer than your hours, which describes almost everyone in week one of a search. It is the cheapest of the three by a wide margin, it needs nothing but a posting, and it is the only one that will talk you out of work. Move to the tailoring pass when you have a posting you want badly and a real record to test against it — that is where the evidence discipline earns its keep, and where the difference between an application that sounds like everyone and one that could only have been written by you gets decided. Reach for the pipeline system when the bookkeeping has become the bottleneck, which usually announces itself as the moment you cannot remember whether you followed up with someone.
They also chain, and the chain is the intended way to run the week. A decode produces the must-have list that the tailoring pass maps evidence against; the tailoring pass produces a tracked application that enters the pipeline at Sent; the pipeline's weekly review sends you back to the decoder with better targeting than you had on Monday. What is consistent across all three is where the line sits. The AI reads, matches, structures, and questions. The specifics come from your record and the final words come from you — not as a compliance gesture, but because those are the only words that hold up when a hiring human reads them and the only ones you can defend, out loud, in a room, three weeks from now.
TAGS:
