§ Wiles et al. 2025 · n=480,948 · Management Science · +8% hires, +10% wages

Ten prompts. Eight worksheets. And the two risks nobody selling prompt packs will mention.

The strongest field experiment on AI-assisted job applications — n = 480,948 — is the reason this kit exists. It's also the reason Part 3 leads with what the study did not test, and with the two failure modes that fall on non-native English speakers and users of the wrong model. That briefing is Part 3. Prompts and worksheets are Parts 1 and 2.

terminal Part 1 · The Prompt System

Ten prompts. One context block.

Paste the context block once. Every prompt below assumes the model already has it. Every one names what it does, gives you the exact text, and closes with a check — the one thing you verify before shipping the output.

context.md
Context for everything that follows in this conversation.

I am job-hunting. My target role is [TARGET ROLE] in [INDUSTRY].
My current or most recent role is [CURRENT ROLE].
I have [N] years of experience.
The seniority I am aiming for is [junior / mid / senior / lead / director].

Three rules for this whole conversation:
1. Never invent a number, a client name, a company, or a result.
   If you need a fact I have not given you, write [I NEED THIS FROM YOU]
   and keep going.
2. Do not use the words: passionate, driven, results-oriented, dynamic,
   seasoned, proven track record, detail-oriented, team player.
3. Write at a plain reading level. Short sentences. No decoration.
01-headline-archetype.md
DoesDecides which of the book's three headline archetypes fits you, then writes three headlines in that formula.
You are a technical recruiter who has screened thousands of LinkedIn
profiles for [INDUSTRY]. You are blunt and you do not flatter.

Step 1. Decide which archetype fits me. Say why in one sentence.

- PRACTITIONER — same function, more senior.
  [Target role] · [Signature skill] · [Metric or outcome]
- BRIDGE — changing function or industry.
  [Prior role] → [Target role] · [Bridging skill] · [Proof]
- SPECIALIST — narrow niche, selling depth.
  [Niche title] · [Signature method] · I help [audience] [outcome]

Step 2. Write THREE headlines using only that archetype's formula.

Rules:
- Hard limit 220 characters. Show the character count for each.
- Each must contain one number, tool, or credential.
- Use only facts I give you below.
- After the three, list what you'd need from me to make them stronger.

My current role:
My target role:
Three things I've done that I can prove with a number:
Check Every number in the output must be one you supplied. If a figure appears that you didn't provide, delete that option.
See it run · one real headline before and after
Before · 148 chars
Passionate marketing professional with a proven track record of driving results, dynamic team player, seasoned strategist across B2B SaaS.
After · Practitioner · 143 chars
Senior Product Marketing Manager · B2B SaaS · Rebuilt launch process from 6 weeks to 11 days across 4 releases in 2025
02-about-hook-proof-cta.md
DoesBuilds the three-paragraph About structure, with the hook engineered to survive the "…see more" fold.
You are writing the About section of my LinkedIn profile using this
exact structure. Do not deviate from it.

PARAGRAPH 1 — HOOK. One or two sentences, maximum 220 characters total,
because LinkedIn truncates after roughly that point and everything after
it is invisible unless someone clicks. It must state what I do and for
whom, concretely. It may not begin with "Passionate," "Experienced,"
"As a," or "With over X years."

PARAGRAPH 2 — PROOF. Exactly three lines. Each line is one accomplishment
containing one number I have given you. No adjectives.

PARAGRAPH 3 — CTA. One sentence naming exactly who should contact me and
about what, plus how to reach me.

Then show me PARAGRAPH 1 alone, so I can see what a recruiter sees before
they click "see more."

Use only what I give you below.

What I do and for whom:
Three accomplishments with numbers:
Who I want to hear from:
How to reach me:
Check Read Paragraph 1 alone, out loud. If it could sit on a stranger's profile without changing a word, it's too generic. Send it back with "make the hook specific to me."
See it run · Paragraph 1 (the fold hook) before and after
Before · 214 chars
As a passionate and results-driven professional with over 8 years of experience across cross-functional teams, I bring a proven track record of delivering excellence in fast-paced, dynamic environments.
After · 187 chars
I write GTM launch plans for B2B SaaS at $10–50M ARR. Four launches in 2025 shortened time-from-brief-to-live from six weeks to eleven days, without adding headcount.
03-metric-excavation.md
DoesFinds numbers hiding in work you thought was unquantifiable. Run this before Prompts 1, 2, and 4.
You are an interviewer who is skeptical of vague accomplishments.

I am going to describe something I did at work in plain language, with
no numbers, because I don't think it had any.

Your job is to ask me up to eight short questions that might surface a
number hiding in it. Consider: time saved, time to complete, frequency,
volume, headcount affected, error rate, cost, revenue, retention,
adoption, before-versus-after states, and how many people did this job
before versus after.

Ask the questions one at a time. Wait for my answer before the next one.

Do not write a resume bullet. Do not suggest numbers. Only ask.

When we're done, list every number I gave you and mark each one:
- HARD (I can document it)
- SOFT (a defensible estimate)
- WEAK (I'm guessing — do not use this on a resume)

Here's what I did:
Check Anything the model marks WEAK stays off the resume. A number you can't defend under questioning is worse than no number.
See it run · what a real dig turns up
Before · you'd have written
Ran onboarding for new hires.
After · what the dig surfaced
HARD · onboarded 27 new hires in 2024 (headcount doc) HARD · shortened time-to-first-PR from 14 to 6 days across 12 engineers (calendar) SOFT · roughly 40% fewer week-two "how do I…" Slack pings (estimate, not measured) WEAK · "team was happier" · do not use
04-bullet-xyz.md
DoesTurns responsibility statements into the book's XYZ formula.
Rewrite each bullet I give you into this formula:

"Accomplished [X] as measured by [Y] by doing [Z]."

You may reorder the clauses so it reads naturally, but all three parts
must be present in every bullet.

Rules:
- Start with a past-tense action verb. Never "Responsible for."
- Maximum 2 lines when printed.
- Every bullet contains a number I gave you. If a bullet has no number,
  output it as: [NEEDS A METRIC — run the Metric Excavation on this one]
- Cut every adjective and adverb.
- Do not merge two of my bullets into one.

For each bullet, show: BEFORE, then AFTER, then one line on what changed.

My bullets:
Check Read each AFTER out loud. If it takes longer than eight seconds, it's too long.
See it run · one bullet, before → after
Before
Responsible for improving the customer onboarding process and mentoring junior team members to ensure a smooth transition for new users.
After · XYZ
Cut time-to-first-value from 14 days to 6 (57% faster) by rewriting the onboarding checklist and pair-shipping it with 4 CSMs across Q3 2024.
05-ats-keywords.md
DoesPulls the vocabulary a specific job posting actually uses, so you can mirror the terms you genuinely match.
Below is a job posting. Extract the vocabulary, do not summarize it.

Produce four lists:

1. HARD REQUIREMENTS — tools, certifications, languages, platforms
   named explicitly.
2. REPEATED PHRASES — any noun phrase appearing more than once. Note
   the count.
3. TITLE VARIANTS — every way the posting names this role or adjacent ones.
4. SOFT REQUIREMENTS — behavioural phrases, quoted exactly as written.

Then produce a table with three columns:
TERM | WHERE IT SHOULD APPEAR (headline / About / skills / bullet) | EXACT WORDING TO USE

Rule: use the posting's exact wording, not a synonym. If the posting says
"stakeholder management," do not write "cross-functional collaboration."

Do not tell me to claim anything. I will decide what I actually match.

Job posting:
Check Cross out every row you can't back up with real experience. Mirroring language you match is preparation. Mirroring language you don't is a story that collapses in the first interview.
See it run · a Senior PMM posting, extracted
Before · what most people copy
"Cross-functional collaborator" · "strategic thinker" · "own the roadmap" · "drive impact"
After · what the posting actually said
HARD · Marketo, Salesforce, ABM REPEATED · "positioning and messaging" (×4), "sales enablement" (×3) TITLE VARIANTS · Product Marketing, PMM, Sr. PMM, GTM Manager Table: "positioning and messaging" → About + first bullet · use exact phrase
06-stranger-test.md
DoesReads your profile the way a recruiter with 250 applicants and no patience would.
You are a hiring manager for [TARGET ROLE]. You have 249 other applicants
and eleven minutes. You are not trying to be fair. You are trying to
reduce the pile.

I'll paste my LinkedIn headline, About section, and top six resume bullets.

Give me:
1. The first thing you'd assume about me in three seconds. One sentence.
2. The three fastest reasons you'd move me to the "no" pile.
3. Anything that reads as filler, in a list, quoted exactly.
4. The single sentence that made you most likely to keep reading — and
   what to do with the rest of the profile given that.
5. One question you'd ask in a screen that my materials do not answer.

Be specific and be harsh. Do not compliment anything to soften it.

My materials:
Check Fix number 5 first. A question your materials don't answer is a gap the recruiter fills with an assumption, and it's rarely a generous one.
See it run · a real "harsh read" answer
Before · your assumption
Solid mid-career candidate. The 8 years of experience carry the story.
After · what the harsh reader said
Three-second read: "generalist marketer, probably remote." No pile because: no vertical named · no team size · no budget managed. Filler quoted: "cross-functional collaborator" · "results-driven" Kept-reading line: "Rebuilt launch process from 6 weeks to 11 days" Question you can't answer: did you own the roadmap or the calendar?
07-post-hooks.md
DoesGenerates opening lines in the book's four post shapes without turning you into a LinkedIn caricature.
You are helping me write the first line of a LinkedIn post. The first
line is the whole job — everything after it is hidden behind "see more."

I'll give you a topic from my actual work. Write four opening lines, one
in each shape:

- OBSERVATION — something true about my field that most people get wrong.
- STORY — drop the reader into a specific moment. Names a time and place.
- NUMBER — leads with a real figure from my own work.
- QUESTION — a real question I don't have a settled answer to.

Rules:
- Maximum 12 words each.
- No "Here's what I learned." No "Let that sink in." No "Unpopular opinion."
- No emoji. No line breaks between single words.
- The NUMBER version uses only a figure I gave you.

Then tell me which of the four fits my topic best and why, in one sentence.

My topic:
The number attached to it, if any:
Check If you'd be embarrassed to have a former colleague read it, don't post it. That reaction is calibration, not fear.
See it run · one topic, four opening lines
Topic
Cutting our GTM launch cycle from 6 weeks to 11 days. Number attached: 4 launches in 2025.
After · four hooks
OBSERVATION · Most launch delays are calendar problems, not craft problems. STORY · Tuesday, 8:47am, four Slack DMs asking where the deck was. NUMBER · Four launches. Six weeks became eleven days. Nothing else changed. QUESTION · Would you rather ship faster, or ship with more people saying yes? Best fit → NUMBER · the topic is a concrete result.
08-cert-triage.md
DoesSorts your credentials into the book's three categories and tells you what to cut.
Sort each credential I list into exactly one category:

- GATEKEEPER — without it I cannot legally or practically hold the role.
- CREDENTIAL-OF-THE-ROLE — hiring managers in this field expect it.
- PORTFOLIO-PROXY — it substitutes for experience I don't have yet.
- CUT — none of the above.

Rules:
- Judge against my target role, not against my history.
- Anything you can't justify in one sentence goes in CUT.
- If a credential is over five years old in a field that has moved on,
  say so.
- Be willing to put most of them in CUT. That is the expected outcome.

For each one: CATEGORY | one-sentence justification | keep on resume yes/no

My target role:
My credentials, with the year I earned each:
Check If nothing lands in CUT, the model is flattering you. Re-run it with "you were too generous; be stricter."
See it run · one sorted list
Before · what's on the resume
CSPO (2019) · Google Analytics IQ (2020) · HubSpot Inbound (2019) · Prosci Change Mgmt (2022) · Coursera "Product-Led Growth" (2023)
After · target = Senior PMM
CSPO · CUT · not asked in PMM postings · no GA IQ · CUT · 5 years old, GA4 is the version hiring now · no HubSpot Inbound · CREDENTIAL-OF-THE-ROLE · shows demand-gen fluency · yes Prosci · PORTFOLIO-PROXY · substitutes for launch-management scars · yes Coursera PLG · CUT · MOOC, not defensible under questioning · no
09-referral-ask.md
DoesWrites the 60-second referral message, in the two versions the book distinguishes between.
Write a LinkedIn message asking for a referral.

Two versions:
A) WEAK TIE — we've met once or twice, or not at all but share context.
B) STRONG TIE — we worked together directly.

Both must:
- Be under 120 words.
- Name the specific role and company in the first sentence.
- Give one concrete reason I'm a fit, using a fact I gave you.
- Make it easy to say no. An explicit out, in plain words.
- Ask for exactly one thing.
- Have no flattery opening. Never "Hope you're doing well!"

The WEAK TIE version must state where our paths crossed in one clause.
The STRONG TIE version may reference shared work directly and be shorter.

Person:
How I know them:
Role and company:
One fact that makes me a fit:
Check Read it as the recipient. If saying no would feel awkward, it's not done. An ask that's easy to decline gets answered; an ask that traps people gets ignored.
See it run · one weak-tie ask, cleaned up
Before · 178 words · closes with "would love to hop on a call"
Hey Priya! Hope you're doing well! I hope this isn't too random — I've been a longtime admirer of the work Acme is doing and I saw you're on the Growth team. I'd love to hop on a quick 15-minute call to hear about your experience. Any chance you have time next week? Would mean the world!
After · 96 words
Priya — we crossed paths on the SaaSGrowth panel last April. Acme just posted a Sr PMM role. I'd like to apply. The angle: I ran launch-cycle work at Beta that cut brief-to-live from six weeks to eleven days, which reads like the "shortening feedback loops" the JD calls out. Would you be open to a quick note over to your hiring manager, or a five-line intro? Totally fine if now isn't the right time — I know these asks land in busy weeks.
10-seniority-test.md
DoesReveals whether your materials describe you or just your job family.
Take the headline and bullets I give you. Rewrite them twice:

A) As if I were one level MORE junior.
B) As if I were one level MORE senior.

Then answer directly: how much had to change?

If the three versions differ only by the word "Senior" or "Lead", say so
plainly and tell me what's missing — the scope, budget, headcount,
autonomy, or decision rights that would actually distinguish my level.

Do not add facts. Show me the hole instead of filling it.

My materials:
Check This prompt is diagnostic. If all three versions read the same, the problem isn't wording — you haven't told anyone what you actually own.
See it run · what a diagnostic gap looks like
Before · your current headline
Marketing Manager · B2B SaaS · Ex-Beta
After · junior / current / senior read the same
JUNIOR: Marketing Coordinator · B2B SaaS · Ex-Beta CURRENT: Marketing Manager · B2B SaaS · Ex-Beta SENIOR: Senior Marketing Manager · B2B SaaS · Ex-Beta Difference: one word. Missing signal: budget owned, headcount managed, revenue attributed. Those three additions would separate the levels — none is a wording fix.
edit_document Part 2 · The READ-DO Worksheets

Eight worksheets. One sitting each.

The checklists in the book verify work you've already done. These do the opposite: they walk you through it in sequence, one step at a time, while you read. This distinction comes from checklist design in aviation and surgery, where the two formats are kept deliberately separate. Read-do is for a task you haven't internalised. Do-confirm is for one you have.

Rule for all eight: one sitting, no tabs open except the one you're editing.

01

Headline

30 min
  1. Open your profile. Paste your current headline below.
  2. Pick the archetype that fits you.
  3. Write your target job title, exactly as a recruiter would search it.
  4. Write one number from your work. Something you can defend under questioning.
  5. Assemble using your archetype's formula. Aim for one line.
  6. Check the character count above. Under 220? If not, cut the weakest clause.
  7. Publish. Don't refine it further today.
02

About

45 min
  1. Delete your current About into a scratch file. Start from blank.
  2. Write one sentence: what you do, for whom. This is the hook a recruiter sees before "see more."
  3. Proof line 1, with a number.
  4. Proof line 2, with a number.
  5. Proof line 3, with a number.
  6. Write the CTA. Who should contact you, about what, how to reach you.
  7. Paste all four into LinkedIn. Open your profile in a private window. Does the hook survive the fold?
03

Activity

40 min
  1. Name pillar one — the thing you already know well.
  2. Name pillar two — the thing you're actively learning.
  3. Pick a shape for today's post.
  4. Write the first line only. Twelve words maximum.
  5. Write four more sentences. Stop there.
  6. Publish before 10 a.m. local time.
  7. Comment substantively on three posts by people you'd want to work with.
04

ATS layout

45 min
  1. Save a copy of your resume. Name it resume-v2. Work only in the copy.
  2. Convert to a single column. Delete every table, text box, and icon.
  3. Rename headings to the standard set: Experience, Education, Skills, Certifications.
  4. Check every date format matches. Pick one style and apply it everywhere.
  5. Move contact details into the body. Out of the header and footer.
  6. Save as .docx. Then export a .pdf. Keep both.
  7. Run it through a free ATS checker. Note which title or date didn't parse.
05

Bullets

60 min
  1. List your six most important bullets below. Only six.
  2. For each: what changed because you did it?
  3. For each: what number measures that change?
  4. For each: what did you specifically do?
  5. Assemble as X-Y-Z. Rearrange for natural reading order.
  6. Read each aloud. Over eight seconds? Cut it.
  7. Mark any bullet still lacking a number. Run the Metric Excavation prompt on those.
06

Credentials

30 min
  1. List every credential on your resume, with the year earned.
  2. For each, write one sentence on why it matters for your target role.
  3. Couldn't write the sentence? Cross it out. It's cut.
  4. Sort survivors: gatekeeper / credential-of-the-role / portfolio-proxy.
  5. Early career or career-switching → certifications go above Experience. Otherwise below.
  6. Write your one-line explanation for any gap over six months.
  7. Delete the cut ones from the live document. Now, not later.
07

Referrals

60 min
  1. List 20 target companies. Twenty, not fifty.
  2. For each, search LinkedIn for one second-degree connection. Write the name next to the company.
  3. Blank rows stay blank. Don't force it.
  4. Mark each contact weak tie or strong tie.
  5. Write five messages using the matching template.
  6. Send all five today. Not tomorrow.
  7. Diarise one follow-up, seven days out. One only.
08

The sprint

14 days
  1. Put twenty-eight 30-minute blocks in your calendar now. Two a day.
  2. Days 1–5: Worksheets 1, 2, 3.
  3. Days 6–8: Worksheets 4, 5.
  4. Days 9–10: Worksheet 6.
  5. Days 11–13: Worksheet 7.
  6. Day 14: re-run the Visibility Scorecard. Compare with your first score.
  7. Book a recurring 30-minute weekly block. Keep it until you sign an offer.
menu_book Part 3 · Honest AI Briefing

What the evidence supports.

Every prompt pack you've seen promises AI will get you hired. Here is what the strongest study says, what it does not say, and the two failure modes nobody selling prompts mentions.

What the evidence supports

The strongest study on this question is a field experiment with 480,948 job-seekers, published in Management Science in 2025. Candidates given algorithmic writing assistance on their resumes were hired 8% more often, at 10% higher wages. Employers were no less satisfied with the people they hired.1

That's a real result from a large sample, and it's the reason this kit exists.

What it does not support

The same study tested grammar and clarity correction — not a language model drafting your bullets from scratch. As of now, no published experiment measures callback or hire rates for generative-AI-written resumes against human-written ones. Not one.

Clearer, error-free, well-structured writing gets hired more often, and AI is one way to get there. The claim "ChatGPT will improve your callback rate" has never been tested. Anyone stating it as fact is guessing.

This distinction matters because it tells you what to use these prompts for. Use them to sharpen writing you could have written yourself with more time and a better editor. Don't use them to manufacture a professional identity you can't defend in a room.

Risk one · the self-preferencing trap

A 2025 study found that when language models are used to evaluate resumes, they favour resumes generated by the same model — self-preferencing in 67% to 82% of comparisons. Candidates whose resume came from the same model doing the screening were 23% to 60% more likely to be shortlisted, with content held equivalent.2

This is a preprint and hasn't completed peer review, so hold it loosely. But if it holds, the implication runs against everything prompt packs tell you. Optimising heavily for one model's style is a bet on which model your next employer happens to screen with. You can't know that. Write for a human reader, keep your own voice in the output, and you're not exposed to the bet either way.

Risk two · detectors, and who they punish

AI-text detectors are unreliable in a way that isn't evenly distributed. A peer-reviewed analysis found accuracy ranging from 55% to 97% depending on text type, length, and language — with the errors falling disproportionately on non-native English speakers.3

Read that again if English isn't your first language. Writing that is careful, correct, and slightly formal — exactly the register a second-language professional writes in — is the register these tools most often misclassify as machine-generated.

There is no clean defence, and pretending otherwise would be dishonest. What reduces exposure: keep specific details, real names, real numbers, and the odd irregular sentence that no model would produce. Generic polish is what gets flagged. Specificity is both better writing and better cover.

What AI cannot do for you

  • Know what you did. It has no access to your work. Every number comes from you.
  • Judge what's defensible. It'll happily produce a metric you can't survive being questioned on.
  • Understand your market. It doesn't know what your industry pays or values this year.
  • Make the ask. No model sends the referral message. That's still you.

Five rules

  1. Never ship a number you didn't supply. If it appears in output and not in your input, it's invented. Delete it.
  2. Read every line aloud before it goes live. If it doesn't sound like you speaking, it won't sound like you in the interview.
  3. Keep the irregularities. The slightly odd phrasing that's genuinely yours is worth more than smooth text.
  4. Use AI on the draft, not the decision. What to claim, where to apply, who to ask — yours.
  5. Assume a human reads it last. Because one does, and that's the only reader who makes an offer.

Bookmark this section. Come back to it every time you're tempted to paste raw output into your profile.

§ Notes
  1. Wiles, E., Munyikwa, Z., & Horton, J. J., "Algorithmic Writing Assistance on Jobseekers' Resumes Increases Hires," Management Science, 2025. Field experiment, n = 480,948, on an online labour market. pubsonline.informs.org — also NBER Working Paper 30886, nber.org/papers/w30886
  2. Xu, J., Li, G., & Jiang, J. Y., "AI Self-preferencing in Algorithmic Hiring: Empirical Evidence and Insights," arXiv preprint 2509.00462, September 2025. Controlled experiment across 24 occupations. Preprint — not yet peer-reviewed. arxiv.org/abs/2509.00462
  3. "The accuracy-bias trade-offs in AI text detection tools and their impact on fairness," peer-reviewed, 2025. ncbi.nlm.nih.gov

Run one prompt today. Read Part 3 before you paste any output into your profile.

§ About this tool

A compilation, not an opinion. The Wiles 2025 study, the Xu 2025 preprint, and the PMC 2025 detector analysis are the load-bearing sources — cited in the footnotes on this page. Maintained by the editorial team — corrections or questions go to hi@thevisibilitystack.com. How this book is compiled →

This kit is part of LinkedIn Hiring Book. Read chapter 1 free →

v2026-08-19 · sources current as of Aug 2026