WEEK 106 :: 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 Offer — Negotiating From Evidence, Not Anxiety."
This is Week 7 of an eight-week series on running a job search with AI — the negotiation week, and the emotional and financial peak of the series, exactly as it was in the car series. An offer is on the table. The reader's heart says grab it before they change their minds; six weeks of artifacts say otherwise. Employers broadly expect candidates to negotiate — yet first-timers routinely accept the opening number, and the cost of that compounds over a career the way a bad interest rate compounds over a loan: every future raise, bonus percentage, and next-job offer anchors on the number the reader accepts this week.
The reader is not walking in empty-handed. Week 1 set the compensation baseline. Week 2 wrote the deal-breakers. Week 4 built the company dossier. Week 6's interviews revealed how much they want this reader. Negotiation is where the whole series' evidence gets spent.
The deliverable the reader should walk away holding: a negotiation plan with a researched target range, a walk-away number, and scripted responses for the conversations and emails ahead.
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 total offer. Base, bonus, equity, benefits, and the fine print translated into real comparable dollars — vesting schedules that gate the money, bonus targets versus bonus reality, what the benefits actually replace — from the offer documents the reader supplies, ending in a one-page picture of what is actually being offered.
- Benchmark it against evidence the reader gathers. A research structure for building the market case: which pay-transparency postings, published salary data sources, and leveling comparisons to pull, how to weigh them for the reader's role, level, and metro, and how to assemble them into a defensible target range — the AI organizing evidence the reader collects, not asserting what the role pays.
- Script the negotiation itself. Word-for-word: the counter delivered with warmth and a number, the email version, responses to "that's the best we can do" and "what number would get you to yes," the scenario tree for competing offers — and the walk-away line the reader decides on in advance, when they are calm.
At the advanced tier, the strongest version of this week is the complete negotiation playbook — decoded offer, evidence-backed range, scenario tree with scripted branches, and competing-offer orchestration handled honestly (real timelines, no bluffing) — built so the reader can run the conversation from a page instead of from adrenaline.
The constraint that defines this week. AI models cannot see current market compensation, and a negotiation anchored on an invented number is worse than no research at all. No prompt may ask the AI to state what a role, level, or metro currently pays as fact, to value a specific company's equity, or to promise that a tactic will produce a result. The market case is built from sources the reader pulls — pay-transparency postings, published salary data, people in the role. Equity in a private company gets a framework for the questions to ask, never an AI-asserted valuation. And where an offer raises legal questions — non-competes, clawbacks, contract terms — the prompts point to a lawyer, not to the AI's reading. Scripts are coaching, not guarantees, and none of this is financial advice: the plan organizes the reader's decision; the decision stays theirs.
Design the prompts so the AI does what it is genuinely good at: decoding dense documents, structuring an evidence file, and drafting language the reader can rehearse and make their own. The reader supplies the offer, the market evidence, and the walk-away point; the AI supplies the decode, the structure, and the words to start from. "Word-for-word" means a complete starting draft, not a finished performance — and this is the week that rule matters most: the final pass belongs to the reader. A counter delivered in someone else's voice collapses at the first follow-up question, and the negotiation email lands on the desk of the exact hiring humans this series' contract is about. Every scripting prompt should end by handing the language back for the reader's own rewrite — rehearsed until it is genuinely theirs — and the post should say why.
Series dependency chain, for the Metadata block: Week 7 consumes Week 1's compensation baseline (the floor), Week 2's deal-breakers (what cannot be traded away), Week 4's company dossier (the other side's position), and Week 6's interview signals (how much they want this reader). Week 7 produces the negotiation plan and its outcome — consumed by Week 8, where the final numbers meet the final decision.
Because readers may arrive at this post with an offer in hand and none of the earlier artifacts, the prompts should work from the offer letter and the reader's own sense of their floor, while making clear the negotiation is strongest when it spends six weeks of accumulated evidence.
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 first-time negotiator with a single offer and sweaty palms, a reader weighing startup equity against a bigger base elsewhere, and one holding two offers on different timelines 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 — above all, no salary numbers. If a prompt or example needs a wage figure to work, restructure it so the reader's own researched number is the input.)
## 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: 7` 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 current compensation for a role or metro as fact, value a specific company's equity, promise a negotiation outcome, or interpret contract terms that belong with a lawyer. The market case is built from evidence the reader gathers; the walk-away number is the reader's own. Also confirm every negotiation script and email prompt ends with the reader rewriting and owning the final words (the series contract's final-pass rule) — a counter in someone else's voice collapses at the first follow-up.
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.
The offer email arrives and your body decides before your brain does — grab it, say yes, don't be difficult — while six weeks of accumulated evidence sit in a folder saying otherwise. The number you accept this week anchors every raise, bonus percentage, and next-job offer that follows it, which is why this post gives you three prompts at three depths: a decoder that turns the offer letter into plain English, a research structure that turns scattered market data into a range you can defend, and a full negotiation playbook with a scenario tree, a walk-away line, and a sparring partner who pushes back before the recruiter does. None of them will speak for you. All of them make sure that when you speak, you know exactly what you are holding.
The Offer Decoder
See exactly what the offer contains before you feel anything about it.
An offer letter is a persuasive document wearing an administrative costume. It leads with the biggest number, states the pleasant things concretely and the unpleasant things by reference, and puts the parts that determine whether the money is real — vesting cliffs, bonus discretion, accrual schedules, plan documents you have not been sent — in the places least likely to be read on a Friday afternoon. Most first-time negotiators respond to the base salary and nothing else, because base salary is the only part written in a form a human can hold in their head. Before you can negotiate an offer, you have to be able to see it. That is a reading problem, and reading dense documents carefully is the single thing an AI does better than a nervous person at nine on a weeknight.
Why this matters now
You are probably reading this within seventy-two hours of receiving an offer, which is precisely when your judgment is worst and the pressure to answer is highest. Recruiters often ask for a decision inside a few days, and the fastest way to lose the negotiation is to spend those days feeling instead of reading. This prompt costs you ten minutes and converts the letter from a source of anxiety into a list of known facts, unknown facts, and specific questions — and the act of sending good questions buys you legitimate time without a single word of pushback. It also does something quieter: it shows you which parts of the offer are conditional, and conditional components are where negotiation is often easiest.
You are acting as my private compensation analyst. I am going to paste the text of a job offer I received. Your job is to translate it into plain language so I understand exactly what is being offered. You are not evaluating whether it is good, and you are not filling in anything the document does not say.
Read the offer and produce four sections.
1. WHAT I WOULD ACTUALLY RECEIVE. List every component separately — base pay, bonus, equity, retirement contribution, health coverage, paid time off, stipends, and anything else named. For each one, state in one sentence what it means in practice and when I would actually have that money or that benefit in hand.
2. WHAT IS GUARANTEED VERSUS WHAT IS CONDITIONAL. Sort every component into guaranteed, conditional on something, or at the company's discretion. Where it is conditional, name the condition. Where the condition is defined in a document I have not been given, say that.
3. WHAT THE DOCUMENT DOES NOT SAY. List the specific facts a reasonable person would need in order to evaluate this offer that are missing, vague, or defined only by reference to a plan document, handbook, or agreement I have not seen.
4. QUESTIONS TO ASK. Write 5 to 8 plain, non-confrontational questions I can send to the recruiter to close those gaps. Number them. Keep them short enough to send in one email without sounding like a deposition.
Rules that override everything above. Do not estimate or assert what this role pays in my market, and do not put a dollar value on the equity — you cannot see current market data or this company's financials, and an invented number would be worse than no number. Where a term carries legal consequences, including non-compete, non-solicitation, clawback, arbitration, or assignment of prior work, flag it in section 3 and tell me to have an employment lawyer read it rather than interpreting it yourself. If the offer is silent on something, say it is silent rather than filling it in.
Here is the offer: [PASTE THE FULL TEXT OF YOUR OFFER LETTER AND ANY BENEFITS SUMMARY YOU RECEIVED]
How the AI reads this prompt
Practical examples from different industries
A first-time negotiator with sweaty palms. A program coordinator at a mid-sized nonprofit receives their first salaried offer after four years of hourly work, and the number is more than they have ever earned, so their instinct is to accept within the hour. They paste the letter and the benefits summary. The decode puts the annual bonus in the discretionary bucket, notes that paid time off accrues monthly rather than arriving as a bank on day one, and flags that the retirement match is described as subject to plan terms in a document not attached. The questions section produces six short lines, one of which asks how the bonus has actually paid out for this role over the last two years. They send the email, buy four days, and discover the number they were about to accept as guaranteed was roughly a tenth smaller in practice.
A reader weighing startup equity against a larger base elsewhere. A senior engineer holds an offer from a private company where the option grant is described as a number of shares and nothing else. The decode does not tell them what the equity is worth — it cannot, and the prompt forbids it — but section 3 lists everything absent from the letter that would determine whether the grant means anything: total shares outstanding, strike price, the price of the last preferred round, the vesting schedule and cliff, the post-termination exercise window, and whether the shares are ISOs or NSOs. Those six absences become six questions. The recruiter's answers turn an emotional argument about upside into a comparison the reader can actually run against the competing base salary.
A shift-based offer with variable scheduling. A veterinary technician returning to work after a caregiving gap receives an offer built from an hourly rate, a shift differential, an on-call rotation, and a sign-on bonus with a repayment clause. The decode separates the guaranteed hourly floor from the differential income that depends entirely on which shifts they are assigned, and flags the repayment clause as a term with legal consequences that belongs with a lawyer rather than with the AI. The questions section asks how many differential-eligible shifts the last person in the role averaged. That single question changes the offer from an attractive headline rate into a schedule negotiation, which is the negotiation that was actually available.
Creative use case ideas
Decode your current job before your next review. Run the same prompt on your existing comp letter and last year's bonus statement. Most people discover at least one component they have been misremembering, and the guaranteed-versus-discretionary sort is a better preparation for a raise conversation than any amount of rehearsal.
Read the open-enrollment packet for once. Benefits documents are written in the same evasive register as offer letters and are re-issued every autumn to people who never open them. The prompt works unmodified on a plan summary, and the what-this-does-not-say section is where the surprises about deductibles and coverage tiers surface.
Help someone else read their first offer. This works well for a teenager's first real job, a partner's promotion letter, or a friend re-entering the workforce, and it gives you something concrete to offer instead of vague encouragement. Strip the names before pasting.
Turn it on a lease, a contractor bid, or a service agreement. Any document that promises something conditionally, at a price, with terms defined elsewhere, has the same structure as an offer letter. A home renovation quote run through the guaranteed-versus-conditional sort produces an unusually clarifying afternoon.
Decode a volunteer or board commitment. Nonprofit board invitations and volunteer leadership roles carry real obligations described in warm language — hours, fundraising expectations, liability, term length. The same four sections apply, and the questions section is far easier to send than to compose from scratch.
Adaptability tips
For an internal promotion offer, add a line asking the model to compare the new components against your current package and list only what changed, since internal letters often restate existing benefits at length and bury the two lines that are actually new. For a contract-to-hire arrangement, add the conversion terms to the input and ask for a fifth section covering what happens at the conversion point and what is guaranteed if it does not happen. For an academic or research position, ask it to treat startup funds, teaching load, and summer salary as compensation components rather than working conditions, because they are. For an offer in a country you have not worked in before, ask the model to name which components are typically statutory there versus negotiable, then verify that answer with a local source rather than trusting it — this is one place where a general-purpose model's knowledge is genuinely thin. And if the offer arrives as a PDF you cannot cleanly copy, upload the file rather than retyping it, then ask the model to quote the exact sentence behind any component it lists, so you can spot extraction errors.
Pro tips
Run it twice. Once with only the offer letter, once with the letter plus the benefits summary and the employee handbook. The difference between the two what-this-does-not-say sections tells you precisely how much of your decision depends on documents you were not sent.
Re-run section 3 after the recruiter answers. Paste their reply back in and ask which gaps closed, which were answered partially, and which were answered with language that sounds like a yes but commits to nothing. Recruiters are not being dishonest when this happens — they are answering in the language their legal team gave them — but the distinction matters to you.
Ask for the sentence. When a component's description surprises you, ask the model to quote the exact clause it drew from. This catches both misreadings and your own hopeful skimming, and it takes five seconds.
Strip identifying details first. Replace the company name, your name, and any employee identifiers before pasting. You lose nothing analytically, and the document you are handling is one of the more sensitive ones you will ever paste into a chat window.
Prerequisites
You need the complete text of the offer letter, and ideally the benefits summary, equity grant documentation, and any plan documents you were given. You do not need any of the earlier weeks in this series to run this prompt, though if you have Week 1's compensation baseline in front of you, the decode will land harder. You should know your own answer to one question before you start: what you currently earn in total, not just in base pay, so you have something honest to compare against. Finally, decide before you paste which identifying details you are willing to put into a chat window, and remove the rest.
Required tools
Any general-purpose AI chat tool will run this prompt, including free tiers of the major assistants. The only real requirement is enough context capacity to accept the full offer letter plus a benefits summary in one paste, which every current mainstream assistant handles comfortably. If your documents are PDFs, a tool with file upload saves time and reduces transcription errors, though pasting extracted text works fine. No plugins, no browsing, and no paid tier are required for this variation.
Frequently asked questions
Is it safe to paste my offer letter into an AI tool? Treat it the way you would treat any sensitive document going into a third-party service. Remove your name, the company name, employee or requisition identifiers, and anything else that would make the document attributable, since none of it changes the analysis. Check whether your chosen tool trains on conversations by default and turn that off if it does — most offer a setting for it. If your employer or prospective employer has a policy on AI tools, follow it; this is backstage preparation, but the document is theirs as much as yours.
What if the AI gives me a salary figure anyway? Ignore it and re-run with the rule restated. Models drift on constraints in long conversations, especially after a large pasted document pushes the original instruction far back in context, and a confident-sounding market number is exactly the kind of output the model is good at producing and bad at grounding. The number is generated from patterns in text, not from current market data, and it cannot see your metro or this quarter. Every real number in your negotiation should come from Variation 2, where you gather it yourself.
I do not have any of the earlier weeks in this series. Does this still work? Yes. This variation was built to work from the offer letter alone, because plenty of readers arrive here with an offer in hand and no accumulated file. What you lose is the comparison layer — a baseline from Week 1 tells you whether the guaranteed portion clears your floor, and Week 2's deal-breakers tell you which conditional components you cannot accept regardless of the number. You can reconstruct rough versions of both in twenty minutes, and it is worth doing before you reply.
Should I send all the questions at once or space them out? Send them in one short email. Spacing them out reads as stalling, extends the timeline, and gives the impression of a candidate who keeps finding new problems, while a single well-organized list reads as someone doing exactly what a careful professional does before signing. Say plainly that you are excited about the role and want to make sure you understand the package fully, then list the questions. That framing is true, costs nothing, and buys you the days you need.
Recommended follow-up prompts
Run Variation 2 in this post next. The decode tells you what is on the table; the evidence file tells you what to ask for, and doing them in that order keeps you from researching components that turn out to be fixed anyway.
Ask for a benefits-replacement check. Prompt the AI to list, for each benefit named in the offer, what expense it replaces and what you would pay for the equivalent yourself if the benefit did not exist — then price those yourself. This turns the benefits section from a list of perks into a comparison you can actually run against a competing offer.
Revisit Week 4's company dossier prompt with the offer in hand. The dossier was built to understand the company as an employer; reading it again after an offer arrives surfaces which of the conditional components are likely to be real at this specific company, based on what you already learned.
Tags and categories
Tags:
job offer, offer letter, compensation, benefits, equity, total compensation, document analysis, negotiation preparation, career, job search
Categories:
Career and Job Search, Document Analysis
Citations
IRS Publication 525, Taxable and Nontaxable Income — for understanding which offer components are taxed and when.
IRS Tax Topic No. 427, Stock Options — for the distinction between incentive stock options and non-statutory options referenced in the equity example.
U.S. Bureau of Labor Statistics, Employer Costs for Employee Compensation — the official series measuring wages and benefits as components of total compensation, useful as background for why benefits belong in the decode at all.
The Evidence File
Turn scattered salary data into a range you can defend.
Here is the trap most people walk into. They ask an AI what the role pays, get a confident range, walk into the conversation with it, and get asked one question — where did that come from, what level, what location, what year — that they cannot answer. The number collapses, and their credibility goes with it, because a candidate who cited something they cannot source has now told the recruiter exactly how much preparation is behind everything else they will say. The fix is not to research harder. It is to change what the AI is for. It cannot see the market; it can see structure, and structure is the thing amateur research actually lacks. Let it design the collection and weigh the results, and go get the evidence yourself.
Why this matters now
Pay transparency has changed what is possible here in the last few years. A meaningful share of job postings now carry published ranges by law, which means the evidence for your negotiation exists in public and did not a decade ago — but it exists as scattered postings across jurisdictions, titles, and leveling schemes that do not line up with each other or with you. Assembling thirty half-matching data points into a defensible position is genuinely hard, and it is hard in a way that rewards structure over effort. This prompt gives you a week's worth of research discipline in one sitting, and more importantly it tells you which of your data points are weak enough that raising them would damage your case. That second half is what separates a range you can hold from a range you can state.
Act as a research analyst helping me build the evidence file behind a salary negotiation. You cannot see current market pay data and I do not want you to estimate it. I will gather the data points myself and bring them back to you. Your job is to design the collection, then organize and stress-test what I collect.
My situation.
Role and level: [TITLE AND LEVEL AS THE OFFER STATES IT]
Metro or work location: [CITY, or REMOTE plus the company's headquarters location]
Industry, company size, and stage: [FOR EXAMPLE, HEALTHCARE SOFTWARE, ABOUT 200 PEOPLE, PRIVATELY HELD]
Years doing this kind of work: [NUMBER]
Three things I do that a typical person in this role does not: [LIST THEM]
My floor, decided before this offer arrived: [THE TOTAL BELOW WHICH I WOULD DECLINE]
Work in two phases and stop after phase one.
PHASE ONE — THE COLLECTION PLAN. Give me four things.
(a) A ranked list of 6 to 8 source types to pull from, strongest evidence first, with one sentence on what makes each strong or weak for my specific role, level, and location.
(b) For each source type, the exact search string I should run or the exact question I should ask a person, written out so I can use it directly.
(c) A blank comparison sheet I can fill in, described as a list of column headings, including at minimum source, date collected, exact posted title, level, location, company size, posted range, and how closely the match fits me.
(d) Three ways a data point can look relevant and not be, so I do not collect evidence that falls apart under one follow-up question.
Then stop and wait. Do not begin phase two until I paste my filled-in sheet.
PHASE TWO — after I return my collected data, weigh it. Group the points by how well they match my role, level, and location. Tell me which two or three I should lead with and which are weak enough that raising them would hurt me, and say why in each case. Then propose a target range with a specific opening number, stating in one line what evidence each end of that range rests on. Finally, name every gap a well-prepared recruiter could push on, and tell me what evidence would close it.
Rules. Do not tell me what this role pays — only what my collected evidence supports and how strongly. If my data is too thin to support a range, say that plainly and tell me what to go collect rather than producing a range anyway. Do not attach a value to any company's equity. Where the offer raises legal questions, point me to an employment lawyer rather than interpreting the terms.
How the AI reads this prompt
Practical examples from different industries
A remote worker whose company pays by its own headquarters. A data analyst living in a smaller metro takes a fully remote role at a company headquartered in a considerably more expensive city, and the offer is benchmarked to neither location clearly. In phase one they specify remote plus the company's headquarters city, and the collection plan tells them to gather two parallel sets of postings — remote-designated roles at similarly headquartered companies, and in-office roles in their own metro — and to record which category each row belongs to. In phase two the weighting surfaces that their strongest points come from remote-designated postings at comparable companies, and that their local in-office data actively undercuts them. They lead with the first set and never mention the second, which is a decision they would not have reached by intuition.
A career changer whose title does not exist in the market. A former high school science teacher moving into instructional design finds that half the postings call the role learning experience designer, a quarter call it curriculum developer, and the levels are numbered differently at every company. The collection plan handles this by asking for searches across all three title families plus a leveling comparison based on scope described in the posting rather than on the number attached to it. Phase two groups their thirty rows by actual scope and shows that their strongest evidence sits in a title family they had not been searching, which changes both their range and the language they use to describe themselves in the conversation.
A reader in a state with no transparency requirement. A logistics coordinator in a state without a pay-disclosure statute assumes the research is impossible and nearly skips it. The collection plan points them to postings from employers who publish ranges regardless of where the job sits, to national data broken out by metro, and to the two people they already know in the role at other companies, with the exact question to ask each — a question phrased about ranges for the role rather than about the person's own pay, which is what makes it answerable. Phase two flags the geographic adjustment as their largest open gap and tells them precisely what would close it, which turns an anxiety into a task.
Creative use case ideas
Set your freelance rates from evidence rather than nerve. Replace role and level with your service and client type, and the collection plan produces a survey structure for published rate cards, public project budgets, and marketplace listings — plus the same brutal sort into points you can lead with and points that would undercut you.
Prepare an internal raise conversation. The structure works unchanged when the other party already employs you, and phase two's gap analysis is especially useful because internal conversations get pushed back on with internal comparisons you have to anticipate rather than external ones you can research.
Evaluate graduate program stipend offers. A doctoral applicant weighing funded offers across three programs and three cities has exactly this problem in a different costume — scattered, half-comparable numbers with different cost bases and different guarantees about duration. The collection plan adapts by swapping in stipend, funding duration, and required teaching load as the columns.
Help a community organization set a stipend or a salary band. Small nonprofits, co-ops, and volunteer boards routinely set compensation by whatever the last person was paid, which reproduces every past inequity. The same evidence structure produces something a board can actually defend to its members, and the weakness sort keeps the conversation off anecdote.
Build the case for a budget you keep having to justify. Any recurring request that gets denied for lack of comparables — a program budget, a tool subscription, headcount — has the same shape. Collect the comparables, weigh them, and lead with the two strongest rather than the whole pile.
Adaptability tips
Change the unit of comparison when the money is not the point. If the negotiation you care about is really about title, scope, remote flexibility, or start date, swap the posted-range column for whatever you are actually trading in, and phase two will weigh it the same way. Add a time dimension for volatile fields by asking the model to weight recent data points more heavily and to flag anything older than a stated cutoff as background rather than evidence. If you are negotiating across a border, add a column for statutory versus negotiated components, because in many countries a meaningful share of what an American reader would negotiate is set by law. For senior roles where posted ranges are rare, replace the posting column with sources that do exist at that level — proxy filings for public companies, published nonprofit compensation disclosures, and the people you know — and ask phase two to say plainly how much weaker that evidence base is. And if you run this more than once, keep the sheet. The same file, updated, is the starting point for every future negotiation you will have.
Pro tips
Date-stamp everything and mean it. Two of the three most common ways an evidence point gets destroyed in a live conversation are staleness and level mismatch, and the date column is the cheap defense against the first one. Anything you cannot date, treat as background.
Ask for the recruiter's counter-case. After phase two, ask the model to write the strongest good-faith argument against your range using only your own data. What comes back is roughly what you will hear, which means you have already answered it once before you hear it.
Run phase two twice with different framings. Once asking for a defensible range, once asking for the most conservative range your evidence strictly supports. The distance between them is your actual uncertainty, and knowing that distance is what lets you hold a number calmly.
Keep the sheet outside the chat. Chat sessions end; your evidence file should not. Put the collected rows in a spreadsheet or a document you own, and paste it back in when you need it.
Prerequisites
You need a few hours across a few days for the actual collection — this is the variation with homework, and the homework is the point. You need your floor decided before you start, ideally from Week 1's compensation baseline, and you should write it down somewhere before the offer's number is in your head. It helps enormously to have two or three people you can ask about ranges in your role, so line them up before you begin rather than during. And you need the decoded offer from Variation 1, or at least a clear sense of which components are fixed, so you do not spend research effort on parts of the package that were never going to move.
Required tools
Any general-purpose AI assistant handles this, and no browsing capability is required — in fact this variation is designed on the assumption that the AI cannot browse, because you are the one collecting. A spreadsheet or notes app to hold the comparison sheet between sessions is genuinely useful, since the collection happens across several days. If your assistant supports saved projects or persistent conversation memory, use it, so phase two can be run days after phase one without re-establishing your context. Free tiers are sufficient throughout.
Frequently asked questions
Why not just let the AI tell me the range? Because it does not know, and the way it fails is the dangerous kind. A model generates a plausible number from patterns in text it was trained on, which means the answer is shaped by whatever was written about your role in years past, in aggregate, without regard to your metro, your level, or this quarter. It will sound authoritative, which is exactly what makes it costly — you will state it with confidence and be unable to source it when asked. Every number you bring should be one you can point to.
How many data points do I actually need? Fewer high-quality points beat many weak ones, and phase two is built to tell you which is which. As a working target, aim for a handful of close matches on role, level, and location rather than dozens of loose ones, and treat anything that requires an explanation of why it is comparable as a point you will not raise. If phase two tells you the file is too thin to support a range, believe it and collect more — that outcome is the prompt working, not failing.
What if I find data that says the offer is already generous? That is a real and useful result, and it should change what you ask for rather than whether you ask. If the base is genuinely strong for your level and market, the conversation moves to the components that are not — start date, title, remote flexibility, sign-on, review timing, professional development budget. Knowing that the base is strong also removes the anxiety that makes people accept everything else without discussion.
Is it awkward to ask people I know about pay ranges? Less than you expect, if you ask about the role rather than about them. Asking what range a company posts for a given level, or what someone would expect that role to pay in that market, is a question most people answer readily, because it does not require disclosing their own number. The collection plan writes these out for you specifically so you are not composing them in the moment.
Recommended follow-up prompts
Move to Variation 3 in this post once you have a range. The playbook takes the decoded offer and the evidence summary as direct inputs, and it is noticeably better with the real thing than with an estimate.
Build a components-versus-cash prompt. Ask the AI to structure a comparison between the non-salary items you could request and their rough value to you personally, so that when the base is genuinely fixed you already know your priority order among everything else.
Revisit Week 2's deal-breakers with the evidence in hand. Deal-breakers written before an offer existed read differently once you can see what you would be trading them for, and the reconciliation is worth doing on paper before you are asked to do it out loud.
Tags and categories
Tags:
salary research, market data, pay transparency, benchmarking, evidence, target range, negotiation, leveling, compensation, job search
Categories:
Career and Job Search, Research and Analysis
Citations
U.S. Bureau of Labor Statistics, Occupational Employment and Wage Statistics program — official occupational wage data by metropolitan area, useful as one input among several rather than as a benchmark on its own.
State and municipal pay-transparency statutes, including Colorado's Equal Pay for Equal Work Act, New York City Local Law 32, California SB 1162, and Washington's Equal Pay and Opportunities Act — the legal basis for the published ranges this variation's collection plan depends on.
The Complete Negotiation Playbook
Run the conversation from a page instead of from adrenaline.
The failure mode of a well-prepared negotiator is not being under-researched. It is having exactly one script — the opening counter — and nothing at all for the fourteen seconds after the recruiter says something unexpected. That silence is where people give the number back. They fill it with a nervous concession, or an apology, or the phrase but I'm flexible, and six weeks of evidence evaporates in a sentence nobody planned. The counter is the easy part. What you need is a map of where the conversation can go, a drafted response waiting at each destination, and a line you decided on while calm for the branch where the answer is no. Then you need to have said all of it out loud, in your own words, to something that pushed back.
Why this matters now
This is the week the series' central rule earns its keep. Everything the AI produces here is language that a hiring human will read or hear, which means none of it can go out as written — a counter delivered in a voice that is not yours collapses at the first follow-up question, and recruiters have become extremely good at recognizing text that was not written by the person sending it. So the playbook is built to be rewritten. The AI maps the branches, drafts the starting language, plays the opposing side, and then scores you on how you actually sounded, which is the part no amount of solo preparation provides. Used this way — analyst backstage, you on the page — the preparation is invisible in the room and entirely yours, which is precisely the point.
You are my negotiation coach and sparring partner for a live job offer. Nothing you write will be sent as written. Everything you draft is a starting point I will rewrite in my own words before it reaches a human, and I want you to hold me to that at the end.
My inputs.
DECODED OFFER: [PASTE the plain-language breakdown of the offer, or the offer text itself]
EVIDENCE SUMMARY: [PASTE your researched target range and the two or three strongest data points behind it]
MY FLOOR AND MY DEAL-BREAKERS: [The total below which you decline, and the terms you will not trade away]
WHAT THEY REVEALED: [Signals from the interviews about how much they want you — urgency, scope, who you would be replacing, what they said about the timeline]
OTHER PROCESSES IN FLIGHT: [Other companies, with their real stage and real dates only. Leave this blank if there are none.]
MY CONSTRAINTS: [Decision deadline, start date, visa, relocation, anything genuinely non-negotiable]
Build a negotiation playbook in five parts.
PART 1 — THE ASK. State my counter as one sentence containing the number, the two or three non-salary terms I am also asking for, and the priority order among them. For each item, name which of my inputs supports it. Where an input is missing or thin, say so rather than inventing support.
PART 2 — THE SCENARIO TREE. Map the conversation as branches. Cover at minimum these six: they accept, they counter below my target, they say the number is fixed, they ask what number would get me to yes, they ask whether I have other offers, and they go quiet for several days. For each branch give me four things — what that response most likely signals, my objective in that branch, a drafted reply in spoken form, and the shorter version I would send in writing. Where I do have another process in flight, use its real stage and real timeline. If a branch would only work by implying an offer I do not have, say so and give me the honest version instead.
PART 3 — THE WALK-AWAY. Draft the line I would say if the final number lands below my floor, in a form that leaves the relationship intact and the door open. Then list the three conditions that would legitimately justify revising my floor, so I can check my own reasoning in the moment instead of afterward.
PART 4 — THE REHEARSAL. Interview me as the hiring manager. One message at a time, in character, pushing back the way an experienced negotiator would rather than an easy one. Do not coach me while in character. Stay in character until I write STOP.
PART 5 — THE CRITIQUE. When I write STOP, drop character and score me against these criteria, quoting something I actually said for each and giving one concrete fix. Did I name a number without apologizing for it. Did I stop talking after making the ask. Did I trade against my own stated priority order. Did I claim leverage I do not have. Did I say anything I would not want repeated to my future manager on my first day.
Rules that override everything above. Do not state what my role, level, or metro currently pays — the range comes only from my evidence summary. Do not assign a value to this company's equity; give me the questions that would determine its value instead. Do not interpret non-competes, clawbacks, or any contract language, and flag those for an employment lawyer. Do not promise that any tactic will produce a result; scripts are preparation, not guarantees, and none of this is financial advice. When the playbook is complete, tell me to rewrite every scripted line in my own words and read them aloud until they sound like me.
Start with Part 1 and wait for me before continuing to Part 4.
How the AI reads this prompt
Practical examples from different industries
Two offers on genuinely different timelines. A product manager holds one offer with a Friday deadline and a second company that has just scheduled a final round for the following week. They fill in the other-processes field with the real stage and the real date, and the scenario tree produces something they would not have written themselves — a branch in which they tell the first company plainly that another process concludes on a specific date and ask whether the deadline can move to that date, with the drafted spoken version and a three-line email version. Because the timeline is real, the ask is verifiable, and the branch where the answer is no comes with its own script. What they avoid is the version everyone reaches for under pressure, which is implying a competing offer that does not yet exist.
Startup equity against a larger base elsewhere. An infrastructure engineer weighs a private company's grant against a considerably higher base at a public one. The playbook refuses to value the equity, which is the correct behavior and initially frustrating, and instead builds Part 1 around asking for the specific facts that would let the reader value it themselves — outstanding shares, strike price, the last preferred round, the exercise window after departure. The counter that emerges asks for a base increase and those disclosures together, on the stated reasoning that they cannot compare the two offers without them. That is a request a recruiter can act on, and it is far stronger than an argument about how much the equity might be worth.
A single offer, no competing process, no leverage but the work. A grants manager at a small foundation has one offer, no other interviews, and a floor that the offer clears by a narrow margin. Their honest inputs produce a tree with no bluffing branch in it at all — instead, Part 1 builds the ask around the scope signals from the interviews, where the hiring manager described a portfolio noticeably larger than the posting implied. The rehearsal in Part 4 is where the value lands. Three rounds of a hiring manager saying the range is set teaches them to say their number once and stop talking, which is the skill that actually moves the outcome when there is no external leverage to lean on.
Creative use case ideas
Rehearse any conversation you are dreading. The Part 4 and Part 5 structure — in-character pushback, exit token, evidence-based critique — transfers to a landlord about a deposit, a contractor about a change order, a family member about money, or a doctor's appointment where you keep leaving without asking the thing you came to ask.
Negotiate scope instead of price. Freelancers and consultants can swap the offer inputs for a project brief and run the same tree against the branches that actually cost money — the extra revision round, the expanded deliverable, the deadline that moved.
Prepare an advocacy conversation you are having on someone else's behalf. Managers arguing for a report's promotion, parents in a school accommodation meeting, or anyone accompanying a relative to a benefits appointment can build the same tree, and the walk-away section becomes the point at which you escalate rather than the point at which you leave.
Run it as a post-mortem instead of a preparation. After a negotiation that went badly, paste in what was said and ask for Part 5 only. The scored critique against the same five criteria is uncomfortable and unusually instructive, and it is the fastest way to be better at the next one.
Build the playbook for a decision that is not a conversation at all. Whether to take a role, leave one, or accept a project has the same structure — conditions decided while calm, branches mapped in advance, a threshold with defined revision conditions. Skip Parts 2 and 4 and the remainder still works.
Adaptability tips
Split it if your assistant loses the thread. On a long session, run Parts 1 through 3 in one conversation and save the playbook as a document, then start a fresh conversation for the rehearsal with the playbook pasted in as context. The rehearsal is more realistic when the model has not just spent an hour agreeing with you. For an internal negotiation, replace the interview-signals input with what you know about the budget cycle and your manager's own constraints, and add a branch for the response that only occurs internally, which is a promise about the next cycle. If your negotiation involves an agency recruiter rather than the employer directly, add a note about it — the incentives differ, and the branch where the recruiter is negotiating against their own commission needs its own handling. Where the offer involves relocation, visa sponsorship, or an unusual start date, put those in constraints rather than in the ask, because they change which trades are available to you rather than being trades themselves. And run Part 4 more than once with different opponent personalities specified — a warm relationship-builder and a blunt time-pressed manager will expose different weaknesses in the same script.
Pro tips
Rehearse out loud, not in the chat window. Typing your responses to Part 4 will make you look far more composed than you will sound, because typing lets you edit. Read your replies aloud as you type them, or use voice input if your tool supports it, and the critique in Part 5 becomes considerably more useful.
Do the critique twice. Once after the rehearsal and once after the real conversation, pasting in what was actually said. The gap between how you performed against a simulated hiring manager and how you performed against a real one is the most specific coaching you will get anywhere.
Print the tree. The entire premise is running the conversation from a page instead of from adrenaline, and a page you have to scroll on a phone during a call is not that page. One sheet, branches down the left, your line for each on the right.
Write your walk-away line in your own words first, before the model drafts one. Compare them afterward. If the model's version is better, take the structure and keep your words — this is the one line where sounding like yourself matters most, because it is the line you will say under the most stress.
Prerequisites
You need the outputs of Variations 1 and 2, or their equivalents — a decoded offer and an evidence-backed range with the strongest supporting points identified. You need a floor you decided before the offer arrived and deal-breakers you wrote down when nothing was at stake, ideally from Weeks 1 and 2 of this series. You need honest notes on what the interviews revealed about how much they want you, and honest facts about any other processes, including the fact that there are none. Set aside roughly ninety minutes, because the rehearsal is the part people skip and it is the part that works. And if the offer contains a non-compete, a clawback, an arbitration clause, or an intellectual property assignment that concerns you, have an employment lawyer read it before the negotiation rather than after — that review belongs with a professional, not with any of these prompts.
Required tools
This variation benefits from a paid or higher-capacity tier of a general assistant, mainly for context length — the playbook, the rehearsal transcript, and the critique in one session add up, and a model that loses the early instructions will drift on exactly the rules that matter here. Voice input helps considerably during Part 4 if your tool offers it. A document or notes app for the printed playbook is genuinely necessary rather than optional. No browsing, plugins, or specialized software are required, and the prompt is deliberately built so that the AI never needs external data access, because everything factual comes from you.
Frequently asked questions
Isn't a scripted negotiation obviously scripted? Only if you deliver the script. The playbook exists so you have decided in advance what you want and what you will say in each direction, not so you can read sentences aloud. Rewriting every line in your own words and rehearsing until they sound natural is the difference between a prepared candidate and a candidate reciting, and it is why the prompt ends by handing the language back to you. A negotiator working from a well-internalized map sounds calm, not scripted.
Can I use this to write the negotiation email itself? You can use it to draft one, and then you must rewrite it. This is the exact point where the series' rule is most load-bearing — a negotiation email lands on the desk of the hiring manager and recruiter you will work with for years, many of whom now recognize machine-written text on sight, and being pegged as someone who outsourced their own counter costs you twice. Use the draft for structure and for the moves you would not have thought of, then write it in your voice, and check that every claim in it is true.
What if I have no leverage at all? Then the playbook is more valuable, not less. Most of what changes an outcome in a single-offer negotiation is not leverage but composure — naming a number once, not filling the silence, not trading against your own priorities, and knowing your floor well enough that you are not deciding it live. Parts 3 and 4 are built for exactly that reader, and the third example above is that reader.
Should I tell them I have another offer if I do? Yes, if it is real, stated accurately, and framed around timing rather than as a threat. What you should not do is imply, exaggerate, or invent one, which is why the prompt refuses to script a bluff and substitutes an honest version instead. Details get checked more often than people expect, industries are small, and the downside is not a lost negotiation but a lost offer. If you have nothing in flight, leave the field blank and let the playbook build the case that actually exists.
What if the whole thing goes badly and they pull the offer? Rescinded offers over a professionally delivered counter are rare enough that the fear does far more damage than the risk, mostly because employers who invested weeks in a hiring process do not restart it over a respectful question. What genuinely does raise the risk is a counter delivered as an ultimatum, a bluff that gets tested, or a candidate who negotiates repeatedly after a term is settled — all three of which this playbook is structured to avoid. Ask once, clearly, with warmth and a number, and then let them answer.
Recommended follow-up prompts
Build the acceptance-and-confirmation prompt. Once terms are agreed, ask the AI to structure a short confirmation email listing every agreed term for the record — and then write it yourself. Verbal agreements about start dates, review timing, and title have a way of not appearing in the final paperwork.
Run a first-ninety-days prompt against the scope signals. Everything the hiring manager revealed about what they need from this role is also the raw material for what you do after you start, and the negotiation notes are the freshest source you will ever have on it.
Take the outcome into Week 8. The final numbers and everything the negotiation revealed about the company feed directly into the last week's decision, which is where the offer stops being a number and becomes a choice.
Tags and categories
Tags:
salary negotiation, counter offer, negotiation scripts, scenario planning, walk-away number, competing offers, rehearsal, roleplay prompt, equity, career, job search
Categories:
Career and Job Search, Advanced Prompting
Citations
Roger Fisher and William Ury, Getting to Yes — the source of the best alternative concept underlying this variation's floor and walk-away structure.
Anthropic, prompt engineering documentation at docs.claude.com — for role assignment, explicit constraints, and structuring multi-part instructions.
OpenAI, prompt engineering guide in the OpenAI platform documentation — for the general principles of specificity and staged instruction used in the phase and exit-token design.
Which of the three should you use?
These three prompts are not one idea at three lengths, and they are not ranked. They do different jobs at different moments in the same week, and the sequence they form is the real deliverable. The Offer Decoder is a reading task — it takes a document that was written to be persuasive and turns it into a list of facts, absences, and questions, with no strategy in it at all. The Evidence File is a research task, and it is the one with homework attached, because the only trustworthy market numbers in this entire process are ones you collect yourself. The Negotiation Playbook is a performance task, and it assumes both of the others are done. Run in order, they take you from a document you are afraid to read to a conversation you have already had once.
Where they overlap is the constraint that shapes all three. None of them lets the AI tell you what anything is worth. The decoder refuses to price equity, the evidence file refuses to estimate a range, and the playbook refuses to promise an outcome — and in each case the refusal is the feature. What the AI contributes is decoding, structure, drafting, and opposition, which are the four things it does genuinely well and the four things a nervous person at eleven at night does badly. The facts stay yours because they have to be defensible under a question from someone whose job is asking them.
Choose by where you actually are. If the offer arrived in the last day and you have not yet read it properly, start and stop with Variation 1 tonight; it is a ten-minute job and it will let you sleep. If you have read it, know what is fixed, and have three or four days before you respond, Variation 2 is where the negotiation is actually won, because a range you can source is what makes everything in Variation 3 hold together. If the conversation is scheduled and you have your evidence, go to Variation 3 and give the rehearsal the ninety minutes — it is the section readers skip and the one that changes how the call goes. And whichever you run, the last step is the same one: the words that reach a human have to be yours, out loud, until they sound like you.
TAGS:
