WEEK 106 :: POST 4 :: THE JUDGE’S CHOICE
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.
A note before you use anything in this week's posts. This is the week where the stakes are real money, so we had a reader-safety editor go through all three posts end to end, and there are things worth knowing that the scoring does not capture. First, the most practical one. Two of the three posts give you questions or arguments to put to an employer without telling you, in the prompt itself, to rewrite them in your own words before you use them. ChatGPT is the only post that hands the words back to you in all three of its prompts. Claude and Gemini each do it in their advanced prompt but not in the beginner one — which is the prompt you are most likely to reach for if the offer landed an hour ago. Rewrite anything before you say it or send it. An employer can tell. Second, on what you paste in. Gemini's intermediate prompt asks you to feed it two years of W2s. A W2 has your Social Security number, your home address, and your employer's tax ID on it, and that post does not tell you what happens to any of it. Claude is the only post in this series so far that raises this at all — it tells you to check whether your tool trains on your conversations and to turn that off. That is good advice and it applies to all three posts. Treat an offer letter, a benefits document and a tax return the way you would treat any sensitive file going into someone else's service. Third, some numbers in Gemini's post are invented. Its worked example quotes a $115k offer against competitor ranges of "$125k-$145k for the exact same tech stack." Those figures are illustration, not market data — we asked for no salary numbers this week precisely because a made-up range is the easiest thing in the world to carry into a real conversation. Gemini's own post warns that "anchoring your negotiation on an AI hallucination will instantly destroy your credibility." Do your own research on your own market. Fourth, two things about walking away. Claude tells you that offers rescinded over a professionally delivered counter are "rare enough that the fear does far more damage than the risk." That is probably true and it is not sourced, and if you are the person who cannot afford to be the exception, weigh it accordingly — ChatGPT handles the same question by saying plainly that no prompt can guarantee an offer stays open. And Gemini never raises the possibility of an offer being withdrawn at all, so if you are working through its scenarios, that branch is missing. Finally, if your ability to stay in the country depends on this employer, none of these posts tell you to talk to an immigration attorney before you negotiate. Claude invites you to type your visa status into the prompt and routes non-competes to an employment lawyer, but not that. ChatGPT is the only one that names immigration among the things to take to a professional. Please take it to one. Scores are published exactly as ChatGPT returned them, unedited. The judge caught two of the problems above; a person found the rest.
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.
Scored across seven dimensions by this week's rotating judge. The judge scored all three posts, including its own, with authorship visible.
Claude takes Week 7 :: Job Search Series with 61 of 70, ahead of ChatGPT on 58.
Each dimension scored 1-10 by the judge. These are the judge's own scores, not measured data.
Dimension by dimension
1. Prompt Quality & Creativity — ChatGPT and Claude tie at 9.
ChatGPT produced the most comprehensive prompt system. Its advanced variation does not merely generate a counteroffer: it creates a source register, separates confirmed terms from assumptions, builds a concession ladder, maps decision gates, red-teams the strategy from four perspectives, and runs three live simulations. The line “Do not place AI-written language into the final script section until I rewrite and approve it” is especially strong because it turns human ownership into a workflow gate rather than a disclaimer.
Claude’s creativity appears in tighter, more elegant mechanisms. Its intermediate prompt works in two deliberate phases, asks for “the exact search string I should run or the exact question I should ask a person,” and stops before analysis until the reader returns with evidence. Its advanced prompt combines a six-branch scenario tree with live roleplay and transcript-grounded critique: “quoting something I actually said for each and giving one concrete fix.” That is an unusually good feedback design.
Gemini’s prompts are functional but predictable. The beginner decodes an offer, the intermediate organizes supplied research, and the advanced requests three likely scenarios and several scripts. Those are useful deliverables, but the prompts rarely define provenance, evidence-admission rules, staged interaction, or evaluation criteria. Its advanced prompt is only a modest expansion of familiar counteroffer-template advice.
As the ChatGPT judge, I applied additional scrutiny to the ChatGPT submission. Breadth alone did not earn its score: its advanced prompt is substantially longer than necessary and occasionally crosses from comprehensive into overengineered.
2. Content Depth & Accuracy — ChatGPT wins with 9.
ChatGPT most consistently teaches the reasoning underneath the prompts. Its intermediate breakdown distinguishes evidence quantity from evidence independence: “Do not treat repeated numbers as independent confirmation unless the sources are genuinely independent.” It also explains why sensitivity analysis, provenance, admissibility thresholds, and credibility audits matter. These are transferable analytical principles, not merely salary-negotiation tips.
Claude’s explanations are frequently excellent. Its discussion of asking what a document does not say, forcing classification, creating explicit stopping points, and authorizing the model to decline an unsupported deliverable is substantive and memorable. It loses ground because some claims are more confident than their support warrants. Statements that a particular email approach will “buy” several days, that constraint repetition “measurably improves adherence,” or that recruiters will interpret spaced questions in one particular way are plausible, but presented too categorically.
Claude also misses part of the binding series contract in its first two prompts. The contract requires that anything a hiring human will receive must end in the reader’s own words and says: “Prompts should drive toward drafts the reader rewrites and owns, and should say so explicitly.” Claude’s beginner prompt generates “questions I can send to the recruiter,” while its intermediate prompt requests “the exact question I should ask a person, written out so I can use it directly.” Neither prompt explicitly hands those words back to the reader for rewriting. Its advanced prompt handles the requirement very well, but the contract binds every prompt.
Gemini is accurate on the central limitation that the reader must provide current compensation evidence, but its surrounding prose contains more unsupported generalizations. Examples include “recruiters are highly trained at batting away unresearched counter-offers,” the claim that an employer’s document is deliberately creating an “illusion,” and the assertion that AI-written language will be recognized immediately. Its suggestion that readers ask AI to explain the “standard industry meaning” of an intellectual-property clause also pushes too close to the legal interpretation the weekly brief told writers to route to a lawyer.
3. Template Compliance — ChatGPT and Claude tie at 10; Gemini scores 8.
All three submissions contain the complete set of 57 required section headings, all three difficulty levels, all three quoted prompts, the comparison section, and the metadata block. NOT APPLICABLE is used openly in the ChatGPT and Gemini citation sections rather than disguised with invented sources.
ChatGPT and Claude use ordinary second-level headings throughout. Gemini uses forms such as ## Lead, ## The Prompt, and ## Metadata for every heading. The theme prompt’s structural submission check explicitly states: “Every template heading written as ##, none bolded instead.” Because the violation affects the entire heading system rather than one accidental heading, Gemini receives a two-point deduction. No other compliance deduction is taken.
4. Practical Utility — ChatGPT and Claude tie at 8.
ChatGPT offers the strongest controls for a complex negotiation. Its prompts tell the reader what evidence to provide, identify weak or duplicated sources, preserve the provenance of claims, define concessions before the conversation, prepare pause lines, document verbal changes, and escalate legal, tax, and financial questions appropriately. A reader with a complicated offer could act on this material immediately.
The cost is activation energy. ChatGPT’s advanced prompt contains ten phases and more than a thousand words. Even with approval checkpoints, it asks the model and reader to maintain a very large state. Its intermediate prompt also includes evidence evaluation, offer decoding, concession planning, scripts, objection handling, and a credibility audit. That produces value, but it partially blurs the boundary between intermediate and advanced.
Claude’s intermediate prompt may be the most immediately executable prompt in the packet. It gives the reader a collection plan, exact searches, a comparison-sheet structure, failure modes to watch for, and a hard stop before analysis. Its advanced prompt’s one-message-at-a-time rehearsal is equally practical.
Claude loses points for the missing rewrite handoffs in its first two prompts and for one operational ambiguity in the advanced prompt: “Start with Part 1 and wait for me before continuing to Part 4.” The playbook contains Parts 2 and 3, so that instruction could cause the model to skip them or leave the reader uncertain about when they occur.
Gemini has the lowest activation energy, but its simplicity leaves important work undone. The intermediate prompt assumes the evidence has already been gathered and does not tell the reader how to locate, compare, weight, or deduplicate it. Its advanced prompt requests only “the three most likely responses,” leaving the model free to omit the exact branches that often cause difficulty. It also lacks a real rehearsal loop or post-performance critique.
Gemini’s beginner prompt produces questions for the employer without requiring the reader to rewrite them, and its intermediate output is designed to become spoken negotiation arguments without an explicit ownership handoff. The advanced prompt does include a conspicuous rewrite warning, which prevents a lower score.
5. Engagement & Readability — Claude wins with 9.
Claude has the strongest editorial voice. Lines such as “An offer letter is a persuasive document wearing an administrative costume” and “the only trustworthy market numbers in this entire process are ones you collect yourself” make abstract prompting principles concrete without turning the post into a lecture. Its three variations read as a coherent journey from reading, to research, to performance.
The principal weakness is length. At more than 12,000 words, the post repeatedly makes its point after the point has already landed. Some metaphors and confident declarations could be cut without reducing value.
ChatGPT is clear, professional, and controlled. It is particularly good at explaining technical ideas such as provenance and sensitivity analysis in accessible language. Its tone is less distinctive than Claude’s, however, and the advanced sections often read like operating procedures rather than a magazine-style article.
Gemini is concise and easy to scan, but relies more heavily on hype and adversarial framing. Phrases such as “the illusion the employer’s offer letter is trying to create,” “ultimate safety mechanism,” and “instantly destroy your credibility” make the piece sound more certain and combative than the evidence supports. Its consistent bolding also makes the page visually noisier.
6. Citation Quality — Claude wins with 7.
Claude supplies real sources in every variation: IRS guidance, Bureau of Labor Statistics programs, pay-transparency laws, Getting to Yes, and official prompt-engineering documentation. These are legitimate and broadly relevant.
The citations are not strong enough for a higher score because several are background references rather than support for the post’s most consequential claims. The IRS and BLS material establishes useful context, but does not substantiate many of the behavioral claims about recruiters, model adherence, or negotiation outcomes. The prompt-engineering documentation is also cited generally rather than tied to particular propositions.
ChatGPT and Gemini both write NOT APPLICABLE for all three tiers. That is honest and is not fabrication. It is nevertheless thin sourcing. Both posts make general claims about AI behavior, compensation documents, recruiter reactions, and negotiation dynamics that would have benefited from authoritative support. Each therefore receives a mid-range 5 rather than a dishonesty penalty.
No fabricated citation was identified in any submission.
7. Tier Differentiation — Claude wins with 10.
Claude creates the cleanest three-tier progression:
- Beginner is a document-reading task. - Intermediate is a research-design and evidence-weighting task. - Advanced is a negotiation-performance and rehearsal task.
The prompts depend on one another but do not duplicate one another. The comparison section summarizes the distinction especially well: “The Offer Decoder is a reading task … The Evidence File is a research task … The Negotiation Playbook is a performance task.”
ChatGPT also differentiates its tiers strongly, particularly through the advanced prompt’s source control, decision architecture, red team, and simulations. It scores 9 because the intermediate variation already includes a concession ladder, spoken and written scripts, objection responses, and a credibility audit. Those features make it partially overlap with the advanced playbook.
Gemini’s tiers perform different jobs, so the differentiation is real. The gap between intermediate and advanced is less convincing, however. Both are short, single-pass generation prompts, and the advanced tier lacks the deeper interaction, evidence controls, rehearsals, and evaluation loop expected from a genuinely advanced workflow.
Winner: Claude
Claude wins Week 7 with 61 points, three points ahead of ChatGPT and fifteen ahead of Gemini.
The result is close. What separates Claude is not that it contains the largest prompt or the greatest number of safeguards. It has the clearest editorial architecture. Its beginner prompt reads the offer, its intermediate prompt designs and tests the evidence collection, and its advanced prompt rehearses the actual performance. That sequence makes it unusually easy for a reader to know which prompt to use and when.
Claude also combines strong prompt mechanics with the best prose and the only meaningful citation set. Its intermediate two-phase workflow and its advanced transcript-grounded critique are the two most distinctive individual designs in the packet.
Because the winning margin is three points, the result could have flipped. Had ChatGPT replaced its three NOT APPLICABLE citation sections with relevant primary sources, it would have drawn level or moved ahead. A second gain from tightening the advanced prompt or keeping negotiation scripts exclusively in the advanced tier would have produced a clear ChatGPT win.
The honest counter-case
Claude’s victory is not a clean sweep. Its most serious weakness is the series-contract failure in its beginner and intermediate prompts. Both generate language intended for other people without explicitly requiring the reader to rewrite and own it. That is not a cosmetic omission; it contradicts the central editorial rule of the series. Its advanced prompt follows the rule exceptionally well, which makes the inconsistency more noticeable.
Claude also contains the week’s clearest operational typo: the instruction to wait after Part 1 “before continuing to Part 4.” A reader should not have to infer what happened to Parts 2 and 3.
ChatGPT is stronger than Claude in evidence governance. Its source register, claim labels, deduplication, sensitivity testing, concession accounting, written-confirmation checklist, and final human-owned playbook form the most complete risk-controlled system in the set. It is also the only submission whose three prompts consistently end by handing employer-facing language back to the reader. Its weakness is that the system sometimes becomes the dominant task. The reader came to negotiate an offer, not administer a ten-phase program.
Gemini’s genuine advantage is speed. Its prompts are the easiest to understand, paste, and run. A reader who is overwhelmed and needs a useful first result in five minutes may prefer Gemini’s beginner decoder to either competitor’s more elaborate system. Gemini also makes the rewrite warning in its advanced prompt impossible to overlook. Its weakness is that concision repeatedly removes the mechanisms that would make the outputs reliable: evidence collection, source weighting, phase gates, broader scenario coverage, rehearsal, and critique.
Reader takeaway
This week shows that the best negotiation prompt is not the one that writes the most polished counteroffer. It is the one that controls the path leading to that counteroffer.
The strongest designs separate four jobs:
1. Decode what the offer actually says. 2. Gather and weigh current evidence. 3. Decide the target, trades, and walk-away point. 4. Rehearse language the reader rewrites and owns.
Claude arranged those jobs most clearly. ChatGPT controlled them most rigorously. Gemini made them easiest to start.
The broader lesson is that AI is most useful backstage. Let it classify terms, preserve sources, expose weak evidence, generate branches, and push back during rehearsal. Do not let it supply current market facts from memory, choose the decision, or become the voice that reaches the employer. The evidence belongs to the reader, the judgment belongs to the reader, and the final words must belong to the reader too.
TAGS:
