AI Guides › Workbench
By Nigel Guy · 7 min read
Most people ask an AI "how do I earn more?" and get a tidy list: negotiate, upskill, build a personal brand. It reads like advice and fits anyone, which is exactly why it is useless. The assistant knows nothing about your work, so it fills the gap with the average career.
The rule: never ask for advice until the assistant has your evidence, and never accept a conclusion until it has argued against it. The five prompts below run in order. The first three build the evidence, the fourth attacks it, and the fifth turns what survives into one small, testable move.
| # | Prompt | What it does | Cost | Best for | Catch |
|---|---|---|---|---|---|
| 1 | Evidence Dump | Turns your history into a structured record of what you did and what changed | Free tier of any major assistant is enough | Starting cold | Only as good as the detail you paste in |
| 2 | Underused Scan | Finds skills and outcomes you use but never get credit or pay for | Free tier | People who feel "just ordinary" | Flatters you if you are vague |
| 3 | Market Check | Turns your claims into questions to test against real pay and job data | Free tier, plus your own searching | Anyone considering a pay rise or move | The assistant cannot see live pay data reliably |
| 4 | Argue-With-Itself | Makes the assistant build the case against its own findings | Free tier | Stopping false confidence | It will sound convincing both ways |
| 5 | One-Move Test | Converts survivors into one 30-day experiment | Free tier | Getting out of analysis | Needs you to actually do the thing |
On cost: at time of writing, Claude has a free plan, Pro at $20 a month (about £15 at time of writing; check the £ price at checkout) and Max from $100 a month. You do not need a paid plan for this. A longer career history and several back-and-forth turns may hit free-plan limits sooner, which is the only reason to consider Pro. I could not verify current ChatGPT pricing for this guide, so check its pricing page; the prompts work in any assistant.
Gather, in one document: your CV, your last two or three appraisals or feedback emails, your current job description, and a rough list of what you are paid and what you have been offered before. Remove anything you would not want stored: colleague names, client names, employer confidential figures. Describe those generically ("a retail client with about 40 stores").
Use one conversation for prompts 1 to 3 so the evidence carries forward. Start a new conversation for prompt 4, pasting only the findings. That matters; see the trap below.
Fill in: your career history and the extra detail in the square brackets.
You are a careful career analyst. You work from evidence only.
Context: I am a [JOB_TITLE] in [INDUSTRY], based in [UK_REGION]. I have [YEARS] years of experience. My current pay is [SALARY_OR_RANGE]. My goal is [WHAT_I_WANT_MORE_OF: money / seniority / flexibility / security].
Here is my material: [PASTE_CV_APPRAISALS_JOB_DESCRIPTION].
Task: build a structured record of what I have actually done.
1. List every role, with the problems I was handed, what I did, and any outcome that could be counted (time saved, money, error rates, people managed, customers served).
2. Mark each outcome as STATED (it is in my material) or INFERRED (you are guessing).
3. List the gaps: places where an outcome is probably there but my material does not say.
Rules: do not invent figures. Do not give advice yet. If my material is too thin to do this properly, ask me up to eight specific questions first and wait for my answers.
Before you reply, check that every STATED item can be traced to a line I gave you.
Answer its questions properly. The quality of the next four prompts depends on this one.
Fill in: nothing new, if you stay in the same conversation. Add what you do outside work in the brackets.
Using only the evidence record above, plus this extra context: [UNPAID_OR_SIDE_WORK_YOU_DO_AND_ANYTHING_PEOPLE_ASK_YOU_FOR_HELP_WITH].
Find three kinds of mismatch:
A. Things I do regularly that are not in my job title or pay band.
B. Skills I have that colleagues or clients rely on me for informally.
C. Outcomes I produced that someone else got the credit or the budget for.
For each, give: the evidence line it rests on, how confident you are (high / medium / low), and what you would need to know to firm it up.
Output as a table. Do not pad. If you find fewer than three in a category, say so rather than stretching.
Fill in: the mismatches you think are real.
Take these claimed mismatches: [PASTE_YOUR_SHORTLIST].
For each one, do not tell me what it is worth. Instead write the questions I need to answer with real sources:
- Which job titles would explicitly list this skill?
- What would I search for on job boards, and which pay bands would I look at?
- Who could confirm it (a former manager, a peer in the same field)?
Then list any assumptions you are making about the UK labour market and label each as something I must verify. Do not quote salary figures from memory.
This is the step where you leave the chat. Two free, official places to use: the ONS Annual Survey of Hours and Earnings, which publishes pay by occupation (the 2025 edition came out on 23 October 2025; check for a newer one), and the National Careers Service job profiles on GOV.UK, which cover typical salary, qualifications and progression. Add a handful of live job adverts for the roles you are eyeing. Bring the real numbers back to the chat.
Open a fresh conversation. Paste the shortlist, the real numbers you found, and nothing else.
You are a sceptical hiring manager and a blunt career coach, in two voices.
Here is a claim about my career: [PASTE_ONE_FINDING], with the evidence behind it: [EVIDENCE] and the market data I found: [PAY_OR_JOB_AD_DATA].
Task:
1. Hiring manager voice: give the three strongest reasons not to pay me more or hire me for this.
2. Coach voice: say which of those reasons I could fix, which I could not, and which are just not true given my evidence.
3. State what single piece of new evidence would change your mind either way.
Rules: be specific to my evidence. No generic advice such as "network more". If my evidence is not enough to judge, say what is missing instead of guessing.
Before answering, check that each objection refers to something in my material or my market data.
Run it once per finding. Anything that survives with two or fewer unfixable objections goes forward.
Based on the surviving finding: [PASTE_FINDING] and the objections I could fix: [LIST].
Design one experiment I can finish in 30 days with about [HOURS_PER_WEEK] hours a week and a budget of [£_BUDGET].
Give me: the move, the exact first action this week, what result would count as success, what result would count as a no, and who I would show it to. Output as a short checklist.
Rules: one move only. No courses unless I name a specific one. If you need more information to be specific, ask first.
Agreement. An assistant tends to build on whatever you have already said, so a long single conversation drifts toward flattering you. That is why prompt 4 runs in a fresh chat, with the evidence rather than your hopes. A second trap is vagueness: "I manage projects" gives it nothing; "I ran a six-person team through a system migration" does.
Short on time: run prompts 1, 3 and 4. Prompt 2 is the one that finds hidden value, so keep it if you suspect you are underpaid. If you are happy and just curious, prompt 5 on its own is a waste: it needs the earlier findings.