AI Guides › Workbench
By Nigel Guy · 7 min read
The usual move is to paste a CV into Claude and type "make this better". You get smoother adjectives, a few power verbs and a document that is no more likely to get an interview, because nobody told the model what the CV is being measured against. It feels like progress while the real problems (a missing requirement, a layout the parser mangles, bullets that describe duties rather than results) stay exactly where they were.
The rule: never ask for a better CV; ask for a diagnosis against one specific job advert, then fix only what the diagnosis finds.
An applicant tracking system (ATS) collects applications, parses your file into fields, and lets recruiters search and sort. Workday's own documentation says resume parsing populates application fields from the file and lets the candidate review the data, and that it does not auto-fill some fields (languages and skills are named). Parsing results vary with layout and word order. Third-party guidance consistently advises plain structure without images, tables or columns; treat that as sensible rather than official.
The popular claim that "the ATS rejects 75% of CVs" is not well supported. One recruiter survey, published as a press release by a CV-builder company (so treat it as a vendor source), reported that most recruiters said their system did not auto-reject on content. The part that is real is knockout questions: yes/no form fields such as right to work, a required licence or location. Those can reject you, and no CV rewrite changes your answer. Beyond that, the usual enemy is volume: a human skims quickly or searches parsed fields for keywords. So the audit below targets two readers: a parser and a tired person.
| Prompt | What it does | Cost | Best for | Catch |
|---|---|---|---|---|
| 1. Match diagnosis | Compares CV with one advert, lists what is missing or unclear | Free tier works | Starting every application | Only as good as the advert you paste |
| 2. Gap closer | Turns missing requirements into questions about your real experience | Free tier works | Honest gaps you can evidence | It must not invent; you supply facts |
| 3. Bullet rebuild | Rewrites each weak bullet as action, scope, result | Free tier works | Duty-list CVs | Needs real numbers from you |
| 4. Ten-second read | Simulates a skim and a keyword search | Free tier works | Final check | A simulation, not a real ATS |
Claude can read PDF and DOCX uploads, according to its support documentation. At time of writing the Free plan is $0 and Pro is $20 a month ($17 a month billed annually), roughly £15 and £13 at time of writing; check the £ price at checkout, as it varies by region and tax. You do not need Pro for this.
Collect: your current CV (as text or DOCX), the full job advert, and a "facts sheet" of things you have done that are not on the CV yet. Remove your home address, date of birth and referee details before uploading anything to any AI tool. Read Anthropic's current privacy settings for your plan before you paste personal data.
You are a careful recruiter who screens CVs for [ROLE_TYPE] roles in [COUNTRY].
Goal: tell me exactly why this CV may not progress for this job, so I can fix the right things.
Inputs:
- CV: [PASTE_CV_TEXT]
- Job advert: [PASTE_FULL_ADVERT]
Steps:
1. List every requirement in the advert, marked "essential" or "desirable" using the advert's own wording.
2. For each, say whether the CV shows it clearly, shows it weakly, or does not show it. Quote the CV line that supports your verdict, or write "no evidence".
3. Flag layout or wording that a parser or a quick skim could misread (odd headings, dates in inconsistent formats, job titles that differ from the advert's terms).
4. Rank the five most damaging problems.
Output: a table (requirement, status, CV evidence), then the ranked five.
Rules: do not rewrite anything yet. Do not assume experience that is not in the CV. If the advert or CV is missing or truncated, ask me before starting.
Before answering, check that every "clearly shown" verdict has a quoted line behind it.
Fill in: role type, country, CV text, advert.
You are an interviewer helping me recall genuine experience, not invent it.
Context: the diagnosis below found these gaps between my CV and the job. Gaps: [PASTE_RANKED_GAPS_FROM_PROMPT_1]. My facts sheet: [PASTE_FACTS_OR_WRITE_NONE].
Goal: find out which gaps I can honestly close with experience I already have.
Steps:
1. For each gap, check my facts sheet for matching evidence.
2. Where there is none, ask me up to three short questions per gap (what I did, what changed, how I know).
3. After I answer, mark each gap: close with evidence / wording issue only / real gap.
Output: a list of gaps with status and, for the ones I can close, one draft line using only what I told you.
Rules: never add a tool, qualification, number or employer I did not state. Real gaps stay real; suggest I address them in a covering letter or by learning the skill, not by claiming it.
Self-check: confirm every draft line traces to something I wrote.
Fill in: gaps from Prompt 1, your facts sheet.
You are an editor who writes CV bullets for [ROLE_TYPE] applicants.
Goal: rebuild each bullet so it shows what I did, how big the job was, and what changed.
Inputs: bullets for [JOB_TITLE_AND_EMPLOYER]: [PASTE_BULLETS]. Target advert terms to reuse where truthful: [PASTE_KEY_TERMS].
For each bullet:
1. Name what is weak (duty only, vague, no result, buried verb).
2. Ask me for any missing figure (team size, volume, time saved, money, before-and-after). Do not guess one.
3. Write one rewrite in this shape: strong verb, what I did, scale, result. One line each.
4. If no honest result exists, say so and keep the bullet short and plain.
Output: a table with columns original, problem, rewrite, question for me.
Rules: UK English, no first-person pronouns, no buzzwords, no invented numbers. Keep the facts identical; only improve the structure.
Before answering, check that no rewrite contains a number I did not give you.
Fill in: role type, job title, bullets, key terms.
You are a recruiter with ten seconds per CV and a keyword search box.
Input: my revised CV: [PASTE_REVISED_CV]. The job advert: [PASTE_ADVERT].
Steps:
1. Using only the top third of the first page, say what role you think I am and whether I look relevant. Be blunt.
2. List the twelve search terms you would type to find this candidate, then say which appear in the CV and which do not.
3. Name the single line you would cut and the single line you would move up.
Output: three short sections matching the steps.
Rules: do not praise. If the CV is too long or too short to judge, say so and ask for the missing part.
Fill in: revised CV and advert.
Run all four in order for a job you really want. For a quick application to a role you are only half-keen on, run Prompts 1 and 4 and skip the rest. Start a fresh conversation for each advert so one job's wording does not leak into the next. Save the final version as a clean DOCX or simple-layout PDF, then paste it into a plain text editor: if the order of sections survives, a parser will probably cope.