AI Prompts for Career Coaching: A CV and LinkedIn Review Workflow

Coding Liquids tutorial cover featuring Sagnik Bhattacharya for AI Prompts for Career Coaching: A CV and LinkedIn Review Workflow.
Coding Liquids tutorial cover featuring Sagnik Bhattacharya for AI Prompts for Career Coaching: A CV and LinkedIn Review Workflow.

Six prompts cover a CV and LinkedIn review: a targeting check against real job adverts, an evidence audit, an applicant-tracking keyword check, a headline and About pass, a CV-to-profile consistency check, and a question list for the client. The coach still judges direction, honesty, voice and what to leave out.

The order matters more than the wording. Consent and an intake form come first, so every prompt runs on the client's own facts rather than invented ones. Each prompt is also built to ask the client for missing numbers instead of making them up, because a CV that overstates is a risk to the client and to you. One illustrative client runs through all of it.

Follow me on Instagram@sagnikteaches

Consent, and what you remove before pasting anything

A CV is a bundle of personal data: name, contact details, employers, dates, sometimes a photograph or a date of birth. Before an assistant sees any of it, the client should have agreed, in writing, to how you use it. A short clause in your engagement letter is enough. Adapt this wording to what you actually do, and only promise what you will do (illustrative, not legal advice):

Connect on LinkedInSagnik Bhattacharya

To give you faster and more detailed feedback, I sometimes use AI writing tools on your documents. Before anything is entered I remove your name, contact details and other identifying information. I use an account where the provider does not train its models on my content, and I delete the chat when your review is finished. I will not enter anything you have marked private. If you would rather I did not use AI on your documents, tell me and I will review them by hand. By signing, you confirm you have read this.

If you also want to use anonymised before-and-afters in your marketing, ask for that separately, as an opt-in line the client can decline without affecting the review. Folding it into the AI clause makes both consents weaker. Mentioning AI use in client contracts and proposals gives more wording options, and keeping coaching sessions confidential with AI note takers covers the session side of the same promise.

Subscribe on YouTube@sagnikteaches

Strip these before pasting: name, email, phone, address, date of birth, photograph, LinkedIn address, referees' details and any salary figures. If the client is searching confidentially and their employer is identifiable, replace the employer's name with [EMPLOYER] too. Keep each client in a separate chat or Project, with memory switched off, so that one client's details cannot surface in another's review. On the tool side, an individual plan needs its model-training switch turned off, and the business plans from the main vendors state that they do not train on business content by default. Setting labels change, so read the screen rather than a guide, and note that a business plan costs at least two seats on both ChatGPT Business and Claude Team.

The Complete Excel Course with AI Integration thumbnail

The Complete Excel Course with AI Integration

From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.

By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner

Buy now

The intake form that feeds every prompt

The intake form is what stops the prompts from guessing. It collects the target, the evidence and the constraints once, in the client's words, and every later prompt draws on it. Here is a filled-in version for the illustrative client, an operations coordinator at a 40-person HVAC contractor who wants to move into project management (all details invented):

FieldClient's answer
Target role and levelProject manager, junior to mid level, building services contractors.
Three target advertsPasted as adverts A, B and C, with the client's note on which they like best.
Current roleOperations coordinator for six years; started as a dispatcher.
Numbers the client can proveSchedules 11 technicians; about 70 jobs a week; coordinates 4 subcontractors on larger installs; missed appointments fell from about 9 a month to 3 after a shared dispatch calendar (client's own tracking sheet).
QualificationsHealth and safety certificate (valid); project-management foundation course (in progress).
Gaps or worriesNo formal project title; a four-month gap in 2019 the client is happy to explain.
ConstraintsOne month's notice; will not relocate; salary floor shared with the coach only.
ConsentSigned on [date]; AI allowed for CV and LinkedIn only; identifiers removed.

The "numbers the client can prove" row is the one that matters most. It becomes the source list for every claim in the finished documents, and it is what the claims ledger later in this tutorial checks against. If the client cannot back a number with a record, a payslip, a report or a message, it stays out of the CV. The "gaps or worries" row is for you, not the assistant: it records what you will coach, not what the tool should hide.

Prompt 1: does the CV fit the roles the client wants?

Before polishing a single bullet, find out whether the CV is aimed at the right target. This prompt compares the CV with the three adverts and lists each requirement, where the CV shows it and how strong the evidence is. It generates questions for the client rather than fixes.

I am a career coach. Compare the CV below with the three job adverts.
Use only the text I paste. Do not assume anything about the client.

ADVERTS: [A], [B], [C]
CV (identifiers removed): [paste]

Return a table with four columns:
1. Requirement that appears in at least two adverts
2. Where the CV shows it (quote the line) or NONE
3. Strength: Clear, Thin or Missing
4. A question I should ask the client (never an answer)
Then list requirements the CV mentions that none of the adverts ask for.

Illustrative output, abridged to four rows:

Requirement in the advertsWhere the CV shows itStrengthQuestion for the client
Schedule and coordinate trades"Scheduling technicians; coordinating subcontractors on larger installs"ClearHow many, and over what size of job?
Budget or cost trackingNONEMissingDo you track costs for any job, even informally?
Client communication"Handling customer enquiries"ThinWhat volume, and any escalations you resolved?
Project-management qualification"Foundation course in progress"ThinWhich course, and when does it finish?

The judgement checks are the coach's. The assistant can say "Missing"; only you can say whether the gap is fatal or fixable in the time the client has. Ask whether the three adverts are a fair sample of the target (three roles at similar firms tell you little about a different kind of employer), whether the client is aiming too high, too low or at the wrong sector, and whether a "Missing" item is a real gap or just something the client has done and not written down. The last case is common, and it is why the output is a list of questions.

Prompt 2: the evidence audit, and one bullet from start to finish

Most weak CVs describe duties, not results. This prompt sorts each bullet into one or the other and drafts a stronger version using only facts already in the CV. Where a number or outcome would help, it inserts a bracketed question instead of a figure. That one rule is what keeps invented achievements out.

Review each CV bullet below. For each one:
1. Say whether it describes a duty or a result.
2. Rewrite it using ONLY facts that appear in the bullet or in the intake
   form. Where a number, scale or outcome would strengthen it, put a
   bracketed question in its place, for example [HOW MANY TECHNICIANS?].
   Never supply a number yourself.
3. List the questions for the client in one place at the end.

INTAKE FORM: [paste, identifiers removed]
BULLETS: [paste]

Here is the main worked example, one bullet in four stages.

  • Original: "Responsible for scheduling technicians and handling customer enquiries."
  • Assistant's first rewrite: "Scheduled [HOW MANY TECHNICIANS?] technicians across [HOW MANY JOBS A WEEK?] jobs and handled customer enquiries, [ANY MEASURABLE RESULT?]." The questions at the end ask for the headcount, the weekly volume, and any change in missed appointments, response time or complaints that the client introduced or measured.
  • Client's answers: 11 technicians; about 70 jobs a week; missed appointments fell from about 9 a month to 3 after the client set up a shared dispatch calendar, with the owner's approval. The tracking sheet is on file.
  • Final bullet: "Scheduled 11 technicians across about 70 jobs a week; set up a shared dispatch calendar that cut missed appointments from about 9 a month to 3."

The final bullet is no longer a duty: it has scale, an action and a result, all traceable to the intake form. The coach's judgement still has two jobs here. First, causation: the calendar was introduced and appointments fell, but was anything else changing at the time, such as a new hire or a quieter season? If so, "cut" claims too much and "alongside" is safer. Second, ownership: "set up" is accurate only if the client did the work, so ask. These are conversations, and a prompt cannot have them for you.

Run the audit on all of the CV's bullets, but cap each client at one round of questions per bullet. Clients answer a short, specific list well and a long one badly. If a bullet cannot be strengthened because the client has nothing to measure, say so and move on. A plain, accurate duty line is better than an inflated result.

Prompt 3: an ATS keyword check that stays honest

Applicant tracking systems (ATS) store applications and let recruiters search and filter them. How a particular employer has set theirs up is invisible to you, so the useful question is not "will this pass?" but "does the CV use the same words as the advert wherever the client can truthfully claim them?" This prompt compares the CV with the adverts and labels each term claimable or not.

From adverts A, B and C, list the skills, tools, qualifications and
phrases that appear in at least two of them. For each, check the CV text
and the intake form and return:
- Exact phrase in CV: yes or no
- Variant in CV: quote it, or NONE
- Can the client truthfully claim it? Answer YES only if the intake form
  or CV shows it. Otherwise write NOT CLAIMABLE.
- Suggested placement, only for YES items.
Do not suggest adding any term marked NOT CLAIMABLE.

Illustrative output, abridged:

Term in the advertsIn the CVClaimable?Action
Project schedule"Scheduling" onlyYES: client builds schedules for larger installsUse "project schedules" in the subcontractor bullet
Subcontractor management"Coordinating subcontractors"YESChange to "subcontractor management" in one bullet
MS ProjectNot presentNOT CLAIMABLE: client uses a shared calendar and spreadsheetNone; discuss a short course
Project-management certification (desirable)"Foundation course in progress"ONLY AS "IN PROGRESS": the intake form shows a course under way, not a passName the course, say "in progress" and add the expected date

The "NOT CLAIMABLE" column is what separates a check from keyword stuffing. Adding a tool the client has never opened is a risk the moment an interviewer asks about it. A match score from any tool is an estimate of relevance, not the employer's result, so use scores as prompts for conversation. Add two practical tests that cost nothing. Copy all the text out of the client's file into a plain text editor: if the order is jumbled or content is missing, a parser may struggle too. And check that dates, titles and contact details are in the body of the page, not in a text box, header or footer, which some systems can skip. Resume formatting AI tools for recruiters describes the employer-side tools that rewrite layouts, and how AI screens CVs is worth sharing with a client who fears a robot will reject them.

Prompt 4: LinkedIn headline and About, inside the limits

LinkedIn fields have length limits. LinkedIn's own advice on profile summaries puts the About section at 2,600 characters. Third-party guides give 220 characters for the headline and 2,000 for each role's description, but LinkedIn's help pages do not state those two, and the profile editor shows a live counter. Treat that counter as the authority. Then preview the profile as a visitor would, on a phone and on a laptop: wherever the About section is cut short, the opening lines carry most of the weight.

Using only the intake form and the CV below, draft:
1. Three LinkedIn headline options, each under 220 characters, built
   from the client's current role, target role and one proven number.
   Do not use any job title the client has not held.
2. An About opening of two or three sentences, first person, plain words.
   No claims that are not in the intake form. No phrases such as
   hard-working, motivated or passionate.
For each option, state the character count.

Illustrative output for the headline, with the original for comparison:

VersionHeadlineCharacters
OriginalOperations Coordinator at [EMPLOYER]36
Option AOperations coordinator moving into project management | Scheduling 11 technicians and 4 subcontractors in building services123
Option BBuilding services operations coordinator | Dispatch, subcontractor coordination, customer scheduling | Seeking project manager roles132
Option CProject coordinator, building services | Scheduling, subcontractor coordination, customer communication | Health and safety certified133

Option C is the trap. It reads well and fits the length, but "project coordinator" is a title the client has never held, and a recruiter who checks it against the CV will notice. Strike it, and add the "no titles not held" rule to the prompt. Options A and B differ in a way only the coach and client can decide: A signals a move, B announces a search. If the client's employer follows their profile, even "seeking" or "moving into" may be more public than they want, so ask before anything goes live. Count the characters yourself in the editor, because a model's count can be a few off.

For the About section, here is a before and after. The original: "I am a hard-working and motivated professional with six years' experience in the HVAC industry, looking for new challenges." The draft opening: "I keep 11 technicians and about 70 jobs a week on schedule for a 40-person HVAC contractor. After six years in dispatch and coordination, I am moving into project management, where the same habits apply on a larger scale." The coach's edit is to ask the client to read it aloud. If "habits" sounds like a word they would never say, they change it. Voice is not a nice-to-have in a profile that is supposed to sound like a person.

Prompt 5: the CV and the profile must tell the same story

A recruiter can check a CV against the LinkedIn profile, and mismatches in titles or dates raise questions that have nothing to do with ability. This prompt lists differences without deciding which version is right, because only the client knows.

Compare the CV and the LinkedIn text below. List every difference in
job titles, employers, dates, qualifications and skills. For each,
quote both versions. Do not say which is correct. Add a question for
the client.

Illustrative output:

ItemCVLinkedInQuestion for the client
Title, 2021 to 2024Senior coordinatorCoordinatorWhich was the official title, and do you have a letter or payslip showing it?
Start dateMarch 2019June 2019Which date is right?
SkillsNo project software listedMS ProjectHave you used it for real work?
CertificateHealth and safety certificateNot listedAdd it to the profile?

Never let the assistant resolve a mismatch. Choosing "senior coordinator" because it sounds better is exactly how a CV becomes untrue. The client confirms the facts, and you record the answer on the intake form so later prompts use the corrected version.

Prompt 6: one list of questions for the client, not five

By now prompts 1 to 5 have each produced questions, and sending the client five overlapping lists is a good way to get five half-answers. The last prompt merges them into one short list, ordered by importance, with the reason for each question and the proof that would settle it.

Below are the questions generated by prompts 1 to 5. Merge duplicates
and write one list for the client, in plain words:
- Maximum eight questions, most important first.
- For each: why it matters in one line, and what proof would help
  (a record, a message, a figure from a report).
- Neutral wording only. Do not suggest the answer.

Illustrative output, abridged to four of the eight:

  1. How many people or firms did you coordinate on larger installs? Why: the adverts ask for trade and team coordination. Proof: a job sheet or rota.
  2. Do you track costs on any job, even informally? Why: the adverts mention budget tracking and the CV has nothing on it. Proof: a spreadsheet or report.
  3. Which date did you start in your current role? Why: the CV and the profile disagree. Proof: a contract or payslip.
  4. Which project-management course are you taking, and when does it finish? Why: a qualification is listed as desirable. Proof: the enrolment email.

The coach edits this list before it goes out. Questions about motives, gaps or confidence are better asked live than emailed. Some documents the client should not be sending around. And watch for leading wording: "Did you lead the subcontractors?" invites a yes, while "What was your role with the subcontractors?" invites the truth.

What the coach judges that no prompt can

Everything above produces drafts and questions. The decisions below are why a client pays for a coach and not for a subscription.

  • Whether the target is realistic. The assistant will tailor a CV to any role you name. Only you can tell a client that the gap between their experience and the advert is two years wide.
  • Which story to tell. A career change needs a through-line that the client believes. The tool offers phrasing; you choose the narrative.
  • What to leave out. Old roles, irrelevant qualifications and inflated responsibilities all need a human to cut them.
  • How to handle gaps and short tenures. The right answer depends on the client's circumstances and on how much they want to disclose.
  • Whether a claim is true and provable. This is the claims ledger below, and it is a person's job.
  • The client's voice and comfort. A profile the client cannot read aloud without wincing will not survive an interview.
  • What the profile signals to a current employer. That is a risk assessment about this client's workplace, which no prompt has seen.

How an invented achievement slips in, and the ledger that catches it

Here is the failure to expect, as an illustration. A draft CV comes back polished, and one bullet now reads "led a team of 12 subcontractors on commercial installs". The client's own words were "coordinated subcontractors on bigger jobs", and the intake form says four. The assistant rounded up, the sentence sounds good, and the client, reading quickly and pleased, nearly keeps it. It would have come out at interview, the first time someone asked about the team of 12.

The defence is a claims ledger, run after every rewrite and before the client sees it. Ask the assistant to list every number, title and named tool in the new CV next to the intake line that supports it, or the word NONE. You then check the NONE rows.

Claim in the new CVSource on the intake formDecision
Schedules 11 techniciansNumbers the client can prove, item 1Keep
About 70 jobs a weekNumbers the client can prove, item 2Keep
Missed appointments cut from about 9 a month to 3Client's tracking sheetKeep, with the date range added
Led a team of 12 subcontractorsNONE (the form says 4 subcontractors coordinated)Remove, restore to "coordinated 4"

The ledger takes about five minutes, and it works because it makes the assistant's confidence irrelevant. The question is not whether a claim sounds plausible but whether the client has a record for it. A claim without one comes out, however good it sounds.

Keeping the client file tidy after the session

When the review ends, three things happen. Delete the chat or the Project, as your consent clause promised. Save the final CV, the final profile text, the completed intake form and the claims ledger in your own client file, where the client's consent and the facts sit together. And write down the decisions you made, such as which headline the client chose and why, because those are the part of the work that is yours. AI session notes and follow-ups for coaches covers turning that record into the next session's plan.

A rough sum for one client, labelled illustrative: by hand, intake review 15 minutes, targeting check 30, bullet rewrites 60, keyword check 20, LinkedIn 40, consistency check 15, which comes to 3 hours 0 minutes. With the prompts, the same steps take about 10, 10, 25, 8, 20 and 5 minutes, plus a 10-minute claims ledger, for about 1 hour 30 minutes. The saved time is best spent on the conversations that make the review worth paying for: the target, the story and the voice.

Career coach questions about AI, consent and ATS checks

Do I need client consent before using AI on a CV?

Get it in writing before you start. A CV holds personal data and employment history, so tell the client which tool you use, what you remove first, whether the provider trains on content, how long you keep the chat, and that they can opt out. Ask a data-protection adviser which rules apply to your practice.

Can an ATS keyword check guarantee an interview?

No. An applicant tracking system stores applications and lets recruiters search and filter, and how a given employer has set it up is invisible to you. A keyword check tests relevance against the advert and catches formatting problems. A match score from any tool is an estimate, not the employer's result.

Should the AI write the client's LinkedIn About section?

Let it draft options from the client's own facts, then have the client rewrite them in their voice. The test is whether they would say those sentences aloud in an interview. Also check what the profile signals to their current employer before anything goes live, because that is a judgement for you and the client.

How do I stop the AI inventing achievements?

Tell it to ask for missing numbers in square brackets instead of supplying them, then run a claims ledger: every number in the new CV must trace to a line in the client's intake form. Anything without a source comes out, however good it sounds, because the client will be asked about it.

Further reads

Sources: OpenAI, Anthropic, Google Workspace and Microsoft privacy pages (business-plan training defaults); ChatGPT Business and Claude pricing pages (two-seat minimums); LinkedIn's own advice on profile summaries (About section limit); third-party guides for the headline and role limits, which readers should confirm in the profile editor.

Want a CV and LinkedIn review workflow for your practice?

On a 1:1 call we'll map your current review steps, decide which of them suit AI, and set up a consent wording, intake form and prompt set you can run with every client.

Book a 1:1 call with me