How Recruiters Use AI to Write Candidate Summaries Clients Read

Give the AI three inputs, the client's brief, the CV and your own screening notes, and ask for a fixed one-page format that leads with fit against the client's must-haves, backed by evidence, then concerns and logistics. Then check every claim against the source before it goes out. Clients read summaries that answer "why this person, for this role".

Your screening notes are what make the summary worth reading. Without them, AI can only rephrase the CV, which the client already has, and it tends to smooth over gaps and upgrade vague achievements into confident ones. The judgement you formed on the call is what the client is paying for; the AI's job is to present it clearly and quickly.

Follow me on Instagram@sagnikteaches

What a hiring manager does with your summary in 90 seconds

A busy hiring manager reads the top, glances at the middle and decides whether to open the CV. Write for that. The format that holds up across most roles has five parts, in this order:

Connect on LinkedInSagnik Bhattacharya

1. Headline (two sentences). Who they are and why they fit this role, not their life story. "Store manager with six years in specialist retail, currently running a team of nine, looking for a larger store with more say over buying."
2. Must-haves, with evidence. Each of the client's four to six requirements, and the specific evidence against it from the CV or the call. Where there isn't evidence, say so.
3. What to probe. The gaps and doubts you'd want the client to explore at interview. Clients trust recruiters who name these.
4. Motivation and logistics. Why they'd move, notice period, salary as agreed with the candidate, location or working pattern constraints.
5. Your view (one or two sentences). In your own voice, clearly yours.

That's 250 to 350 words. Anything longer gets skimmed, and the probing section is the first casualty.

Subscribe on YouTube@codingliquids

Gathering inputs the AI can't invent

The quality of the summary is decided before you open the AI tool. Three inputs matter.

The brief, turned into must-haves

Client briefs are often a job description plus a phone call. Before the first summary for a role, write the must-haves down in a line each and confirm them with the client. For an illustrative sports equipment shop hiring a store manager, the list might be: runs a team of at least six; accountable for a store's sales and stock results; experience of a specialist or technical product range; comfortable with weekend trading; can train staff on product knowledge. Save it in the client's project in your AI tool (on business plans, both ChatGPT and Claude let a team share a Project) so every summary for that role uses the same yardstick.

Structured screening notes

Free-form notes produce free-form summaries. Take notes under fixed headings, even rough ones. A filled-in set from a 25-minute call:

Role: Store manager, [client], ref 118        Call: 25 min
Current: Assistant manager then manager, outdoor retailer, 6 yrs; manager 2.5 yrs
Team: 9 (4 FT, 5 PT). Hires and rotas herself. Lost 2 staff last yr, both to promotion.
Results: Store up ~12% last FY; she says "mostly footfall from a new retail park,
  partly our bike servicing push". Stock loss "below target" - no figure given.
Product: Trained team on bike fitting and tents. Runs quarterly product nights.
Why move: Wants buying input; current employer buys centrally. Not salary-driven.
Concerns: Never managed a store over 10 staff. Sunday trading fine, not late evenings.
Notice: 4 weeks.  Salary: agreed to share "looking for at least current + 10%".
My view: Strong on people and product. Numbers claims are modest and credible.

Notice what the notes capture that a CV never would: the candidate's own attribution of the sales growth, a missing figure, and a working-pattern limit. Those are the details that make a client trust you.

Transcripts, if you record calls

Some recruiters record screening calls and transcribe them with a tool such as Otter (Pro lists at $8.33 per user per month billed annually, or $16.99 monthly). Tell the candidate at the start and give them the option to decline. A transcript is useful for checking quotes, but it's a poor input on its own: it's long, rambling and full of small talk, so the AI weighs everything equally. Use it to verify your notes, not to replace them. The trade-offs between tools are covered in AI meeting note-takers compared.

The summary prompt

Keep this in the client's project or your shared prompt library, and paste in the three inputs each time:

You are helping a recruiter write a candidate summary for a client hiring manager.
Use ONLY the three inputs below. Do not add facts, figures, titles or dates
that are not in them.

Write these sections, 250-350 words in total:
1. HEADLINE: two sentences on who they are and why they fit this role.
2. MUST-HAVES: for each must-have in the brief, one line of specific evidence
   from the CV or notes. If there is none, write "Not evidenced - [what to ask]".
3. TO PROBE AT INTERVIEW: the gaps or doubts in the notes, stated neutrally.
4. MOTIVATION AND LOGISTICS: why they would move, notice, salary exactly as
   written in the notes, working-pattern limits.
5. RECRUITER'S VIEW: copy the "My view" line from the notes, lightly tidied.

Rules: plain English, no adjectives unless the evidence supports them
("strong", "proven", "exceptional" need a fact beside them). Keep the
candidate's own attribution of results. Never mention age, health, family,
nationality, religion or other personal characteristics, even if they appear
in the inputs.

BRIEF: [must-haves]
CV: [CV text]
NOTES: [screening notes]

A sample output, and what the recruiter fixes

Run against the notes above and the candidate's CV, an illustrative first draft of the must-haves section might come back as:

MUST-HAVES
- Team of 6+: Manages a team of 9 (4 full-time, 5 part-time), including hiring
  and rotas.
- Store results: Proven track record of driving growth, delivering a 12%
  sales increase last year.
- Specialist range: Trained staff in bike fitting and tents; runs quarterly
  product evenings.
- Weekend trading: Comfortable with weekend and evening trading.
- Staff training: Two team members promoted in the last year.

Three things need fixing, and they're typical:

• "Proven track record of driving growth" has dropped the candidate's own explanation that the rise was mostly footfall from a new retail park. The client will ask, and the candidate will tell them, and your summary will look inflated. Fix: "Store sales up about 12% last year; she credits mainly new footfall and partly a bike-servicing push she led."
• "Weekend and evening trading" is wrong. The notes say Sundays are fine and late evenings are not. Fix it, and move it to the probing section so the client can decide whether it matters.
• "Two team members promoted" is presented as evidence of training. The notes say two staff left for promotions elsewhere, which may say something good about her coaching but is a stretch as stated. Fix: "Trains staff on product knowledge; two of her team left for promotions last year."

The prompt told the AI not to add facts, and it mostly didn't. What it did was reframe facts in the most flattering way, which is harder to catch because nothing is strictly invented. That's why the check below compares meaning, not just facts.

With those fixes made and the other sections filled from the notes, the version that goes to the client reads like this (illustrative):

Store manager, ref 118

Headline: Store manager with six years at an outdoor retailer, two and a half of them as manager, running a team of nine. Wants a store where she has a say in buying, which her current employer handles centrally.

Must-haves:
Team of 6+: manages 9 (4 full-time, 5 part-time); does her own hiring and rotas.
Store results: sales up about 12% last year; she credits mainly new footfall from a new retail park, partly a bike-servicing push she led. Stock loss "below target"; no figure given.
Specialist range: trained her team on bike fitting and tents; runs quarterly product evenings.
Weekend trading: Sundays fine; see below on evenings.
Staff training: trains staff on product knowledge; two of her team left for promotions elsewhere last year.

To probe at interview: she hasn't managed more than 10 staff, so ask how she'd approach a larger team. Ask for a stock-loss figure. She's happy with Sundays but not late evenings; check that against your trading hours.

Motivation and logistics: moving for buying input, not salary. Four weeks' notice. Looking for at least her current salary plus 10%.

My view: strong on people and product. Her claims about results are modest and credible, which I'd take as a good sign.

Two things to notice. The probing section does real work for the client: it hands them three interview questions they'd otherwise have to find for themselves. And the salary line keeps the candidate's own terms, "at least current plus 10%", rather than a figure the AI worked out.

The five-minute claim check

Before any summary goes to a client, read it against the CV and your notes with these questions. It takes about five minutes once it's a habit:

• Titles and dates: is every job title exactly as on the CV? AI likes to promote "coordinator" to "manager".
• Numbers: is every figure in the source, with the same scope (the store, not "the region")?
• Attribution: where the candidate said "we" or "the team", does the summary say "she"?
• Logistics: are notice and salary stated exactly as the candidate agreed you could share them?
• Personal characteristics: is there anything about age, health, family or background, including indirect hints such as graduation years or "young team"?
• Your view: is the last section still yours, or has it become generic praise?

The first check exists because of mistakes like this one, illustrative but typical: a summary for an e-commerce homeware brand described a candidate as "Head of Operations at a fulfilment business". The CV said "Operations Coordinator". The candidate's line manager held the head-of role, and the AI had merged a sentence about reporting lines. The client's first interview question was about leading the function, and the recruiter's credibility took the hit, not the AI's.

Gaps in a CV are where the AI is most tempted to fill in. An illustrative case: the notes say "Career break 2022-2024; happy to discuss at interview", and the first draft reads "took two years out to care for a family member before returning to retail". Nothing in the inputs gave a reason. The model supplied the most common-sounding one, and in doing so put a family circumstance, true or not, in front of a client. The fix is to state only what the candidate agreed to share ("Career break 2022-2024; happy to discuss at interview") and to add a rule to the prompt: "For any gap, use the candidate's wording from the notes; never give or guess a reason."

Keeping bias and protected data out of the process

A summary can steer a client's decision as much as the CV does, so treat it with the same care you'd give a screening tool:

• Strip before you paste. Remove photos, dates of birth, full home addresses and anything about health or family from the CV text before it goes in. The prompt tells the AI to ignore these, but it can't ignore what it's been shown if the wording leaks into tone.
• Watch proxy language. Words like "energetic", "digital native", "mature" or "recent graduate" say more about age than ability. Ask for evidence-based wording only, and search drafts for these words.
• Be consistent across a shortlist. Run every candidate for a role through the same prompt and must-haves. Summaries written with different prompts are hard for a client to compare fairly.
• Know where the rules sit. If you place candidates with clients in the EU, the EU AI Act lists AI used in recruitment, including evaluating candidates, among its high-risk uses; those obligations are now scheduled for 2 December 2027. A summary you write from your own assessment, with AI drafting the words, is a different thing from a tool that scores or ranks candidates, but keep the decision visibly human and document your process. The screening side is covered in bias checks every recruiter should run.
• Leave emotion scores out. Some video-interview platforms rate candidates for "enthusiasm" or "confidence". For clients in the EU, the EU AI Act has banned AI that infers emotions in the workplace, recruitment included, since 2 February 2025, with narrow medical and safety exceptions. Wherever you work, a machine's reading of someone's face doesn't belong in a summary that carries your name; describe what the candidate said and did.

A second pass with the AI makes the proxy-language search quicker. Paste the finished summary and ask:

List every word or phrase in this candidate summary that suggests age,
health, family situation, nationality, religion or another personal
characteristic, directly or by implication. Quote each one and suggest
evidence-based wording. If there are none, say so.

An illustrative reply on a summary for a warehouse supervisor: "'Brings real energy to a young team' may suggest the age of the candidate and of the team; suggest 'motivates a team of 12 on rotating shifts'. 'Graduated in 2019' lets a reader estimate age; remove unless the qualification date matters for the role." Both flags are fair. The model will also flag harmless words now and then, such as "flexible" in a line about shift patterns; keep those when they describe what the candidate actually agreed to.

Adapting the format to three kinds of client reader

The same candidate needs a different emphasis depending on who reads the summary:

Add one line to the prompt for each client type ("The reader is the business owner; they care most about...") rather than keeping three separate prompts, so the structure stays the same across your desk.

For the store manager, only the headline and the order of the must-haves change. For the owner of a single-site shop: "Runs a nine-person store herself, including hiring, rotas and product training; wants more say over what the shop sells." For an in-house talent team: "Evidence against all five must-haves; stock-loss figure and late-evening availability to confirm at interview." The facts are identical. The first answers "can she run it without me?", the second "does she match the spec, and where's the risk?"

What this changes for a small agency's week

For an illustrative three-consultant agency placing retail and e-commerce staff, sending about 25 summaries a week: hand-written summaries took 35 to 45 minutes each, mostly spent re-reading the CV and turning notes into sentences. With structured notes, the prompt and the claim check, a summary takes 15 to 20 minutes, of which five are checking. That's roughly eight to ten hours a week back across the desk, most of it spent on calls rather than writing.

Two conditions make the difference between that and a pile of polished but unreliable summaries. The screening notes have to be taken under the headings, every time. And the check has to happen even when the client is pushing for CVs by lunchtime, because a single inflated summary costs more trust than a late one. If your ATS has AI drafting built in, the same format and rules apply; how to evaluate the AI features in your recruitment software helps you test whether it follows them. For the steps before and after this one, see AI CV screening for recruitment agencies and how much time AI saves per role.

Questions recruiters ask about AI-written summaries

Should I tell clients the summaries are AI-assisted?

If a client asks, say yes and explain the process: your notes and assessment, AI drafting, your check against the CV and call. Many agencies add a line to their terms of business covering AI-assisted drafting. What clients care about is that the assessment is yours and the facts are right, so make sure both are true before you worry about wording.

Do candidates need to agree before their CV goes into an AI tool?

Tell candidates in your privacy notice that you use AI tools to prepare their details for clients, and use a business plan that doesn't train on the content. Get their agreement before you send their details to any specific client, as you would anyway. If a candidate objects to AI processing, write their summary by hand and note it on their record.

Should the summary include salary expectations?

Include what the candidate has agreed you can share, stated exactly as they said it, such as a range or a minimum. Don't let the AI turn a range into a single figure or round it up. Many recruiters put salary and notice in a separate logistics line so the client reads fit first and cost second, which usually produces better conversations.

Further reads

• AI Screening: Bolt It Onto Your ATS or Switch Systems? — Whether to add AI to your current ATS or move systems.
• AI for Recruitment Business Development: Winning New Clients — Using the same tools to win the clients you write for.
• Questions to Ask a Recruitment Agency About Its AI Screening — The questions clients will increasingly ask you about AI.
• AI Bias in Small Business Decisions: Hiring, Pricing and Credit — How bias gets into AI-assisted decisions, with hiring examples.
• How to Build a Shared Prompt Library for Your Team — Keeping one tested summary prompt the whole desk uses.
• How to Use ChatGPT Projects to Keep Client Work Separate — One project per client so briefs and notes don't mix.
• How Staffing Agencies Automate Candidate Follow-Up With AI — The seven moments where candidates go quiet, which to automate first, message templates, AI reply sorting and the consent rules that keep texts welcome.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: EU AI Act Annex III (employment and recruitment uses) and the Digital Omnibus timetable; Otter.ai pricing page; OpenAI and Anthropic business plan data terms.