Marcus runs talent acquisition for a 220-person fintech in Austin, and his team of four recruiters sends roughly 90 candidate messages a week. The problem was never volume. It was that a passive-candidate InMail, a rejection after a final-round interview, and a counter-offer reply all got written with the same harried twenty-minute energy, and all three suffered for it. Outreach strategy in the abstract had never been his weak point; applying it under time pressure was. This lesson walks through Marcus working three realistic situations end to end with AI assistance, so you can see the full cycle in motion: research, insight, draft, review against a checklist, and iterate to something he would actually put his name on.

The Cycle You Are Practicing

Every scenario below uses the same five-step loop. Research the specifics of the person and the situation. Find the one genuine insight that connects the candidate to the role or the decision. Build a prompt that hands the AI that insight plus your constraints. Review the draft against a verification checklist. Edit or re-prompt until it passes. The discipline is in never skipping the review step, because that is where AI-assisted writing either earns or loses your trust. A draft that is eighty percent right is not a finished message when your name and your company's reputation are attached to it.

The three scenarios are deliberately different from each other: a passive candidate outreach, a rejection message after a real interview, and a counter-offer in response to an offer decline. They are the three situations recruiters most often handle badly, and they fail in different directions. Outreach fails by being generic. Rejection fails by being either cruel or so vague it is useless. Negotiation fails by being evasive about a constraint the recruiter has not actually decided on yet. Working all three with the same loop is what makes the loop transferable, because you learn where the loop stays constant and where the content of each step has to change.

Scenario One: Passive Candidate Outreach

Marcus is filling a senior backend engineer role and has identified Alex Chen, a staff engineer at Scale AI who has spent three years there leading their distributed-systems team and contributing regularly to etcd, an open-source consensus system. Alex's GitHub shows consistent, sustained work on infrastructure. Alex has also gone quiet on LinkedIn, which usually signals someone who is content where they are rather than someone scrolling job boards. That changes the entire tone of the message. You are not responding to interest; you are earning the right to interrupt. The task Marcus has set himself is to write outreach that demonstrates genuine research, respects Alex's current status, connects Alex's work to his own, and creates real interest without applying pressure.

Step one is the insight. Before anything else, Marcus asks what the genuine connection is between Alex's work and what his team needs. The answer is concrete: Alex's work on distributed systems and consensus, specifically etcd, maps directly onto the infrastructure scaling problems Marcus's platform team is wrestling with right now. That is the hook, and everything in the message hangs off it. Three years of tenure is a second, softer signal that Marcus notes without leaning on: long enough to suggest contentment, long enough that a genuinely interesting challenge might land. He does not treat that inference as fact, because assuming someone is restless is the fastest way to sound presumptuous.

Step two is the prompt. A prompt that says "write an outreach to a senior engineer" produces template sludge. The prompt that works names the person and role, states plainly what is known about their work, and then sets explicit requirements. Marcus tells the model that Alex is a staff engineer at Scale AI who led their distributed-systems team, is a significant contributor to etcd, has been at the company three years, and is technically focused on infrastructure and scalability. Then he lists what the message must do: reference the specific etcd work because his team faces similar scaling challenges, respect that Alex is likely happy at Scale AI, position the opportunity as interesting without overselling it, and remove pressure. He closes the prompt with a tone brief: peer to peer, genuine, respectful, four to five short paragraphs.

Step three is getting the draft, which is the fastest part and the least important. Step four is the review, and this is the step that earns the whole exercise. Marcus runs five checks before anything leaves his outbox. Accuracy: is the etcd work attributed correctly, and is his own company described truthfully? Authenticity: does it read like a real person wrote it, or like a mail merge? Tone: does it respect Alex's current role rather than implying they should be unhappy, and does any sentence feel pressuring? Company alignment: would his founder be comfortable with how the company is represented? Conversion likelihood: would a busy staff engineer actually want to reply to this?

Step five is editing and improving. Marcus notes what needs changing, then either edits directly or asks the AI to revise, and he iterates until the draft passes every check rather than most of them. On the first pass his draft over-flattered Alex and buried the etcd connection in the third paragraph, where a skimming reader would never reach it. He re-prompted to lead with the consensus connection and cut the flattery entirely. The second draft passed all five checks. That is the normal shape of this work: one substantive revision, not a heroic rewrite, and not a resigned send.

Scenario Two: The Rejection Message

The second scenario is a rejection, which is the message recruiters most often rush and most often get wrong. Marcus interviewed Sarah for a mid-level product manager role. She was genuinely strong on strategy and on thinking through problems, had relevant PM experience, and communicated well. Where she was weak was execution and the ability to move quickly, and the role requires someone who can ship fast with a small team and limited structure. The task is a rejection that is clear about the decision, respectful of her effort, specific in its feedback, and leaves the door open.

Step one is identifying the key elements, which for a rejection means answering three questions in order. What was good about Sarah? Strategy, communication, and relevant experience. Where was the gap? Speed of execution and comfort with ambiguity. Why does that gap matter? Because the role requires moving fast with a small team. Those three answers are the entire content of the message, and writing them down before drafting is what stops a rejection from collapsing into a vague statement about fit. "Fit" is not feedback. "We needed more evidence of shipping under ambiguity" is.

Step two is the prompt, which mirrors that structure. Marcus tells the model he is rejecting a candidate named Sarah for a mid-level PM role, lists her strengths as strategic thinking, excellent communication, and relevant PM experience, names the gaps as execution speed and comfort working independently with limited structure, and explains why it matters: the role requires shipping quickly in a scrappy environment. Then he sets six requirements. Be clear about the decision immediately. Respect her effort and preparation. Give honest feedback about the gap between execution and strategy. Acknowledge her strengths. Leave the door open for future opportunities. Keep the tone respectful and empathetic without being condescending. Three or four short paragraphs.

Step four, the review, is almost entirely about tone here. Marcus reads the draft against three questions. Does it feel condescending anywhere? Does it make Sarah feel she should improve, or does it make her feel she simply was not good enough, which are very different messages? Does it leave the door open genuinely, or dismissively, in the reflexive way that everyone recognizes as a formality? Step five is adjusting on those axes. He tunes the tone where it drifted, and he decides deliberately how specific the feedback should be, because specificity is a choice about how much you want to help this person rather than a fixed rule. If he genuinely wants to hear from Sarah again when a different role opens, he strengthens the open-door line so it reads as an actual invitation.

Specific feedback is also a fairness and legal-hygiene matter, not just kindness. Feedback should describe job-related behavior you observed in the process, never a protected characteristic and never speculation about the candidate's life. "We needed more evidence of shipping under ambiguity" is defensible and useful. Anything about age, family status, or accent is neither. The EEOC's guidance on adverse employment decisions is built on exactly this distinction: decisions and the explanations for them must rest on legitimate, job-related criteria applied consistently across candidates. Marcus keeps his rejection feedback in that lane on purpose, and he applies the same level of specificity to every rejected candidate for a given role rather than writing generously for the ones he liked.

Scenario Three: The Counter-Offer Negotiation

The third scenario tests judgment under a real constraint. Marcus extended an offer to Jordan for a senior engineer role. Jordan received a competing offer with significantly higher compensation, roughly twenty percent more, and wrote back honestly: "I'm torn. I love your company and team. But I have a family to support, and the other offer is too good to pass up. Is there any flexibility?" The task is a response that acknowledges the dilemma, reflects Marcus's actual position, and either opens the door to negotiation or handles the decline graciously if it cannot.

Step one is identifying your constraints, and it comes before any drafting. Can you match twenty percent? Partially? Not at all? That answer determines the entire approach, and a recruiter who starts drafting before deciding will produce a message that hedges, because it is written by someone who is dodging. In Marcus's case the working assumption is that his comp bands let him stretch to about ten percent, not twenty. His company's stated philosophy is that it pays fairly and competes, but is not the highest payer on the market. Jordan is strong and Marcus wants to keep them engaged whatever happens.

Step two is the prompt, which carries the constraint rather than hiding it. Marcus tells the model that a candidate has effectively declined over compensation, that the competing offer is twenty percent higher, that his side can stretch to ten percent but not twenty, and what the company's pay philosophy actually is. Then six requirements. Acknowledge the difficult position, including the family and financial responsibility, without judgment. Offer to negotiate up to the real limit of ten percent. If the gap cannot be matched, explain the philosophy respectfully. Emphasize non-monetary benefits where they honestly apply. Keep the door open for future opportunities. Do not sound bitter or resentful. Tone: honest, respectful, business-like, warm, three or four paragraphs.

Step four here is a review for fairness. Does the message acknowledge Jordan's constraint, a family to support, without a trace of judgment about it? Does the revised number feel justified rather than grudging? Does it keep Jordan engaged even in the likely case that ten percent is not enough to close a twenty percent gap? A response written from a place of "we are insulted that you are negotiating" reads instantly, and it costs you the rehire and the referral. Step five is adjusting to your genuine comfort level. If the draft feels too generous against your actual bands, pull it back. If it reads as too hard, soften it. The non-negotiable is that the final message represents your real offer and your real flexibility, because a message that overpromises to buy a day of goodwill creates a worse conversation later.

Adapting the Frameworks to Your Own Context

Scenario practice is where theory becomes skill. The frameworks are useful, but real recruiting has messy, specific situations that do not fit neatly into three categories, and the point of working these three is not to memorize them. It is to internalize the loop well enough that you can run it on a situation nobody wrote a template for. Your job is to learn the frameworks and then adapt them to your reality: different roles, different company cultures, different comp constraints, and different candidate populations all change the content of every step while leaving the sequence intact.

The adaptation is concrete rather than philosophical. If you recruit for a company that genuinely is the highest payer in its market, the counter-offer scenario inverts and your constraint is no longer money but headcount approval or level. If you hire high-volume roles where you reject forty people a week, the rejection scenario has to survive being run at that scale, which usually means a shared, job-related feedback vocabulary agreed in advance rather than a bespoke paragraph each time. If your candidates are not engineers, the research signals in scenario one change completely, since a designer's portfolio and a clinician's publication record carry the hook that a GitHub profile carries for Alex.

One more piece of sequencing advice: practice on low-stakes scenarios first. Run the loop on a role you are not desperate to fill and a candidate whose reply you can afford to lose, so that the first time you use it under real pressure the mechanics are already familiar. Building confidence on the messages that matter most is an expensive way to learn.

Three Ways This Goes Wrong

Not adapting to the scenario. The first failure is treating all three situations as one, using the same tone and the same structure for a passive outreach and a rejection. It fails because these are genuinely different conversations. Outreach is about attraction, rejection is about respect, and negotiation is about honesty under constraint. Copying the register of one into another produces messages that feel off in ways candidates notice even if they cannot name why. The defense is to identify what makes each scenario unique before you prompt, and to customize both the prompt and the tone brief accordingly rather than reusing last week's.

Insufficient review. The second failure is getting an AI draft that looks decent and sending it. It happens because the draft is plausible and you are busy, and it fails because you will not catch the misattributed fact, the tone problem, or the sentence that misrepresents your company. The verification checklist takes two or three minutes to run in full, which is worth it every single time, and it routinely catches exactly the kind of error that is expensive in public. Always run the full checklist rather than the two checks you happen to remember.

Not iterating. The third failure is accepting the first draft as final. AI generates something, you review it, it is about eighty percent there, and you send it anyway. Eighty percent is not good enough for messages that carry your name. One round of editing or a single re-prompt usually moves a draft to ninety percent or better, and a second pass gets you to the ninety-five percent that reads as genuinely yours. Commit to at least one round of revision as a standing rule, so that iterating is the default rather than a thing you do when you have time.

Practice

Work these in order. The first three rehearse the loop on the worked scenarios; the last two move it onto your own desk.

• Work scenario one completely. Using the passive candidate framework, run the Alex Chen outreach end to end: research, insight, prompt, draft, review against the five checks, edit, final version. Write down which check failed on your first draft.
• Work scenario two. Draft the rejection message for Sarah. Run the verification checklist on it. Does it pass? Where specifically would you improve it, and is your feedback job-related enough that you would be comfortable if she quoted it back to you?
• Work scenario three. Draft Jordan's counter-offer response. Decide first what you would actually offer and why, then check whether the message reflects that decision honestly rather than hedging around it.
• Adapt to your context. Take each of the three scenarios and rewrite them for roles you are genuinely hiring for. What changes in the research, the constraints, and the tone, and what stays the same?
• Create your own scenarios. Identify three recent recruiting situations from your own work, ideally ones you handled imperfectly. Draft responses using the frameworks here, compare them to what you actually sent, and name the specific place where you need to improve.

Reflection

• Which of the three scenarios, passive outreach, rejection, or negotiation, do you find most challenging, and why? What would getting better at it actually require?
• When you have handled similar scenarios in the past, what worked well and what did not? What would you do differently now that you have the loop?
• How much time do you currently spend on these three kinds of communication in a normal week? How much would a repeatable loop save, and where would you spend the recovered time?
• What is one scenario you are facing right now that resembles one of these three? How would you adapt the framework to it?
• Which step of the loop do you skip most often under time pressure, and what would make it harder to skip?

Glossary

• Scenario practice. Working through realistic recruiting situations using the techniques and frameworks you have learned, rather than studying them in the abstract.
• Verification checklist. The five tests you apply to every piece of outreach: accuracy, authenticity, tone, company alignment, and conversion likelihood.
• Iteration cycle. Review, identify gaps, edit or revise, review again. Most pieces of outreach need one or two iterations to reach final-version quality.
• Passive outreach. Reaching out to someone who is not actively looking for a job. Requires stronger research and explicit respect for their current status.
• Negotiation response. A message replying to a request for better compensation or terms. Requires honesty about your constraints and respect for the candidate's needs.
• Genuine insight. The specific, researched connection between a candidate's actual work and your open problem, which is what makes an outreach message land rather than read as a template.

Related Lessons

This lesson is the applied session for a chapter you have been building toward, and each scenario has a companion lesson that goes deeper on its mechanics.

• Passive Candidate Engagement: Drafting with AI supplies the research depth and multi-touch sequencing behind scenario one, including how long to spend per candidate and which timing signals justify reaching out now.
• Rejection Messages and Declining Offers: Tone, Clarity, and Care extends scenario two with more worked language for hard conversations and for declines you did not choose.
• Editing and Verifying AI-Drafted Messages is the discipline behind step four, covering what to check and how to catch the errors AI reliably makes.
• Prompt Anatomy: Structure, Context, and Constraints explains why the prompts in these scenarios are shaped the way they are, with context first and constraints made explicit.
• Iterating with AI: Following Up, Clarifying, and Refining covers step five in depth, including how to ask for a revision that actually changes the thing you wanted changed.
• Scenario Practice: Review and Critique is the natural next session, where you critique drafts rather than produce them.

Closing

You now have a loop that covers every major outreach scenario you are likely to face: research and insight, then a prompt that carries both, then a draft, then a review against explicit checks, then at least one round of improvement. The loop is the same every time. What changes is the content of each step, which is why practicing it on three deliberately different situations is more useful than practicing it three times on the same one.

Practice them, adapt them, and make them your own. This week, identify three real recruiting scenarios you are currently facing and run the full loop on each. The frameworks are only worth anything once they are wearing your context, and the fastest way to get there is to stop reading about the cycle and run it on a message you actually need to send today. From here the chapter turns from crafting outreach to analyzing and summarizing candidate information, where the same insistence on reviewing what the AI produced does even more work.

Key Takeaways

• Run the full cycle every time. Research, insight, prompt, draft, review, edit. The review step is where AI-assisted writing earns or loses your trust, so it is the one you can never skip.
• Different scenarios demand different approaches. Passive outreach, rejections, and negotiations each have their own structure and tone. Outreach attracts, rejection respects, negotiation is honest under constraint. Reusing one voice across all three produces messages candidates can feel are wrong.
• Lead with the genuine insight. For passive outreach especially, the specific connection between the candidate's work and your open problem is the hook. For Alex Chen that was consensus and distributed systems mapping onto a live scaling challenge, not flattery or generic enthusiasm.
• Always run the verification checklist. Accuracy, authenticity, tone, company alignment, and conversion likelihood. It takes two or three minutes and it catches errors and misrepresentations that are expensive once sent.
• Keep rejection feedback job-related and specific. Name what was strong, name the single decisive gap, and explain why it matters for this role. Describe observed, role-relevant behavior, never a protected characteristic or speculation, consistent with EEOC expectations that decisions rest on legitimate criteria applied consistently.
• Decide your real constraint before you negotiate. Work out whether you can match fully, partially, or not at all before drafting a word. Knowing your actual limit keeps the message honest and warm rather than evasive or resentful, and the final message must represent your real offer.
• Iterate past eighty percent. First drafts are rarely final-version ready. One edit or re-prompt usually carries a draft from acceptable to genuinely yours, and messages with your name on them deserve that last pass.
• Adapt the frameworks, and practice low-stakes first. Different roles, cultures, and constraints require different messaging, and building confidence on a scenario you can afford to lose is cheaper than learning on the one you cannot.

Frequently Asked Questions

How many iterations should I expect before a draft is ready? Usually one substantive revision, occasionally two. First drafts tend to land around eighty percent: structurally fine, but over-flattering, generically enthusiastic, or with the important detail buried in the middle. One round of editing or a re-prompt typically gets you to ninety percent, and a second focused pass gets to the ninety-five percent that sounds like you wrote it. If you are on your fourth revision, the problem is almost always the prompt rather than the draft, and the faster fix is to rewrite the prompt with better context than to keep patching output.

Is it dishonest to use AI to write a personal message like a rejection? Not if the judgment is yours. The AI is not deciding that Sarah was strong on strategy and weak on execution speed, and it is not deciding whether the door stays open. You are. What the model does is turn decisions you have already made into clean prose faster than you would write it, which is why the loop puts research and insight before prompting and puts your edit after the draft. The dishonest version is sending a message whose content you have not actually verified or would not stand behind, and that failure is available with or without AI.

How specific should rejection feedback be? It is a deliberate choice rather than a fixed rule, and it scales with how much you want to help the candidate and how confident you are in the observation. Be more specific when the gap was clear, job-related, and something the person could act on. Be less specific when your read is thin or when the decision came down to a comparison rather than a deficiency. What never changes is the boundary: describe job-related behavior you observed in the process, never a protected characteristic and never speculation about the candidate's circumstances, and apply the same standard to every candidate you reject for that role.

What if I cannot move at all on the counter-offer? Then the message becomes a graceful decline rather than a negotiation, and it still deserves the same care. Acknowledge the candidate's position without judgment, be straight that you cannot move rather than implying you might, explain the compensation philosophy respectfully so the number reads as a policy rather than an insult, mention genuine non-monetary value only where it honestly applies, and keep the door open for future roles. The one thing to avoid is any note of bitterness that the candidate negotiated. That is what costs you the rehire and the referral later.

The verification checklist feels like overhead on a high-volume week. Can I shorten it? Run it in full, because it takes two or three minutes and the checks you would drop are the ones that catch the expensive errors. Accuracy and company alignment are the two most often skipped and the two most damaging when they fail, since a misattributed project or an overstated claim about your company is public and permanent. If volume is the real constraint, reduce it upstream by sending fewer, better-researched messages rather than by reviewing more messages less carefully.

My drafts keep coming back sounding like a sales email. What am I doing wrong? Almost always the prompt is missing either the genuine insight or the explicit tone brief. If you do not hand the model a specific, researched connection, it has nothing to write about except enthusiasm, and enthusiasm is what sales email is made of. Give it the specific work you are referencing, state the real reason that work connects to your role, and then constrain the voice directly: peer to peer, respectful, no overselling, no pressure, four to five short paragraphs. Then cut whatever flattery survives, because that is the line the model reaches for when it runs out of substance.