How to Humanize a Recommendation Letter After ChatGPT

Recommendation letters are judged on specificity and credibility—not on how many superlatives fit on letterhead. ChatGPT produces confident praise with no incidents: the candidate is "innovative," "a leader," and "an asset to any organization," with nothing a committee could verify.

Hiring managers and admissions readers have read ten thousand of these. They look for two moments you witnessed and an honest comparison to peers in that role. A humanizer that smooths template prose but leaves invented projects in place makes the letter worse, not safer.

You are the manager (or mentor) who was there. Lock candidate name, role, dates, and two incidents before any model help. Humanize body prose once after the facts are true.

Related: resume and cover letter hub, performance review feedback, informational interview email, ethical AI checklist.

What recommendation readers punish

Element	Weak AI draft	Strong manager draft
Relationship	"I have known them for some time"	"I managed Jordan on the Platform team, Jan 2023–Aug 2025"
Evidence	"Consistently exceeded expectations"	Named project, date, measurable outcome you saw
Ranking	"Top candidate I have ever met"	"Strongest analyst on a team of six in their cohort"
Voice	HR brochure	Sounds like a person who ran the 1:1s
Integrity	Skills the candidate never demonstrated	Matches resume and your calendar

Incident lock before any model

Before ChatGPT, write six lines by hand:

1. Candidate full name (as on the application)
2. Role and employer (exact title)
3. Dates of direct report or collaboration
4. Incident A: One situation, your role, outcome (metric or deliverable if true)
5. Incident B: Different skill dimension—communication, judgment, technical depth
6. Honest ranking: Where they sit among peers you managed in that window

If you cannot fill (4) and (5) from memory and notes, schedule a call with the candidate for facts—not for the model to invent stories.

Do not paste confidential performance data into a consumer chatbot if your employer forbids it. Work from your own bullet notes.

Typical letter structure

Adjust to the form, but most recommendations look like:

1. Opening — who you are, relationship, length of time
2. Role context — team, scope, what success looked like
3. Incident A — situation, action, result
4. Incident B — different strength or growth arc
5. Closing — ranking, willingness to be contacted, signature block

Section	Humanize?	Why
Name, title, dates, company	No	Verification
Incident facts and numbers	No	Integrity
Narrative around incidents	After lock	AI sounds like a template
Closing endorsement	Light, one pass	Often stiff

Paste body paragraphs into Human Writes once. Restore every name, date, and metric if they drift.

Before and after (incidents, not adjectives)

AI body paragraph:

Alex is an exceptional employee who consistently demonstrates leadership qualities and a strong work ethic. They would be an excellent addition to any graduate program or organization seeking a dedicated professional.

After incident lock:

When our checkout API failed during Black Friday 2024, Alex owned the rollback playbook without waiting for direction—restored service in 41 minutes and wrote the postmortem the exec team used in Q1 planning. On a team of four senior engineers, that response set the bar for incident ownership in their year with us.

Second version has a named event, behavior you saw, and context for the ranking.

Manager voice vs candidate voice

	Manager letter	Candidate self-draft
POV	"I managed…" "I observed…"	Sounds like a cover letter
Evidence	Calendar-backed incidents	Resume repetition
Risk	AI template praise	Inflated metrics
Fix	Incident lock first	Candidate supplies fact sheet; you write

If the candidate sends you a draft to "clean up," replace generic lines with your two incidents before any humanize pass.

Workflow

1. Confirm the deadline and submission portal (some forms limit length).
2. Fill the six-line incident lock from HR records and memory.
3. Ask the candidate for a one-page fact sheet—projects, dates, outcomes. Verify, do not copy blindly.
4. Draft opening and two incident paragraphs from the lock—not from "write a strong recommendation."
5. Add honest ranking language ("among the top two analysts I hired that year").
6. One Human Writes pass on stiff narrative prose.
7. Re-read as if you are the receiving committee: every claim defensible in a reference call?
8. Sign on letterhead or upload per instructions. Disclose AI editing if your organization requires it.

For peer references in academic settings, same incident discipline applies. See scholarship essays for the candidate side of the packet.

What not to do

• Sign a letter with achievements you cannot confirm.
• Let the model invent percentages or client names.
• Reuse one template for every report with swapped names.
• Humanize five times until dates or titles change.
• Treat humanizing as a way to hide that you did not know the candidate well.

Scores on detectors are a review signal, not the bar. A hollow superlative fails faster than a percentage. See best practices for humanizing.

Bottom line

Recommendation letters land when you name the role, cite two real incidents, and rank honestly. Human Writes is the voice pass after your fact sheet matches what you would say on a reference call.

Paste stiff body paragraphs on Human Writes when every incident could survive a follow-up from admissions or HR.