Why You Shouldn't Use Claude to Ghostwrite Professional Reference Letters (And the Right Way to Leverage LLMs for Endorsements)

We have all been there. A former junior engineer or product manager you genuinely enjoyed working with drops you a message. They are applying for a dream role and need a reference letter by Friday. You want to support them, but your calendar is absolutely swamped.

Then you look at the blinking cursor of a chat interface on /platforms/claude. You think: I can just feed their CV and the job description into the model, tell it to sound like me, and have a polished letter in twelve seconds.

It is incredibly tempting. But fully delegating this task to a Large Language Model (LLM) is an ethical misstep that undermines the entire purpose of a professional reference. Let’s talk about why you should step away from the "generate" button, and how you can actually use these models to assist your writing without losing your professional integrity.

The Professional Reference as a Sacred Trust

A reference letter is not just a formal bureaucratic box to tick. It is an act of social capital transfer. When you sign your name to an endorsement, you are putting your personal and professional reputation on the line to vouch for another human being.

Recruiters and hiring managers read dozens of these letters. They do not read them to see if the candidate can find someone who knows how to operate an LLM; they read them to find the subtle, authentic nuances that define a candidate's work ethic and character. They want to know what makes a candidate tick, and only a human who has shared a high-pressure environment with them can truly speak to that.

When you use an AI to write the letter from scratch, you strip away the authentic, idiosyncratic voice that makes a reference believable. You trade your genuine personal perspective for a highly polished, sanitised stream of corporate buzzwords. It is a disservice to the candidate, a deception to the hiring manager, and a cheapening of your own professional brand.

Why Fully Generated Reference Letters Fail the Smell Test

LLMs are trained to output the most statistically probable next token. In the context of a reference letter, this means they default to the most generic, cliché-ridden version of a "great employee" possible.

An LLM-generated reference is easy to spot. It will invariably claim the candidate is a "proactive self-starter with exceptional collaborative synergy and a deep commitment to excellence." It reads like a performance review written by a committee of HR robots.

More importantly, fully automated letters lack the power of specific, messy human memories. An LLM cannot remember the rainy Tuesday afternoon when your production database collapsed, and how that specific junior engineer stayed calm, found the rogue query, and kept the team laughing through the stress. It is those specific, unpolished anecdotes that actually get people hired. If you want to know how Claude handles text synthesis in detail, you can explore our guides on /platforms/claude/articles, but rest assured, it cannot manufacture real human memories.

The Authentic Alternative: How to Use LLMs as an Editorial Assistant (Not a Proxy)

You do not have to write every single word from scratch if you are struggling with blank-page syndrome. There is an ethical, highly effective middle ground: using the LLM as an editor and structural guide rather than a ghostwriter.

Instead of asking the model to write the letter for you, write a quick, messy, unformatted brain dump of your actual thoughts first. This ensures the raw data—the memories, the specific skills, and the personal voice—comes entirely from you.

Here is a practical workflow to do this ethically:

1. Write a messy bulleted list: Jot down three things the candidate did exceptionally well. Use your own casual language. Mention specific projects, even if you do not remember the exact dates or formal titles.
2. Define your relationship naturally: State exactly how long you worked together and in what capacity.
3. Use the LLM to format and polish: Pass your raw, unedited bullets to the model and ask it to organise them into a standard, professional letter format. Use a tight system prompt to prevent the model from adding synthetic praise or hallucinating achievements.

A Step-by-Step Ethical Workflow for Recommendation Drafts

To make sure the model does not run wild with corporate fluff, you need to constrain its output strictly. If you want to dive deeper into structuring these types of strict instructions, check out our resource on /glossary for prompt engineering terms.

Here is an example of an ethical prompt you can use to clean up your raw notes:

`text I am writing a professional reference letter for a former colleague. Below, I have provided my raw, unformatted notes about their performance and my genuine experiences working with them.

Your task is to organize these notes into a clean, professional, three-paragraph reference letter.

CRITICAL CONSTRAINTS: 1. Do not add any achievements, skills, or projects that are not explicitly mentioned in my notes. 2. Do not use generic corporate clichés (e.g., "visionary leader", "synergistic team player"). Keep the tone warm but professional, direct, and factual. 3. Preserve my specific anecdotes exactly as described, simply cleaning up the grammar.

Here are my raw notes: [Insert your bullet points here] `

By framing the interaction this way, you remain the sole author of the substance. The AI is acting merely as a sophisticated grammar check and layout engine.

Protecting Your Reputation (and Your Candidate's Career)

Before you hit send on any reference letter, read it aloud. If you find yourself reading a sentence you would never actually say in real life—such as "Their unmatched dedication to paradigm-shifting solutions was exemplary"—delete it immediately.

If a hiring manager decides to call you to follow up on the letter (which good hiring managers frequently do), you need to be able to speak to the exact points in that letter without stuttering or trying to remember what your AI wrote. If there is a massive mismatch between the highly formal, polished letter and your actual conversational style over the phone, it raises immediate red flags.

Keep your recommendations honest, keep them specific, and keep them human. Your former colleagues deserve nothing less.