AI is fast. ERAS deadlines are brutal. Faculty are busy. So yes—using AI to help draft a personal statement or organize material for a letter of recommendation is tempting.
That’s exactly why people get burned.
I’ve seen applicants lose credibility over details they never meant to claim. A polished paragraph quietly invents a publication status. A “helpful” rewrite turns a sincere story into generic mush. A letter draft starts sounding like a robot pretending to be an attending. Bad look. Worse timing.
These documents are not casual writing exercises. They are high-stakes professional representations. If your ERAS statement or LoR contains false, biased, exaggerated, inconsistent, or privacy-breaking content, you can damage interviews, relationships, and trust.
So I’m going to do this the way I always do it: protect you from the dumb mistakes.
Here are the 7 mistakes to avoid when using AI to draft ERAS materials or support LoRs—plus how to catch the red flags before they cost you.
Mistake #1: Letting AI Fabricate—“Helpful” Details That Aren’t True
This is the most dangerous mistake because it looks polished.
AI will confidently generate details you never gave it. Dates. Duties. Author order. Publication counts. Awards. Leadership roles. It doesn’t always “lie” maliciously; it fills gaps. That’s the problem. In an application, gap-filling is not help. It’s risk.
I’ve seen versions of this go bad in very predictable ways:
- A submitted manuscript becomes “published”
- A poster presentation becomes “oral presentation”
- “Participated in QI efforts” becomes “led a hospital-wide initiative”
- A sub-internship timeline gets shifted by a month
- Co-author order is rewritten
- A decent rotation evaluation gets inflated into “ranked among the top students”
Don’t make this mistake: never treat AI output as factual just because it sounds specific.
Specificity is not proof. In fact, in AI-generated text, extreme specificity should make you more suspicious, not less.
Red flags that should stop you immediately
- Claims that don’t match your CV or ERAS entries
- Exact numbers with no source
- Publication details you didn’t provide
- Rotations, dates, or roles stated more confidently than your records support
- Language like “top 5%,” “best student,” or “first author” unless documented
What to do instead
Run a brutal fact check. Line by line.
Use this verification checklist:
- CV match: Every role, title, date, and institution must match your master CV.
- ERAS match: Activities, publications, and experiences must be internally consistent.
- Transcript/rotation match: Confirm clerkships, electives, sub-Is, and dates.
- Publication match: Check PubMed, acceptance emails, PDFs, and author order.
- Employment/HR match: Verify job titles and timeline.
- Reference match: If a recommender mentions something, make sure they actually observed it or have documentation.
If a sentence contains a fact, it needs a source. No exceptions.
Mistake #2: Using AI to Write Your Voice Instead of Enhancing It
Your personal statement should sound like you. Not like “an applicant.” You know the voice I mean—overwrought, glossy, emotionally vague, stuffed with words like passionate, deeply meaningful, and lifelong commitment. That style is dead on arrival.
Programs read tons of these. Generic writing is not neutral. It is actively forgettable.
The trap is simple: AI can produce clean prose, but clean prose is not the same thing as authentic prose. If your application voice doesn’t match your actual training story, interviewers feel the mismatch. Recommenders feel it too.
Don’t make this mistake: submitting language that could belong to any applicant in any specialty.
Red flags of AI-flattened voice
- Repetitive phrasing
- Inflated adjectives with thin substance
- Vague motivations like “I always wanted to help people”
- Sudden emotional claims not grounded in experience
- Unnatural transitions between stories and career goals
- Polished but empty paragraphs
A strong statement has fingerprints on it. Specific moments. Clear observations. Real stakes. Not just “I learned the importance of teamwork.” From what? During what case? What changed in how you think?
How to make AI useful without losing yourself
Feed it real material, not broad prompts.
Better inputs:
- A specific patient encounter you can ethically describe in de-identified form
- One hard lesson from a rotation
- A moment of failure or uncertainty and what changed
- A measurable achievement tied to actual work
- Why this specialty fits your temperament, not just your résumé
Then edit manually. Aggressively.
Ask:
- Does this sound like something I would actually say?
- Is this story specific enough to be memorable?
- Did AI smooth out the tension and make it bland?
- Would a mentor who knows me recognize me in this piece?
Your job is not to sound impressive. Your job is to sound real, sharp, and consistent.
Mistake #3: Not Controlling for Bias, Equity, and Unprofessional Framing
This one catches people because the language often sounds polite.
Polite is not the same as safe.
AI can produce wording that is subtly paternalistic, stereotyped, patient-blaming, or just professionally off. Sometimes it frames communities in deficit-based ways. Sometimes it assigns traits based on identity. Sometimes it overexplains social issues in a way that sounds performative and clumsy. None of that belongs in ERAS or LoRs.
Don’t make this mistake: assuming that because output sounds formal, it’s ethically clean.
Watch for these red flags
- Comments about protected characteristics unless clearly relevant and appropriately framed
- Language implying patients are “noncompliant” without context
- Paternalistic phrasing about underserved communities
- Stereotyping of cultural, racial, disability, gender, or socioeconomic groups
- “Color commentary” that makes you sound politically careless rather than thoughtful
- Personality labels that feel coded or loaded
You need a bias review pass. Every time.
Use patient-centered and inclusive language. If something feels even slightly off, rewrite it. Better yet, have a mentor, advisor, or staff member with good judgment review it for professionalism.
High-risk phrasing doesn’t just look sloppy. It can make you look immature, biased, or unsafe. That’s a heavy price for a paragraph you didn’t even fully write.
Mistake #4: Failing to Verify Accuracy—Dates, Metrics, and “Sourceable” Claims
This overlaps with fabrication, but it deserves its own warning because a lot of damage happens in the small stuff.
Applications fall apart on inconsistencies.
Maybe your personal statement says you spent “two years” on a research project, but your CV timeline shows fourteen months. Maybe a draft says you “authored three papers,” but one is only submitted and another is an abstract. Maybe a letter mentions you were “among the top students,” but that ranking was never actually given.
These are not tiny errors. They raise the question no applicant wants raised: What else here is unreliable?
Don’t make this mistake: leaving numbers, dates, and rank-like claims unverified.
Facts that need proof
- Dates of rotations, jobs, and research
- Publication status
- Number of presentations, posters, or papers
- Awards and honors
- Rank descriptors like “top,” “best,” “highest,” or percentile claims
- Program names, department names, and formal titles
- Scope of responsibilities
Here’s the workflow I recommend, and yes, it’s tedious. Do it anyway.
- Generate the draft
- Compare it line by line against your CV, ERAS entries, transcript, and source documents
- Highlight every factual claim
- Mark each one as verified, unsupported, or wrong
- Replace anything unsupported
- Check chronology
- Do a final consistency scan across all documents
That process saves people. I mean that literally.
If a statement can be sourced, source it. If it can’t, soften it or remove it. Clean and accurate beats flashy and fragile every time.
Mistake #5: Misusing AI for Letters—Breaking the Recommender Relationship
This is where people get ethically sloppy.
A letter of recommendation is not just another document. It is supposed to reflect the recommender’s judgment, observations, and voice. AI can help organize themes or convert your brag sheet into a cleaner packet. Fine. But if you or your recommender are using AI to ghostwrite a letter that pretends to be deeply personal when it isn’t, you’re in dangerous territory.
Don’t make this mistake: treating an AI-generated letter as if it were the recommender’s authored assessment.
That can damage trust. It may conflict with institutional expectations. And honestly, it often reads weirdly—too polished, too broad, oddly authoritative, full of claims no human observer would make.
Red flags in AI-tainted LoRs
- Multiple letters using nearly identical wording
- Sweeping praise unsupported by examples
- Statements about your performance the recommender didn’t personally witness
- A voice that sounds nothing like the faculty member
- Generic superlatives stacked one after another
The better approach is simple.
Give your recommender a support packet, not a fake letter.
Include:
- Your updated CV
- Personal statement draft
- Specialty goals
- Rotation dates
- Specific cases, projects, or presentations they observed
- Concrete accomplishments
- A short bullet list of themes they might emphasize
AI can help you organize that packet. That’s appropriate. Then the recommender should write in their own words, using their own judgment.
That preserves what actually matters: credibility. A strong honest letter beats an elegant fake every single time.
Mistake #6: Ignoring Privacy, Security, and Institutional/Platform Rules
This mistake is reckless. Don’t do it.
If you paste patient identifiers, confidential evaluations, personal identifiers, or sensitive institutional material into an AI tool without proper safeguards, you’re creating a privacy problem. Maybe a policy problem too.
You do not need PHI to draft a compelling statement. You do not need names, MRNs, exact dates, or identifying clinical details. And you definitely do not need to upload confidential documents blindly because a chatbot asked for “more context.”
Never include
- Patient names or initials
- MRNs
- Exact dates tied to identifiable events
- Specific locations that make a patient obvious
- SSN, home address, personal ID numbers
- Confidential contract or HR details
- Internal incident reports or sensitive evaluations
Safer alternatives
- De-identified summaries
- General clinical themes
- Non-identifying outcomes
- Your own reflections on what you learned
- Broad timelines when exact dates aren’t needed
- Sanitized bullet points from experiences
Also check your school’s guidance and the platform’s terms of use. Don’t assume every AI tool handles data the same way. Some do not belong anywhere near sensitive academic or clinical material.
If you’re ever unsure whether something is too identifying, assume it is and strip it out.
Mistake #7: Skipping the Final Human Review—No Proof, No Polish, No Alignment
This is the last stupid mistake, and it’s common because people are tired.
They generate a draft, tweak two sentences, feel relief, and hit send.
No. Not yet.
AI output is not final copy. Ever. It needs a human review for grammar, consistency, ethics, and plain old common sense. I’ve seen drafts with the wrong program name, mixed tenses, repeated lines, and impossible jumps in achievement level—all because nobody did one clean final read.
Final review red flags
- Wrong institution or program name
- Date mismatches
- Broken tense
- Repeated ideas or duplicate sentences
- Abrupt changes in tone
- Claims that are stronger than the evidence
- Language that sounds unlike you or unlike the recommender
Your final checklist
Before anything gets submitted or sent to a recommender, confirm:
- Accuracy
- Authentic voice
- Professional tone
- Inclusive language
- Privacy protection
- Cross-document consistency
- Proofreading completed
- Mentor or trusted reviewer checked it
Read it out loud. That catches more nonsense than people realize. If a sentence makes you wince when spoken, it probably needs work.
Use AI Safely—Draft Faster, But Verify Like a Clinician
AI can absolutely help you draft faster. Good. Use the efficiency. But don’t hand over judgment. That’s the mistake behind all seven problems.
Here’s the rule I want you to keep: AI is a drafting assistant, not a verifier, and not the true author of a recommender’s letter.
Build a “verify before submit” routine:
- Check every fact
- Restore your real voice
- Review for bias
- Protect privacy
- Keep LoRs ethically grounded
- Get a final human read
Do that, and AI becomes useful instead of dangerous.
Skip it, and you’re gambling with your credibility. That’s a bad bet.
