Use AI without losing your voice

Writing tools can be helpful when you are looking for a clearer sentence or working in another language. The important question is whether the message still represents your understanding, experience, and intentions. A polished email is useful when its content is accurate and specific enough for a research conversation. Smooth wording cannot supply a connection you have not yet worked out.

My advice is to do the reading and draft the research connection yourself before asking a tool to help with presentation. You do not need to make the writing sound less polished to prove it is yours. You need to be able to explain what you mean and stand behind the claims the email makes.

Understand why a generic draft falls short

When a tool is given a CV and a faculty biography, it can produce a plausible introduction, an enthusiastic sentence about the research, and a request for supervision. But the finished paragraph may say very little about what you understood. It might repeat a topic from the biography and describe a broad “alignment” without identifying a question, a method, or a finding. A human-written template can have exactly the same weakness.

Read the research paragraph on its own. Could it be sent to a different professor after replacing the name and a keyword? If so, the next step is more thinking and reading. Explain what interested you in the work, why that matters for the question you hope to explore, and what preparation you bring. The preparation article describes how to build that connection from actual sources.

Do not ask a generator to invent a thesis pitch simply because it appears to match a lab. An impressive-sounding proposal can leave you committed to a topic you do not understand or want to study. Your initial direction can be modest and provisional. It is more useful to identify a question you can discuss honestly than to promise a finished research agenda you could not explain in a meeting.

Keep the thinking in the first draft

Write a few sentences without trying to make them elegant. Describe which program you are considering, what you read, what you noticed, and what you want to ask. Add one relevant experience and explain your role in it. If you get stuck at the point where the research should connect to your background, leave a note to yourself about what you need to learn. That gap is useful information about your preparation.

Then revise for clarity. A trusted reader can tell you where a sentence is confusing or where an idea needs evidence. Where the applicable rules permit it, a writing tool can help identify repeated phrases, spelling errors, or overly long sentences. Review each change yourself. An edit that changes “I helped analyse the data” to “I led the analysis” has changed a factual claim, even if the second sentence sounds more confident.

For permitted editing support, a limited request is more useful than “write me a winning supervisor email.” Ask the tool to identify unclear sentences in your draft and explain why they are unclear, without adding experiences, research claims, or praise. Make the final choices yourself. This keeps the revision focused on communicating ideas you already understand.

Revise the connection before the adjectives

Consider this invented sentence: “Your groundbreaking research perfectly aligns with my passion for artificial intelligence.” It conveys enthusiasm, but a supervisor cannot yet tell which work you read or what you would like to investigate. Replacing “groundbreaking” with another adjective would not answer those questions.

A more useful fictional revision might say: “I was interested in the difference your study found between reported trust and the checks participants performed. In my course project, I compared errors in AI-assisted reports, and I would like to learn how teams decide what to verify.” This refers to the invented study in the writing example, not to a real publication. The revision provides an observation, some relevant experience, and a direction for learning.

The improvement comes from the substance. For your own message, replace the example with something you actually read and did. If you only understood part of the work, keep the claim within that understanding. You do not need to imitate the vocabulary of a research paper to explain why it interested you.

Check every research claim and factual detail

Open each paper or project page you mention. Verify the title, authors, date, and the finding you attribute to it. Check whether you are discussing published work, a preprint, or an announced project. A generated description may combine ideas from several sources or confidently assign a finding to the wrong study. Use the original source to establish what it actually says.

Check your own details with the same care. Degree names, dates, project roles, skills, and award status should agree with your CV. An application for a scholarship is different from an awarded scholarship; an ongoing project is different from a published paper. Make the distinction clear so that a potential supervisor can assess the information you intend to provide.

Be thoughtful about what you share with an external writing service. You can ask for feedback on a short excerpt without uploading identity documents, confidential research, an entire email thread, or someone else’s unpublished work. Use the services and settings your institution permits, and remove personal details that are not needed for the editing task.

Read the rules for the material you are preparing

An informal inquiry, a research proposal, a personal statement, and an interview can have different requirements. Do not assume that a tool use permitted for one task is permitted for every part of an application. Check the program’s current guidance and any authorship or disclosure requirements before submitting.

For example, Oxford’s graduate-application guidance permits formatting and spelling or grammar support, while requiring applicants’ own thinking and expression. It distinguishes initial research assistance from having AI generate application ideas or write the materials, and specifies disclosure for AI-assisted initial research. Those are Oxford’s rules for its graduate applications; they are not a universal policy governing every supervisor email. Consult the instructions that apply to your situation.

Do not turn style into a verdict about a person

A courteous greeting, a common subject line, or careful English does not establish that someone used AI. Students learn conventions from examples and may receive help from teachers, friends, translators, or accessibility tools. Evaluate whether the message is accurate and meaningful rather than treating familiar wording as evidence of dishonesty.

Research also gives reasons to be cautious about automated judgments. Liang and colleagues’ 2023 study of GPT detectors documented errors and bias against non-native English writing in the tools and samples they tested. That study does not evaluate a particular applicant’s email or identify the software used to write it. A respectful response to a vague message is to explain what information would help, rather than making an unsupported accusation.

For an applicant, the practical aim is straightforward: keep the research connection yours, use support in ways the relevant rules allow, and send a message you could explain in a conversation. You can be clear without sounding grand, and you can ask for help without surrendering responsibility for what you say.

Linked guidance and research were reviewed on 13 October 2026. Institutional AI rules vary and may change. The revisions above are fictional teaching examples; no individual student’s correspondence is reproduced or classified.