College Counselling Business: Can You Use AI on Student Essays?

Yes, for feedback, no for authorship. A college counselling business can use AI to ask questions, spot gaps, flag clarity problems and check mechanics on an essay the student wrote. Help becomes ghost-writing when the tool supplies the ideas, structure or sentences that end up in the submission. Put the line in writing and tell families.

The trouble is that there is a grey zone between those two, and it is where most of the risk sits. An application platform may treat "substantive" AI-generated content as fraud without defining how much that means. A parent may assume you are using AI because the essay improved overnight. A student may paste your AI-polished paragraph into the portal and not realise it counts. A written routine keeps the work honest, defensible and still quicker than doing everything by hand.

Follow me on Instagram@sagnikteaches

The spectrum from coaching to ghost-writing

Rather than a single yes or no, think of six rungs. Where each task sits decides whether you do it, and how you record it.

Connect on LinkedInSagnik Bhattacharya

Most reputable counsellors operate between rungs 0 and 3. Rung 4 is where well-meaning help turns into authorship, because a rewritten sentence is the AI's sentence even if the student pastes it in with a change of one word. The cleanest rule is: the AI may say what is wrong; the student must say it better.

Subscribe on YouTube@sagnikteaches

What the platforms and schools say, and how to keep up

Application platforms and universities each set their own rules on AI. The Common App's fraud policy (last updated August 2023) lists among its examples of fraud intentionally misrepresenting as one's own original work "the substantive content or output of an artificial intelligence platform, technology, or algorithm". It does not define how much help makes content substantive. Two details matter for a counselling business. The policy's registered users include counsellors, advisers and other professionals who work with applicants, so it is not only the student's problem. And if the platform finds fraud, it may share the applicant's name and its findings with the colleges on that applicant's list. Read the live policy before you advise a student, because wording can change, and treat my summary as a pointer, not a rule.

Individual universities can add their own wording on top: a school's admissions page might rule out generative tools entirely, allow brainstorming and grammar checks, or ask for a written statement that the work is the applicant's own. Because these change each admissions cycle, build a small habit: before your first meeting each autumn, list the student's target schools and open each one's admissions page on AI. Save the date and the wording in the student's file. If a school's page is silent, assume the platform's general fraud language applies and stay at rung 3 or lower. The same caution applies to other pathways, such as scholarship applications and graduate programmes, which often have their own rules.

This is not legal advice. Where a family asks whether a particular practice is allowed, the honest answer is "check the school's current policy, and here is where I found it."

Disclosure: what to tell students and parents, and when

Disclose before the first essay session, in the engagement letter. It protects the family, the student and you. A usable clause, to adapt with your own legal adviser:

Use of AI tools. We may use AI software to prepare our own questions, to check
a draft against a marking guide and to flag spelling and punctuation. We never
use AI to write, rewrite or suggest wording for any essay or answer that the
student submits. Every word submitted is written by the student. Student drafts
are only entered into AI tools with a parent or guardian's written permission,
with names and identifying details removed. You may ask us at any time which
tools we use and what we have entered.

Give the same message to the student in plain words, face to face: "I might use a tool to spot problems in your draft, but you will fix them, and you must not use one to write any of it. If you do use one for anything, tell me." Students are more likely to admit a slip to a counsellor who has been clear and calm about the rule. Add a one-line declaration to your intake form that the student signs at each application stage: "I confirm the essay I submit was written by me; I have told my counsellor about any AI tools I used."

A safe review routine, step by step

1. Read first, with no tool. Form your own view of the draft. If you ask the AI first, its framing will anchor yours.
2. Remove identifiers. If you use AI at all, strip the name, school, town and any detail that points to the student. Use an initial or "the applicant".
3. Run a feedback-only prompt that forbids rewriting (below).
4. Filter the output. Cut any suggested wording. Keep only observations and questions.
5. Turn observations into questions for the student, in your own voice. "Your second paragraph moves from your grandmother to the school play without a link. What connects them?"
6. Student revises alone and sends the new version.
7. Compare versions. Check that changes are in the student's voice and that nothing sounds suddenly different.
8. Log it. Keep the dated versions and a line on what was discussed.

A feedback-only prompt that does not rewrite

You are a marking assistant for a college counsellor. Below is a draft essay
written by a student. Do NOT rewrite, rephrase or suggest replacement wording.
Do NOT write any sentences the student could paste in. List only:
1. The three clearest moments, quoted in under 8 words each.
2. Up to four problems of focus, order or clarity, each as an observation.
3. Up to four questions the counsellor could ask the student to help them
   deepen the piece.
4. Spelling or punctuation errors, listed by line, without corrections.
Word limit for the essay is 650. Tell me the word count.
[draft with identifying details removed]

An illustrative reply quotes three moments, observes that the opening anecdote runs to 150 words before the point, notes that the conclusion repeats the thesis and asks "what did you actually change about how you run the club after that meeting?" Your edits: check the quotes exist, drop any observation you disagree with and replace its questions with your own where yours are better. If the tool slips and offers "a stronger opening might be…", delete it. Models often volunteer rewrites even when told not to, so scanning for them is part of the routine.

A worked example: thirty students in the autumn rush

Say an illustrative counselling business has two counsellors and 30 students with main essays due in November. Each student produces about three drafts, so roughly 90 reviews. Done by hand at 25 minutes per review, that is 37.5 hours. With the routine above the counsellor still reads each draft, but the AI marking pass takes about two minutes and the question-writing is faster because the observations are listed. Suppose that brings a review to about 17 minutes. That is 90 times 17, so 25.5 hours, a saving of 12 hours across the season. The figures are assumptions, not measurements, so time five of your own reviews first. The point is that the saving comes from the sorting and listing, not from any writing. Spend the saved hours on what AI cannot do: the longer conversation that draws out a better story.

Before and after: the sentence the student fixed

A student's draft sentence: "I learned a lot about leadership from being captain and it taught me many things." The AI observation: "Generic claim; no concrete moment; 'taught me many things' says nothing." The counsellor's question: "What is one decision you made as captain that you would make differently now?" The student writes: "In March I benched my best player for missing practice, and we lost the next match. I still think I was right, but I would have spoken to him before the whole team saw it." The new sentence belongs to the student, because the idea, the memory and the words are theirs. Compare that with rung 4, where the tool suggests "As captain, I learned that leadership requires difficult decisions", a smooth line that could belong to a thousand applicants and which the admissions reader has seen before. The first path produces an essay that sounds like a person. The second produces one that sounds like everyone.

The student whose voice suddenly changes

An illustrative situation: a student's school English homework is plain and slightly clumsy, then draft two arrives with polished metaphors and no errors. You are not an investigator, and detectors are not the answer. OpenAI withdrew its own AI-text classifier in July 2023, citing its low rate of accuracy; in OpenAI's own launch tests it had caught only 26 per cent of AI-written text and wrongly flagged 9 per cent of human writing, and detection tools since then still carry caveats. Instead, have a conversation. Ask the student to talk through the second paragraph: why that image, what happened next, what they cut. A student who wrote it can answer easily. A student who did not will struggle. Handle it as a coaching moment, not an accusation: "This reads differently from your first draft. Tell me how you wrote it." If they used a tool, you reset the rule and ask for a rewrite from their own notes. Log the conversation. Never tell a family their child cheated on the strength of a score.

For more on why detector scores mislead, see the tutorial on trusting AI detectors.

Handling student data in AI tools

A personal essay is deeply personal: family circumstances, health, money, identity. It often belongs to a minor. Before you enter any text into an AI tool, check three things. First, the plan: a business plan that does not train on your content by default is safer than a personal one, and on a personal plan you must switch off training in the settings. Second, consent: the engagement letter should say student drafts may be entered with identifiers removed, and a parent or guardian should have agreed. Third, minimisation: paste the paragraph that needs feedback, not the full application, and never paste transcripts, financial details or recommendation letters. If a family declines, you simply do not use AI on that student's work. Keep that list of declining families somewhere you will see it before opening a tool.

Using AI for everything around the essay

The safest and most productive uses are the ones that never touch a student's words. Deadline trackers per school and portal, checklists for each application, parent update emails, a worksheet of questions for the brainstorming interview, a reading list. Building an AI-assisted tracker shows one approach, and seven consultant jobs from proposals to reports lists more. These tasks create most of the time saving with none of the integrity risk, which is why I would start there and keep essays for later.

When a request to tighten a paragraph became ghost-writing

An illustrative counsellor, short of time the night before a deadline, pasted a student's paragraph into a chatbot and asked it to "tighten" it. The tool returned a crisper version with a new closing sentence. The counsellor pasted it into an email as an example of what to aim for. The student copied the whole paragraph into the application. Nobody meant to ghost-write, yet that is what happened, because the example was a finished text. The fix is procedural: never put AI-written text into anything a student will see, not even as an example. If you want to show what concise looks like, write the example yourself from a different topic.

Short answers and supplements need the same rule

Supplemental essays, activity descriptions and short "why this school" answers are where students are most tempted, because they are short, numerous and tedious. The risk is higher too: a 100-word answer written by a tool is easy to spot because every one starts "I am drawn to…". Apply the rubric prompt to a batch, but keep the rule that the student writes each answer from their own notes. A practical device is the notes-first sheet: before drafting, the student lists three real things about the school (a course they read about, a professor's project, a club they would join) and one personal link to each. The counsellor checks the facts on the school's own site. Then the student writes. A filled-in line looks like this: "Course: the first-year seminar on urban design; link: I redrew our school's car park layout for a project and want to know why it failed." That is specific, true and theirs, and no tool could have made it up on their behalf.

When a parent wants more help than you will give

Some parents will ask you outright to "just fix it with AI". Having a script ready keeps the conversation calm. An illustrative reply you can adapt: "I understand the pressure of the deadline. Here is what I do: I tell [first name] what is working and where the essay loses the reader, and they rewrite it. I do not write or rewrite sentences, by hand or with software, because admissions readers are trained to notice and some applications ask the student to confirm the work is their own. What I can do is give faster, clearer feedback this week. Would you like me to book an extra session on Thursday?" Notice that it offers something, does not lecture and ties the limit to the family's own interest rather than to your principles alone. If the family insists, decline the work. A business built on ghost-written essays is exposed to complaints, platform sanctions and a damaged reputation, which cost far more than one fee.

A filled-in marking guide the AI can apply

A guide you write yourself makes the AI's pass more useful and more consistent across students. Four criteria are usually enough, each with a plain test.

Paste the guide into the prompt and ask the model to apply it. The answer should list, per criterion, one quoted line that passes and one that fails. Because the guide is yours, the feedback reflects your standards, not the model's taste, and the student gets a consistent message from week to week.

Checking the routine after the first cycle

At the end of the season, pull five students' files and look for three things. Does each essay have dated versions showing real change from one draft to the next? Does every file hold a signed declaration and, where AI was used, the parent's permission? And did any feedback record contain suggested wording that slipped through? If the answer to the last is yes, tighten the prompt, not just the habit. Ask the students too: what was most useful, and did anything feel like it was written for them? Their answers are the best evidence that the line is in the right place, and they give you honest material for next year's letter to families.

What to put in your own policy

• The rung limit: nothing above rung 3.
• That students write every submitted word and declare it.
• Consent and identifier-removal for any AI use on student text.
• Which plan or tool you use, and that training on your content is off.
• Keeping dated versions for each essay.
• A yearly review before each application season opens, when platform and school policies are most likely to change.

If you want help drafting that policy and the family letter, or mapping which parts of your admissions calendar are worth automating, that is what my AI implementation consultation covers. The rule that keeps it all simple is the one above: the tool may point at a problem, but the student fixes it, in their own words, and you can show the trail.

Questions college counsellors ask about AI and essays

Can I use AI to check a student's grammar?

Most counsellors treat spelling and punctuation correction as acceptable, but schools differ and platforms do not define the line precisely. Ask the student to accept or reject each change themselves, keep the original, and check the target schools' own statements on AI help.

Should I run essays through an AI detector?

Not as proof. Detectors give false positives and false negatives, and OpenAI withdrew its own text classifier in July 2023 for low accuracy. A better check is a conversation: ask the student to explain choices in the essay and compare it with their other writing.

Do I have to tell parents I use AI?

If you use it on their child's materials, yes, plainly and in writing before work starts. Put it in the engagement letter: what you use it for, what you never use it for, and that the student writes every sentence they submit.

Can I put a student's essay into ChatGPT?

Treat it as personal data about a minor in many cases. Use a business plan with no training on your content, get parental permission, remove names and identifying details, and keep the copy you paste in as short as the task allows.

Where does brainstorming stop and writing start?

If the student chooses the idea and writes the words, brainstorming is coaching. If the tool supplies the story, the structure or the sentences that end up in the essay, you have crossed into writing it for them, however much the student edits afterwards.

Further reads

• AI Automation for Education: Five Workflows, Tools and Costs — Five education-business workflows worth automating first.
• Opening a Tutoring Centre: Costs, Room Plan and AI Admin — Costs and admin when setting up an education business.
• How Much to Charge for Tutoring: Set Rates and Packages With AI — Set rates and packages for coaching services.
• How Tutoring Centres Use AI to Write Parent Progress Reports — Parent progress reports written without losing the personal touch.
• AI for Private Schools: Admissions, Fees and Parent Communication — How schools handle admissions communication with AI.
• Is AI Worth Paying For in a Small Tutoring Business? — Whether paid AI tools earn their cost in a small education business.
• How to Start a Language School: Accreditation, Agents and AI — What a new language school costs in year one, how accreditation and agents fit together, and where AI saves admin time from the first term.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Common App Fraud Policy (commonapp.org, last updated August 2023), OpenAI's note on its AI text classifier and its July 2023 withdrawal, checked November 2026.