AI can produce a personal statement in seconds. Give it the course you want to study, a few details about your qualifications and perhaps a couple of hobbies, and it will usually return something polished, grammatical and reasonably convincing.
That doesn’t necessarily mean it is a good personal statement.
There is an awkward tension at the heart of using AI for university applications. Artificial intelligence can be genuinely useful when you’re staring at a blank screen, struggling to organise your ideas or wondering how an apparently unrelated Saturday job could demonstrate skills relevant to your course. But the whole point of a personal statement is that it is supposed to reveal something about you.
And AI is remarkably good at producing writing which sounds personal without actually being very personal at all.
There is another complication. Applicants increasingly worry about whether universities can detect ChatGPT or other AI tools in their applications. AI detectors do exist, but their results aren’t proof of authorship, and both false positives and false negatives are possible.
So where does that leave you if you want to use AI responsibly?
First: the UCAS rules matter more than an AI detector
For applicants using UCAS, the first question shouldn’t be “Will this get detected?”
It should be “Am I allowed to submit this?”
The UCAS personal statement changed for 2026 entry onwards. Instead of one continuous piece of writing, applicants now respond to three questions:
- Why do you want to study this course or subject?
- How have your qualifications and studies helped you to prepare for this course or subject?
- What else have you done to prepare outside education, and why are these experiences useful?
The overall limit remains 4,000 characters, and admissions teams consider the answers together rather than as three separate mini-applications.
UCAS does not say that all use of artificial intelligence is forbidden. In fact, its guidance suggests that AI can be useful for brainstorming, structuring ideas and checking readability.
There is, however, a very important line.
UCAS says that generating all or a substantial part of a personal statement with an AI tool, then copying and submitting that text as your own words, could be regarded as cheating by universities and colleges. Applicants must also declare that their personal statement has not been copied or provided by another source, including AI software.
That means an AI-generated draft should never be treated as something you simply need to make “undetectable”.
If you use an AI personal statement writer, the sensible approach is to treat its output as writing support: something that can help you see a possible structure, identify points worth developing or show how your experiences could be connected to your chosen subject. Your final UCAS answers should genuinely express your experiences and ideas in your own words.
There is a big difference between using AI and letting AI write your application
“Using AI for a personal statement” can mean several very different things.
One applicant might ask:
What kinds of skills from working part-time in a supermarket could be relevant to a psychology degree?
Another might provide their own notes and ask AI to suggest an order for them.
Another might write a complete draft and ask whether any sentences are unclear.
At the other end of the scale is:
Write me a brilliant personal statement for psychology that will get me into university.
These aren’t equivalent uses of AI.
A useful way to think about it is to ask where the substance comes from.
If the experiences, observations, opinions and reflections are yours, AI can potentially help you organise them.
If the tool is inventing the experiences, supplying the reflections and deciding what you supposedly think about your subject, the result may be beautifully written — but it isn’t really your personal statement.
And invented details create an obvious additional problem: you could later be asked about them.
AI-generated personal statements can be surprisingly convincing
There is now a growing body of research into AI-generated personal statements, although an important qualification is needed. Much of the published research so far concerns medical school, residency and professional programme applications rather than British undergraduate UCAS applications.
We therefore shouldn’t assume that every percentage found in those studies applies directly to an 18-year-old applying for history at Leeds.
The broader findings are nevertheless interesting.
In several experiments, reviewers have found AI-generated personal statements perfectly acceptable and sometimes struggled to distinguish them from genuine applications. Some AI statements have performed extremely well for readability, grammar and organisation.
In one multicentre study comparing real residency applicants’ statements with AI-generated alternatives, however, the human-written statements performed significantly better for characteristics including originality, how compelling they were, career goals and the applicant’s specific reasons for pursuing the speciality (Lum et al., 2024).
That distinction makes sense.
AI is very good at producing a sentence such as:
Studying biology has strengthened my analytical skills and deepened my fascination with the complexity of living systems.
There is nothing grammatically wrong with it.
The problem is that thousands of applicants could say it.
Compare that with an applicant explaining how an A-level practical produced an unexpected result, what they initially thought had gone wrong, what they discovered afterwards and why that changed their interest in a particular area of biology.
The second example contains information an AI cannot know unless you give it that information.
That’s where a personal statement starts becoming genuinely personal.
The problem isn’t only whether the words are “original”
When people talk about originality and AI, two different ideas can easily become confused.
The first is textual originality: has this wording been copied from somewhere else?
The second is personal originality: does this statement contain experiences, thoughts and reflections which actually belong to the applicant?
A paragraph generated by ChatGPT may be completely new in the first sense. It may not reproduce a paragraph that exists anywhere else on the internet.
It can still fail badly in the second.
This is one reason conventional plagiarism checking and AI detection solve different problems.
A plagiarism checker generally looks for similarities between your writing and other material. AI detection instead examines characteristics of the writing itself and estimates whether they are consistent with machine-generated text.
An entirely AI-written paragraph could therefore have no meaningful plagiarism match at all.
Can universities tell when a personal statement was written by AI?
Sometimes — but not nearly as simply as you might imagine.
People aren’t particularly reliable AI detectors.
Research involving personal statements has repeatedly found that reviewers can have difficulty separating human and AI writing. Studies in different professional admissions settings have reported human identification rates ranging from around chance level upwards, with human statements sometimes incorrectly labelled as AI-generated (Koleilat et al., 2024; Whitrock et al., 2024).
One study memorably titled Does using artificial intelligence take the person out of personal statements? We can’t tell illustrates the problem rather well: polished AI writing doesn’t necessarily come with an obvious linguistic warning label.
That doesn’t mean admissions tutors are helpless.
Someone reading thousands of applications may notice generic writing, peculiar changes of voice, claims which don’t sit comfortably with the rest of an application or statements which say very little despite sounding impressive.
And if an applicant is interviewed, an unusual claim can simply become a question:
- “Tell me more about that book.”
- “What did you actually do during that work experience?”
- “Why did that experiment change your view?”
A genuine experience is usually much easier to discuss than one inserted because an AI model thought it sounded impressive.
What does an AI detector actually look for?
AI detectors don’t normally “find ChatGPT” in the way a plagiarism checker can find a matching webpage.
They analyse the text.
Large language models generate language statistically, predicting probable words or tokens from the surrounding context. That can leave measurable patterns in the resulting writing.
Different detection systems work differently, but commonly discussed characteristics include:
- how predictable the language is;
- patterns in vocabulary and sentence construction;
- variation in sentence structure;
- relationships between words and phrases;
- recurring linguistic patterns associated with known AI output.
Terms such as perplexity and burstiness often appear in simplified explanations of AI detection. They essentially concern how predictable text is and how much its structure varies.
Modern commercial detectors can use much more sophisticated classifiers than these two measures alone.
The important point is that an AI detector is making an inference from patterns.
It isn’t accessing your ChatGPT history.
It isn’t discovering who physically typed the words.
And it doesn’t possess a database containing every sentence ChatGPT has ever generated.
Are AI detectors accurate?
Some perform impressively in controlled testing. Others perform much less well.
And even a strong detector can perform differently depending on what it is asked to analyse.
Research into personal statements has produced particularly variable findings. Automated systems have sometimes identified AI-generated application writing considerably more accurately than humans, but studies have also found false positives on genuine human statements (Cumbo et al., 2025).
Researchers have also raised concerns about differences in detector performance between writers. Some studies have found that characteristics associated with non-native English writing can increase the risk of human text being classified incorrectly as AI-generated. That doesn’t mean every AI detector has this problem, but it is another reason not to treat the output from an unknown detector as unquestionable evidence.
The underlying technology is moving quickly as well.
A detector tested against one generation of language models cannot automatically be assumed to perform identically against future models. Detection systems themselves are retrained and updated.
The question “How accurate are AI detectors?” therefore has no single permanent answer.
A better question is:
How well does this particular detector perform on this particular type of writing, produced by these models, under these conditions?
A high AI score is not proof that somebody used ChatGPT
This point is worth emphasising.
An AI detector score should not be interpreted as a forensic finding of authorship.
Suppose an AI detector flags part of a statement.
There are several possibilities:
- the passage was generated by AI;
- the applicant wrote it but an AI tool substantially rewrote it;
- the applicant wrote it themselves and it happens to resemble patterns the detector associates with AI;
- a mixture of human and AI writing has confused the classifier;
- the detector has simply made a mistake.
The reverse is also true.
A statement receiving a low AI score isn’t proof that no AI was used.
Detection systems produce evidence about the text, not a historical record of how that text came into existence.
This is why an admissions decision or misconduct allegation should never rest unquestioningly on a percentage from one detector.
Should you check your personal statement with an AI detector?
You can, but be sensible about what the result means.
Running your draft through an AI detector such as PlagPointer can give you an indication of how automated detection software interprets the writing. PlagPointer is particularly interesting in this context because it offers different sensitivity settings, allowing the user to choose a more cautious threshold or a more sensitive check rather than pretending there is one magical dividing line between “human” and “AI”.
But don’t turn the process into a game of repeatedly rewriting perfectly genuine sentences until the percentage reaches zero.
That is missing the point.
If you’ve written the statement yourself and a passage is flagged, first read it as a human being.
- Is it unusually generic?
- Does it contain several polished but rather empty sentences?
- Did Grammarly, ChatGPT or another rewriting tool substantially alter that section?
- Does it actually sound like the way you express yourself?
Sometimes a detector result can draw your attention to a paragraph that really would benefit from becoming more specific and personal.
Other times, the detector may simply be wrong.
Personalise AI-assisted writing for the reader, not the detector
If you’re working from an AI-assisted draft, one of the worst approaches is to search for tricks that supposedly “humanise” AI text.
Changing a few words, introducing deliberate spelling errors or asking another AI to “make this undetectable” doesn’t make the statement more authentic.
Instead, go back to the content.
Imagine AI produced this:
My work experience developed my communication and teamwork skills and reinforced my desire to pursue a career in healthcare.
Ask yourself:
- What work experience?
- Who did I communicate with?
- Was there a particular interaction I remember?
- What did I find difficult?
- What did somebody else do that impressed me?
- What did I understand about healthcare afterwards that I didn’t understand beforehand?
You might end up with something closer to:
During my placement at a residential home I initially found conversations with residents who had advanced dementia difficult, because questions I thought were straightforward sometimes caused frustration. Watching one of the care assistants use familiar songs and photographs to begin conversations showed me how much good communication depends on the individual rather than simply speaking clearly.
That’s better not because it is “less detectable”.
It’s better because somebody actually lived it.
Specificity is one of the strongest protections against the blandness of generated writing.
AI can invent surprisingly plausible details
One of the easiest ways to damage an application is to submit something you haven’t checked carefully.
AI models don’t know that an assertion is true merely because they can express it confidently.
If you tell a system that you’ve completed work experience at a law firm, it might transform that into observing a court hearing, preparing case files or discussing cases with solicitors even if none of those things actually happened.
It can also get books, academics, course modules, university facilities, scientific facts and historical information wrong.
UCAS specifically advises applicants to check information suggested by AI tools.
For a personal statement, fact-checking has an additional dimension:
Did this actually happen to me?
Read every sentence with that question in mind.
Your voice matters more than sounding “academic”
AI has a tendency to make applicants sound as though they’ve suddenly become university prospectuses.
You get sentences about “fostering a profound appreciation”, “navigating the complexities” of a subject and embarking upon “transformative academic journeys”.
None of those phrases proves that AI was involved. Human beings write clichés too.
But they usually don’t tell an admissions tutor much.
Research into generated personal statements has repeatedly identified this tension: AI can produce highly polished material while human statements retain an advantage in genuine personal reflection. In one study, AI-generated statements were difficult for reviewers to identify but tended to rely more heavily on formulaic linguistic patterns (Johnstone, Neely and Sizemore, 2023).
Another study comparing human and AI-generated residency statements found significant differences in areas including originality and the presentation of personal motivations (Lum et al., 2024).
So don’t replace an ordinary sentence that sounds like you with a complicated one merely because the AI version sounds cleverer.
Admissions tutors aren’t marking a vocabulary test.
The new UCAS questions actually make personalisation easier
The three-question UCAS format is well suited to using your own material rather than producing generic prose.
- For the first question, don’t simply tell the reader that you’re “passionate” about the subject. Show where the interest comes from.
- For the second, don’t simply list your A levels. Pick particular topics, projects or pieces of work and explain what you learned from them.
- For the third, don’t produce an inventory of clubs, employment and volunteering. Choose experiences that genuinely add something and explain what changed because of them.
AI can help you ask the right questions.
Only you can supply the answers.
A safer way to use an AI personal statement writer
If you’re using an AI writing tool, this workflow keeps the emphasis where it belongs:
- Start with your own raw material. Write down your subjects, projects, reading, work experience, employment, volunteering, interests and reasons for choosing the course.
- Give it details, not a persona to invent. Don’t ask AI to make you sound impressive by filling gaps.
- Use the first output as a draft, not a submission. Look at how the information has been organised and which connections are useful.
- Delete anything that isn’t true. That includes small embellishments.
- Interrogate every generic sentence. Replace claims such as “this developed my communication skills” with what actually happened.
- Write important reflections yourself. In particular, the reasons you want to study the subject should come from you.
- Check facts independently. Never assume a book title, course detail, case, statistic or scientific claim is accurate simply because AI supplied it confidently.
- Read the finished version aloud. If you would never say anything remotely like a sentence in real life, consider rewriting it.
- Follow the rules of the application you’re actually making. UCAS, overseas universities, postgraduate courses and individual institutions may have different policies on AI assistance.
- Don’t optimise for an AI detector. Optimise for accuracy, specificity and authenticity.
Keep your drafts
There is another simple habit worth adopting: keep your working material.
Save your original notes. Keep earlier drafts. If you write in software with version history, don’t immediately delete it.
This isn’t because you should expect to be accused of anything.
It’s simply useful evidence of your writing process.
If a question ever arises about how a statement was produced, being able to show that a paragraph developed from rough notes through several drafts provides far more useful context than arguing about whether a detector said 17% or 71%.
Don’t forget privacy
There is one less obvious issue with AI personal statement tools: your data.
A personal statement can contain information about your education, employment, family circumstances, health, ambitions and other personal experiences.
Before pasting that information into any AI tool, check what happens to it.
Does the service retain your prompts? Can your data be used for model training? Can you delete it? Is the statement added to any shared database?
You should be just as selective about where you upload a personal statement as you would be about where you upload a CV.
What applicants should remember
AI isn’t inherently the enemy of a good personal statement.
Used carefully, it can be a useful sounding board. It can help you organise a messy page of notes, suggest questions you haven’t considered, identify repetition and show you where an explanation is unclear.
But a personal statement becomes weaker when AI replaces the very thing the document is intended to communicate: the applicant.
Published research gives us an interesting warning here. AI-generated statements can be polished enough to fool experienced reviewers, yet human-written statements can still perform better when originality, motivation and personal detail really matter.
AI detection adds another layer of uncertainty. Some detectors perform well. Some produce false positives. Human reviewers also get it wrong. No score can reconstruct the entire history of a document from the finished text alone.
So don’t build your personal statement around the question:
“Will they know I used AI?”
Build it around a much better one:
“Does this genuinely tell them something worth knowing about me?”
If AI helps you answer that question, it can be useful.
If it starts answering the question for you, you’ve probably given it too much of the job.
Try our personal statement writer and collaborate to make the statement your own.