Statementory vs ChatGPT: Which Should Check Your Personal Statement?

Every autumn, thousands of students paste their personal statement into ChatGPT and type "review this". It is free, it answers instantly, and it usually says something nice. The question is whether what comes back tells you anything true about how an admissions tutor will read your statement.

We build a specialised checker, so you know where we stand. But this comparison is straight: ChatGPT is genuinely useful for parts of this process, and we will tell you which parts. Where it fails, we will show you exactly how and why.

Where ChatGPT genuinely helps

Credit first. A general chatbot is good at:

• Brainstorming. "What could I write about for question 3 if my only experience is a supermarket job?" produces useful raw material.
• Interrogating your draft. Asking "what questions would an admissions tutor have after reading this?" surfaces gaps.
• Explaining the process. The format, the deadlines, what the three questions ask (though verify details; it confabulates).
• A quick companion read. A second pair of eyes at 11pm when no human is awake.

All free, all low-risk, all early-stage. Use it for this without guilt.

We ran the test: ChatGPT scored it 91, we scored it 72

Rather than describe this, we measured it. We took one Economics personal statement in the 2027 three-question format, 3,804 characters, and gave the identical text to both with the same instruction: score this out of 100.

ChatGPT: 91/100. Statementory: 72/100.

Nineteen points apart on the same words. Where that gap comes from is the whole argument of this article, so it is worth being precise.

ChatGPT found the same weaknesses. It just did not price them

This is the part that surprised us. ChatGPT did not miss the flaws. Its own write-up called the closing sentence "the weakest sentence in the statement", and said of the phrases in it:

"genuine passion", "discipline", "time management" and "asset to your university" are generic phrases that could appear in thousands of UCAS statements.

That is an accurate criticism, and close to what our own report said about the same line. ChatGPT then scored the statement 91, gave its lowest category (personal qualities and reflection) 15 out of 20, and concluded that with a sharper ending it would sit "around 94–96/100".

So the failure is not perception. A chat model can see a cliche perfectly well. What it will not do is take marks off for one. Naming a weakness and pricing a weakness are different operations, and only the second one moves a score.

It missed the error that is specific to UCAS

The statement ends "would make me an asset to your university". A UCAS personal statement goes to all five of your choices, so that sentence is read by five universities at once, four of which are not the one being addressed. That is a factual error about how the application works, not a matter of taste.

ChatGPT quoted the sentence twice while criticising its tone, and did not mention the problem once. That is what "no admissions criteria" looks like in practice: it assessed the writing competently and had nothing to say about the mechanism the writing sits inside.

It did not notice the statement was written by AI

We should be straight about this, because it is the sharpest finding here: we wrote the test statement with AI, deliberately, as a control.

The same text through an independent AI-detection tool scored 10 out of 100 for reading as human writing. ChatGPT, reading the same words, never raised it. It praised the voice, called the writing "clear, mature and economical", and gave 18 out of 20 for style.

If you are using a chatbot to check whether your draft reads as machine-written, that is the result to keep in mind. It is not equipped to tell you. Our free AI detector is, and it is the same check that runs inside every paid report.

Why the gap exists

A score means nothing unless it is anchored to a distribution. Ours is calibrated to bands:

Against that scale, 91 places an AI-written first draft in the Oxbridge band. ChatGPT has no such scale. It has never been shown what a 72 looks like beside an 86, so when you ask for a number it produces one that sounds encouraging, and produces a different one the next time you ask.

You can reproduce that last part in two minutes: ask a chatbot to score your statement, then open a new chat and ask again.

The four failure modes when it reviews

1. It praises almost everything

Chat models are tuned to be agreeable. Paste in a 50-scoring draft full of cliches and the typical response opens with "This is a strong and compelling personal statement..." before offering gentle suggestions. Students walk away reassured about drafts that would be flagged all over by an admissions reader. This is the single most damaging failure, because the feeling of having been checked replaces the reality. The test above is this failure with a number attached: 91 for a draft that scores 72.

2. It has no admissions criteria

Ask ChatGPT what a UCAS statement scores and it will invent a number with no basis. It has never been benchmarked against real statements and their outcomes. Its advice is generic good-writing advice: vary sentence length, show don't tell. It is also why it read a statement addressed to a single university, in a system that sends it to five, and said nothing. True, and nowhere near the specific things tutors score: subject insight, evidence of sustained engagement, reflection depth per question.

3. It is inconsistent

Same draft, same prompt, three runs: three different sets of feedback, often contradictory (tighten the opening / the opening is your strongest section). Useful assessment is repeatable; a signal that changes on refresh is noise.

4. It will rewrite, and rewriting is the trap

Ask for improvements and it produces polished, fluent, oddly weightless prose in the house style detectors are trained on. UCAS runs similarity checks and universities screen statements for AI authorship. Paste-in ChatGPT sentences are exactly what gets flagged, and we wrote up what happens when UCAS detects AI. Worried about a draft you have already touched up? Our free AI detector shows you which phrases sound machine-written. And if you wrote every word yourself and it still reads that way, that is a different problem with a different fix: why a statement you wrote yourself sounds like AI.

What a specialised checker does differently

Statementory exists because reviewing a UCAS statement is a narrow, learnable task that general chatbots are structurally bad at. The differences, concretely:

The "never writes it" line is not marketing hedging; it is the design decision the whole product is built around. Your final statement stays your writing, which is what UCAS requires and what keeps detection risk at zero.

The honest verdict

Use ChatGPT for what it is good at: free brainstorming and early-draft questioning. Do not use it to find out whether your statement is good, because it will tell you it is regardless.

When you need the real answer, the specific lines to fix, and a score that means something, that is a specialised review. £7.49 and ten minutes against three years of university is one of the easier trades in this process, and if the report is not useful, you get the money back.

For the wider landscape of every checking option, including human services and free tools, see our comparison of the best personal statement checkers.

Statementory is the publisher of this comparison, and the test statement above was written by us for the purpose. The head-to-head was run in September 2026. Chat models are non-deterministic, so an identical prompt will not return exactly 91; the behaviour it demonstrates, praising a draft it has itself criticised, is the repeatable part.