How GradPilot works
Scores that don't swing.
Using AI to judge high-stakes work — an essay that can decide an admission — only works if the judging is reliable. Ask a chatbot to grade the same essay twice and the score can swing wildly. GradPilot's scores drift a little at most — small, even drift that still shows you the direction: did your revision help, or not.
Scores can drift a little between runs. What they don't do is swing.
The evaluation
Three stages. One fixed job each.
A Full Review is not one big opinion. It runs three separate stages, each with its own fixed criteria — so nothing vague decides your score, and no strength hides a problem.
AI signal
Could any sentence read as AI-written?
Optional. A check before schools run their own detectors — a signal, never proof.
The writing
Is it clear, structured, and in your voice?
Writing quality is scored on its own, so a great story can't hide unclear writing.
The prompt
Does it do the job of your application?
Scored against the rubric for your exact application — a Common App essay and a med school statement are different jobs.
And when a section is already strong, the review says so instead of inventing feedback. In your report, the writing and prompt stages appear as your Foundation and Focus scores.
Every critique quotes the sentence it is about.
The rubrics
The rubric guardrails the AI.
A generic AI grades from a vibe — whatever the prompt and the moment happen to produce. GradPilot's AI is rubric-aware: held to deep, specific criteria built for your exact application. It can't wander, because every judgment has to land on a criterion you can read yourself — and it can't flatter you, because praise has to point at a criterion too. Students call the flattery gradeglazing.
A generic AI
IllustrationSame essay, three tries. A different score and different advice each time — whatever the essay is for.
The GradPilot review
1 of 7 review areas“Then I stepped into a surgical oncology clinic.”
Critique
You never state your role in the clinic — the reader cannot tell why you were present, what you were permitted to do, or whether this was a single observation. That ambiguity weakens an otherwise specific patient moment.
Suggestion
Name your role and stay inside it: how you came to observe the consultation, and what you could responsibly take from it. Admissions readers check this boundary.
The AMCAS personal statement rubric — every area scored, every critique tied to your sentences.
How a rubric gets built
Research
Start from the official prompt, the application's rules, and what admissions readers say they evaluate.
Draft checkable criteria
Every criterion must point at something in your sentences. Vibe words like “memorable” are banned.
Calibrate
Run the rubric on real sample essays and check that scores and feedback land where experienced readers would land.
Audit, then re-audit
A structured review against our design principles plus mechanical consistency checks — repeated after every revision.
Workflows
Tools that run above the evaluation.
Some jobs come after the review. Workflows use the finished evaluation to do them, and they follow one rule: help your score, or leave it alone.
Trim shortens your essay using what the review knows is earning you credit — a generic trimmer or a “humanizer” edits blind, and can cost you points. Reviewer sharing lets people you trust respond to the same evaluation instead of adding one more random opinion.
Trim
Cuts words, never points.
Reviewer sharing
People you trust, on the same evaluation.
The evaluation
Every workflow is held to the evaluation: help the score, or leave it alone.
Stage 01 · AI detection
It's a signal, not proof.
Many schools run AI detectors on essays. GradPilot runs one first, so you're not surprised later. With detection on, a Full Review flags the sentences that might read as AI-written. You see exactly which ones, and you decide what to change.
- This check is for you. Your essay is never reported to anyone.
- Detection never claims to prove who wrote a sentence.
- Pro users can turn it off. Free Quick Reviews don't run it at all.
Detection partner
We use Pangram, a specialist AI detector, instead of asking a chatbot to judge its own writing.
Clear
No sentences stand out.
AI-polished in places
A few sentences worth rewriting in your own voice. Your scores are not affected.
Flagged
Parts read as AI-written. Shown to you, not to schools.
Each flagged sentence names the pattern it shows, for example:
What is allowed
Permission
What you must say
Disclosure
How it may be checked
Enforcement
University AI policies
High-stakes work comes with rules.
Schools treat AI differently. Our directory tracks what each school allows, what you must disclose, and how rules may be checked — so you know the rule before you submit.
Every record links its source and shows when we last checked it. Always confirm the current instructions on the school's own site.
Search 180 university policiesYour words stay yours
A score you can revise against.
Start with a free Quick Review. Make the changes you agree with. Run it again to see if they helped — a Full Review is $5.