Essay Feedback
Paste a paragraph from a college essay and get rubric-based feedback on voice, specificity, reflection, structure and clarity, with evidence highlighted in your own words. A look at where AI helps an admissions writer, and where it doesn't.
1 evidence for each note cliché
What you’re looking at
This is the first pass I’d give a paragraph as a counsellor, done by Claude Sonnet 5. It reads one paragraph and scores five things admissions readers actually respond to:
- Voice: does this sound like a specific seventeen-year-old, or a brochure?
- Specificity: scenes, names, numbers, the oven that runs hot on the left side.
- Reflection: does the paragraph show how the writer thinks, not just what happened?
- Structure: does it move somewhere and end on purpose?
- Clarity: is every sentence pulling its weight?
Each note comes with a quote from your paragraph as evidence. Hover a note to see its evidence light up. Clichés get a wavy underline. Try the three samples back to back: the generic one and the specific one describe similar kinds of experience, but read completely differently.
Designed to coach, not to ghost-write
The model is told never to rewrite the paragraph. It may rewrite one sentence as an example, and it must use [bracketed placeholders] for any detail only the student knows, rather than inventing experiences.
It mostly follows that rule, but not perfectly. In testing, it revised a strong paragraph by adding a sentence about what the student “realized”. That’s a reasonable suggestion, but it’s a feeling the student never wrote. That is exactly the risk with AI and personal essays: the moment a model starts supplying the insight, the essay stops being the student’s. Use the revision to see the kind of change it’s pointing at, then write your own.
What it can’t judge
- Whether it’s true, or yours. An AI can’t tell a lived experience from a plausible one.
- The whole essay. A paragraph that looks weak in isolation can be exactly right as a setup. Structure only makes sense across the full piece.
- Fit. What works for one school’s prompt, or one student’s application story, isn’t something a paragraph-level rubric can see.
- Taste. The scores are consistent, not objective. Two experienced readers would disagree on some of these, and so would the model on a second run.
Guardrails
Each visitor gets 5 reviews a day, requests pass a Cloudflare human check, and the lab has a daily spending cap. Paragraphs aren’t stored by this site. Please don’t paste real student work that includes names or private details.