Most AI safety is a polite request in a prompt that a clever user can talk around. Ours isn't. A separate validation layer inspects every citation the model produces and drops any that don't point to a source the student actually supplied — before it ever reaches the page. The promise holds even if the model misbehaves.
An essay is judged on thesis and evidence; a worksheet on grounded answers; a lab report by its sections. One tool, the right expectations.
Socratic, mechanics-only, editor, critical, encouraging — and Corrections mode, which answers directly and tags every answer's provenance.
It labels an own-knowledge answer versus a grounded one, and every screen carries the reminder: double-check answers — this tool can be wrong.
No new app to learn. It works in Word, where the draft already lives — comments land on the sentence, feedback survives across sessions.
The AI key and every model call stay on the server. The student's add-in never carries a secret, and never talks to the model directly.
The engine is separated from Word by design — the same reviewer can front a Google Docs or web client with no change to the guarantee.
| Essay Helper | Grammarly | Generic AI chat | Turnitin | Human TA | |
|---|---|---|---|---|---|
| Feedback on argument & evidence | ● | ○ | ● | ○ | ● |
| Coaches without writing it for them | ● | ◐ | ○ | ○ | ● |
| Never invents sources or facts | ● | — | ○ | — | ● |
| Works inside the document | ● | ● | ○ | ○ | ◐ |
| Respects academic integrity | ● | ◐ | ○ | ● | ● |
| Available 24/7, at scale | ● | ● | ● | ● | ○ |
● full ◐ partial ○ no — not applicable
Generative AI went from novelty to default in classrooms in under two years. Institutions now face the same question at every level: how do we let students use AI without letting it fabricate, plagiarize, or replace the learning? The tools on offer are either general-purpose chatbots with none of those guardrails, or plagiarism detectors that police students rather than help them.
Essay Helper is positioned exactly in that gap — an AI that administrators can approve because of how it's built. The buyer (a department, school, or university) is actively shopping for a safe, aligned writing tool, and there is no incumbent that both coaches and guarantees integrity.
of students write assignments in Word or Docs every term — the surface the product already runs on.
AI use in coursework is no longer fringe; policy and procurement are catching up fast.
Coaching tools ignore integrity; integrity tools ignore coaching. We do both.
Land with individual students; expand to the institutions that need a tool they can sanction. Figures below are an illustrative starting model.
The institutional license is the revenue engine: it's sold on the one thing individuals can't buy elsewhere — a writing AI the institution can vouch for.
In-Word review, citation coaching, six modes, five assignment types, Corrections mode, conciseness controls, the no-fabrication guarantee, and the honesty disclaimer — all built, tested, and merged.
Compare a draft against its earlier versions — praise real revision, catch regressions, and coach the arc of the writing, not just the snapshot.
Bring the same guaranteed engine to Docs (the core is already client-agnostic), and give instructors a view of where their class struggles.
Rostering & LMS integration, admin dashboards, more citation styles and languages — the surface a site license is sold on.
The engineering risk is behind us — the product is built, safe, and shipped. What it needs now is a first classroom and a short runway to the next surface. Three things unlock the next stage: