evaluchat
Why this design
Consumer AI will generate a fluent essay from a short prompt. That makes the take-home essay a weak signal of student language skill. evaluchat keeps the familiar chat interface, then adds a structural constraint: the model cannot produce the whole assignment on demand. Students must put ideas into language in the dialogue first.
That matters for teaching because prompting is writing. For complex work, result quality tracks prompt quality. Immediate feedback in dialogue develops clarity and vocabulary. When you can read the actual prompts, you can evaluate language development — not only a finished text with no trace of what the student wrote.
Vs. Docs revision history: Google Docs history is appropriate when the goal is a classroom without AI. evaluchat is for classrooms that use AI on purpose — in a way that develops student agency instead of replacing it.
Getting started
Logging in
Visit evaluchat.com and sign in with the email on your teacher account. If you need an invite or help getting started, email hello@evaluchat.com.
The teacher workspace
After login you land on the teacher dashboard
(/teacher). From here you manage classes and assignments,
open submissions, and read dialogue and engagement metrics for completed work.
Creating assignments
Use Create assignment on the teacher dashboard. Each assignment typically includes:
- Title — what students see on their list
- Prompt — the essay question or writing task
- Coach instructions — tone, subject focus, and how tightly the AI should hold the constraint
- Settings — course, due date, free vs premium tier, and which students are assigned
Free assignments use budget AI models and do not spend credits (within the
free caps: 5 open assignments, ≤100 students each). Premium assignments use
a higher-capability model; 1 credit is taken per student when you assign
them (buy Entry credits in the workspace if needed). Once assigned, the task appears on each
student’s /student page so they can start or continue.
Tip: Start with one short assignment so students learn that better prompting unlocks better help — before a high-stakes essay.
Reviewing student work
Submission list
For each assignment you can see:
- Student name
- Status — Not started, In progress, Submitted, Abandoned
- Progress — completion percentage while in progress
- Submission date when completed
Opening a submission
Click a student row to open the review view:
- Chat transcript — the prompts and negotiation (primary language evidence)
- Final canvas — the document at submission time
- Engagement metrics — how drafting sessions unfolded (context, not verdicts)
Read the dialogue first when judging language growth: thesis clarity, vocabulary, willingness to revise ideas in words, and how much conceptual work the student contributed before any co-generation.
Engagement metrics
Metrics highlight patterns in drafting behaviour. They are conversation starters — not automated integrity scores. A high paste ratio can mean research quotes; a quiet session can mean thinking. Always read the transcript with the numbers.
| Metric | What it shows |
|---|---|
| Writing sessions | Distinct working sessions (breaks of 5+ minutes start a new one) |
| Keystrokes | Total key presses in the editor — drafting volume |
| Writing bursts | Rapid typing stretches that often look like fluent composition |
| Canvas edits | Distinct edits to the document |
| Paste count | How often text was pasted into the editor |
| Paste ratio | Pasted characters vs typed characters — worth discussing when high |
Language to prefer: “engagement metrics,” “process signals,” “worth discussing with the student.” Avoid “cheating indicators” or automated guilt flags — that framing is not how evaluchat is designed.
How the constraint works
The student-facing experience is meant to feel like a normal AI writing chat — flexible, conversational, modern. Under that surface, the model is programmatically restricted from drafting the full assignment (or substantial canvas prose) until the student demonstrates enough conceptual and linguistic contribution in the dialogue.
- One-line “write my essay” requests do not produce a finished product
- Students must articulate ideas, evidence, and wording in chat
- When engagement is sufficient, limited co-generation can unlock (proportional scaffolding)
- Students retain authorial control on the canvas — typing, editing, accepting or rejecting help
Behaviour is shaped by the coach instructions you set on the assignment, so you can tune tone to your subject and classroom norms.
Research frame (for academically minded teachers)
If you want the theory and open research questions behind the constraint — not just the classroom workflow — start with the CAMDLE discussion paper (Constrained AI-Mediated Dialogic Language Education). It is a conceptual framework and research agenda, not a completed study or an efficacy claim.
Central question: what counts as sufficient dialogic contribution to unlock drafting support — and how does that threshold vary by task, proficiency, and language background?
Download PDF (18 pp.) Read online
Interested in a co-designed study? Email hello@evaluchat.com. (Research rebate terms, if any, are arranged case-by-case — not an open free pilot.)
Frequently asked questions
Can I see who has started?
Yes. Status shows Not started, In progress, Submitted, or Abandoned.
Can students edit after submitting?
No. Submission locks the document to read-only for the student.
What about abandoned work?
Abandoned sessions stay visible so you can review progress so far and follow up.
Why not just use Google Docs revision history?
Docs history shows how a text changed when students write without AI (or outside AI). evaluchat is for assignments where AI is allowed and expected — constrained so students still produce the linguistic labor, and so you can see the prompts that drove the result.
How is student data protected?
Traffic uses HTTPS. Auth is via Supabase. Evaluchat does not train its own models on customer content; budget AI providers may use inputs/outputs under their terms — see Privacy. Contact hello@evaluchat.com if your school needs a DPA discussion.
How does pricing work?
Free budget tier (early access): up to 5 open assignments and ≤100 students each, on budget AI models, with no credit burn. Closing an assignment frees a free slot. Premium uses Entry credits (~US$5 for 50): 1 credit per student when you assign them to a premium assignment (not when they submit). Buy credits from the teacher workspace at evaluchat.com. For volume discussion, email hello@evaluchat.com.