How does EnlightenAI work, and what should you check before adopting it?

Details about other products were checked on 29 July 2026. Other providers change their features and pricing, so check their site before deciding.

What does an AI marking assistant actually do?

The category is narrower than the marketing suggests. An AI marking assistant takes student responses, produces a draft mark and draft comments against criteria you supply, and hands them back to you to check. It does not decide anything. In every tool worth considering, the teacher is the one who releases the result. That single design choice is what keeps the work defensible when a parent asks how a mark was reached.

Because these tools share that shape, “which one is better” is usually the wrong question. What separates them is narrower and more testable: whether it marks to your criteria or its own scale, how specific its comments are, how much of your curriculum it recognises, and what happens to uploaded student work.

There is also a quiet cost that product pages never mention. Editing a wrong draft takes longer than writing a comment from scratch, because you have to read it, judge it, then repair it. A tool that produces plausible but interchangeable feedback hands you a class set of small repair jobs. Test every claim about hours saved against your own scripts.

What does EnlightenAI do, step by step?

EnlightenAI publishes its own workflow in detail, and its help centre sets out the steps plainly. You build an assignment with a rubric and instructions, then mark five sample submissions yourself. Its documentation states that after five manually graded samples the model reaches full training, and that it then grades the rest “matching your standards and tone”. You review and edit anything before students see it, then click to finish grading.

That calibration step is a real design decision rather than a slogan. Most tools ask you to describe your standard in a rubric. This one asks you to demonstrate it on real scripts first, and its essay grader page says each adjustment you make during review continues to train the system towards your preferences. If your faculty already moderates by marking a common sample, the pattern will feel familiar.

On accuracy, the company reports a quadratic weighted kappa of 0.77 against the teacher who trained the model, from a study it ran with DREAM Charter Schools, and describes that as higher than teacher-to-teacher agreement on the same scoring guide. Treat it as a vendor-reported figure from one setting rather than an independent finding, and note that it measures agreement with one trained teacher, not correctness. Publishing a number you can argue with, and naming the setting it came from, is worth more than another claim about hours saved. That deserves saying plainly.

The published scope is grades 3 to 12, roughly Year 3 to Year 12 here, and written work: essays, drafts, lab reports and research papers. It connects to Google Classroom, Canvas and Clever. An immediate feedback mode lets students receive AI feedback as soon as they submit, then revise and resubmit, with the teacher setting anywhere from one to eight attempts. That is closer to a drafting tool than a marking one. A free plan is advertised, and no pricing page was published at the time of checking.

Which differences actually matter when you compare tools?

Once you look past the demo, a short list of things decides whether a tool earns a place in your workflow. Rank them for your own context before you trial anything, because a tool that is excellent on one dimension and weak on another can still be the wrong choice for a Stage 6 class.

Rubric alignment sits at the top for a reason. A mark has to hold up against the criteria your faculty agreed and, in senior subjects, against a published marking guideline. A tool that can only mark to its own built-in scale may read well in isolation and still drift from the standard your moderation meeting applies. Ask whether you can upload your own rubric, and whether each comment is tied to a named criterion. Many faculties already publish study guides restating those criteria in student-facing language, and a tool’s output should sit alongside them.

Feedback specificity is the one teachers underrate. A tool can agree with you on every mark and still write comments that fit any script in the pile. The comment is what changes what a student does next.

• Rubric alignment: can it mark against your criteria and standards, not a generic scale?
• Teacher control: do you review and confirm every mark and comment before it is released?
• Feedback specificity: are comments anchored in this student's work, or interchangeable?
• Curriculum fit: does it recognise your year level, subject and task type?
• Data privacy: where is student work sent, stored and processed, and for how long?
• Evidence: has the vendor published any agreement data you can interrogate?

How do you run a trial that tells you something?

Take one assessment you have already marked, ideally ten to fifteen scripts spanning the range. Hide your own marks. Run the set through the tool. Then compare three numbers: how many drafts land in the same band as your mark, how many move a band, and how many comments you would send to a student without editing a word. That third number is the one that predicts your time saving.

Include your hardest scripts, not your neatest: the very strong, the very weak, the one that misread the question, the one written by a student still building academic English. Add a task type the tool has not been shown, such as a source analysis or a practical report. Straightforward essays are where every tool is strongest; the edges are where they differ.

Do it with a colleague if you can, for a second read on whether the comments are worth sending. Timely, specific feedback is part of the professional standards teachers already work to. The test is not whether a draft is fast, but whether it is something you would have written yourself.

What should Australian teachers check on curriculum fit and privacy?

Curriculum fit is where a locally built tool and an overseas one separate, and it is easy to check. EnlightenAI’s own pages name AP History and English, STAAR Writing, CAASPP and Smarter Balanced, Common Core, IB Diploma, ACT and SAT Writing, and NY Regents. That is a strong list for a United States faculty and for IB schools anywhere. Those pages do not name the Australian Curriculum, or senior certificates such as the HSC, VCE and QCE. That is not a fault, just who it was built for.

The practical consequence is translation work. Australian achievement standards and syllabus criteria use particular language, and a Year 9 persuasive piece is marked against different expectations from a Stage 6 module response. If a tool does not recognise your year level, subject and task type, you will spend the time you saved rewriting its output into the terms your students and your moderators expect.

On privacy, get your answers in writing before any student work is uploaded. EnlightenAI’s privacy policy states that no AI provider it uses has the right to train models on any data processed on its behalf, which is exactly the commitment you want a vendor to state plainly. That policy is framed around United States education law, addressing FERPA and parental consent under COPPA. An Australian school still has to satisfy the Privacy Act and its own approval process, so ask where work is stored, how long it is kept, and who signed off.

Where does a tool built for Australian marking fit in?

If your classes sit inside the Australian system, it is worth trialling at least one tool designed around it alongside any overseas option. Jeddle is the Australian entrant in this category, and its marking engine is called JeddAI. It drafts against the rubric, success criteria and comment banks you enter, it is built around the Australian Curriculum and senior frameworks including the HSC, VCE and QCE, and the teacher confirms every mark and comment before anything reaches a student.

The bar any of these tools has to clear is expert essay marking: a mark that holds up against a published guideline, and a comment specific enough that the student knows what to do next. Run the same trial on Jeddle that you would run on anything else. Use an assessment you have already marked, keep your own marks hidden, and count how many drafts you would release unedited.

The honest conclusion is unglamorous. No tool here is better in the abstract. The one that fits is whichever needs the least repair on your subject, your task types and your criteria, and that your school will approve. Test more than one and keep what survives your own scripts.