Six tools here give useful feedback on student writing, and the thing that decides between them is not feedback quality. It is where the writing already lives, and who reads the machine's comments first. Brisk Teaching is the strongest free choice for most teachers because it works inside the Google Doc or Word file a student already has open. Olex.AI is the one for an English department that needs a class set of thirty graded in one pass. Grade Direct is the only tool on this page that publishes a dollar price you can budget against without a sales call.
Compare the six at a glance
Prices and free-tier limits below were checked against each vendor's own page on September 18, 2026.
| Tool | The job it does best | Free tier | Paid price, checked September 18, 2026 | Where the work sits |
|---|---|---|---|---|
| Brisk Teaching | Feedback inside the document a student is already writing in | Yes: $0, unlimited use, includes rubric-criteria feedback (plans) | Premium and Intelligence: custom quote, no public figure (plans) | Our directory holds no hosting or compliance record for it |
| Class Companion | Feedback a student acts on mid-draft, with hints rather than answers | Yes: free teacher plan with instant AI feedback (plans) | School and district: book a demo, no public figure (plans) | Vendor lists FERPA compliance as a free-plan feature |
| EnlightenAI | Teaching the AI your rubric using your own anchor samples | Yes: free to start (essay grader) | No published price tiers | Directory record: hosted in the UK, ages 8+, not shared with third parties |
| Olex.AI | Grading thirty English essays at once, handwriting included | No | Quote-based on school size (FAQs) | States student work is never used to train its models (FAQs) |
| Grade Direct (listed as Graded Pro) | Written responses across STEM and humanities on a small budget | Yes: 150 credits, then 50 a month (pricing) | Teacher Pro $25 for 6,000 credits (pricing) | Directory record: hosted in the UK, ages 4+, shares data with third parties |
| Grade It AI | Grading against a named exam board's mark scheme | No, demo only | Starter $129 a year, Pro $249 a year (pricing) | Vendor states compliance with major regulations; silent on training |
Four questions that pick the tool for you
Work down these in order. The first one usually settles it.
Where does the writing already live? If your students draft in Google Docs or Word, a tool that reads those files directly will get used, and a tool that needs uploading will not. This is the single best predictor of whether a feedback tool survives past October.
Who reads the feedback first, the student or you? Two different products hide under one label. Class Companion and Brisk's student extension put comments in front of the student while the draft is still warm. Olex.AI and Grade It AI put a graded set in front of you, and nothing reaches a student until you release it. Pick the one that matches what you actually want to happen.
Does the tool learn your rubric, or bring its own? EnlightenAI is built around calibration: you supply the rubric and anchor samples and it works to those. Most others accept a rubric but weight it their own way. If your department has agreed criteria, calibration is worth paying for.
What happens to the writing afterward? Only Olex.AI, among these six, states outright that student submissions are never used to train or improve its models (FAQs). For the rest you are reading privacy policies, and that is a conversation for whoever signs data agreements in your district, not a box any blog can check for you.
Browse the writing, feedback and academic integrity category if you want to see the full set alongside these six.
Brisk Teaching: the free option that meets students where they write
Brisk Teaching is the best fit for a teacher who wants to start on Monday without a purchase order, because it is a Chrome and Edge extension that layers onto documents that already exist rather than asking anyone to move. It carries a 4.77 rating from 35 reviews in our directory, the largest review count of any tool in this comparison.
The free plan includes three named feedback modes, glow and grow, next steps, and rubric criteria, plus a separate student extension that gives writing feedback directly to the student (plans).
Watch out for: it lives in a browser extension, so a school on locked-down devices needs its IT team to allow it before any of this happens. The paid tiers carry no public figure, which makes a department budget hard to draft.
Pricing: free plan $0 with unlimited use; Premium and Intelligence are custom-quoted (plans), checked September 18, 2026.
Safety: our directory record carries no data-hosting location, no minimum age and no compliance detail for Brisk, so there is nothing here we can tell you. Before student work goes through it, ask the vendor where documents are processed, whether a district data agreement is on offer, and whether student text is retained.
Class Companion: feedback a student gets while the draft is still warm
Class Companion is the best fit for the teacher whose real complaint is that feedback arrives too late to matter, because its free teacher plan puts instant AI feedback in front of the student during the task rather than days later. Its tutor offers hints rather than finished sentences, which is the distinction that decides whether a feedback tool builds writers or replaces them.
Watch out for: the free plan is genuinely free but deliberately basic, and everything at school scale runs through a demo booking. Our directory holds a single review for it, so treat the rating as thin.
Pricing: free teacher plan, $0; school and district pricing is quote-based with no published figure (plans), checked September 18, 2026.
Safety: the vendor lists FERPA compliance as a feature of the free teacher plan (plans). That is the vendor's statement, not a verdict: FERPA obligations sit with your district, and what matters is whether your district will sign an agreement with this vendor for these students. Our directory record has no hosting location or minimum age for it.
EnlightenAI: the one that learns your rubric rather than bringing its own
EnlightenAI is the best fit for a department that has agreed criteria and wants the machine to apply them, because it is built around calibration: you supply the rubric and anchor samples, and it proposes scores you review before anything is released. Its own example of the feedback it produces is specific in the way good feedback is, naming the paragraph and the problem: "Your thesis clearly states a position, but your evidence in paragraph 2 focuses on a different historical period (1850s instead of 1920s)" (essay grader).
Watch out for: calibration is work. Anchor samples and a rubric have to exist before the tool earns its keep, which makes it a department decision rather than an individual one. It also carries no reviews in our directory yet.
Pricing: free to start, and no published price tiers on the vendor's site, so the cost of a school deployment is unverified. Checked September 18, 2026.
Safety: our directory record lists UK hosting, a minimum age of 8, and that data is not shared with third parties. The vendor publishes a data processing agreement; what your district needs from it is a conversation to have before you upload a class.
Olex.AI: thirty essays in one pass, handwriting included
Olex.AI is the best fit for an English department running frequent mock exams, because its SnapSnap feature "marks a whole class of 30 essays at the same time" and its reader accepts photographed handwritten work from a phone or tablet (features). It covers Key Stage 1 through Key Stage 4 and can be configured to different mark schemes (FAQs).
It is also the only tool in this comparison that answers the training question without being asked: "Olex.AI does not use any student work to train or improve its AI models. All student submissions remain securely stored within your school's account and are never shared, repurposed, or used for machine learning" (FAQs).
Watch out for: it is built to the UK curriculum, so a US teacher is adapting rather than adopting. There is no free tier at all, and no published price.
Pricing: quote-based on school size and deployment, from a department to a whole trust; no public figure (FAQs), checked September 18, 2026.
Safety: the vendor describes itself as fully GDPR-compliant, states the no-training position quoted above, and says schools can request deletion at any time (FAQs). GDPR is the relevant framework for UK and EU schools; a US district is asking a different set of questions and should ask them directly.
Grade Direct: the only published price on this page
Grade Direct, the platform our directory still lists under its former name Graded Pro, is the best fit for a teacher who wants to know the cost before a sales call, because it is the only tool here that publishes figures: a free tier of 150 credits plus 50 a month, and Teacher Pro at $25 for 6,000 credits that do not expire (pricing). It grades written responses across languages, humanities, STEM and the arts, and the teacher previews and edits the feedback, adding written or voice notes, before anything returns to a student (pricing).
Watch out for: the rename is live and incomplete. The graded.pro address now redirects to grade.direct, and our own directory entry still carries the old name and the old link. Credits are also an unfamiliar unit: budget by the work, roughly three credits for a short task and about thirty for a full exam paper.
Pricing: free tier $0 with 150 credits plus 50 a month; Teacher Pro $25 for 6,000 credits (pricing), checked September 18, 2026.
Safety: our directory record lists UK hosting, a minimum age of 4, and that it does share data with third parties, which is the one line in this table worth reading twice before a class of elementary writing goes through it. The vendor's site makes no statement about whether student work trains its models.
Grade It AI: grading against a named exam board's mark scheme
Grade It AI is the best fit for a secondary English teacher grading to a specific specification, because it takes your uploaded mark scheme and works to it across AQA, Pearson, Cambridge and others, accepting handwritten and typed work as photos or PDFs (site). The vendor's framing of the division of labor is the right one: "You review, adjust anything you want, and send."
Watch out for: it is scoped tightly to English literature and language at secondary level, so it is the wrong tool for anything else. Gradings are metered, so a teacher with heavy classes should count the year's essays before subscribing. Our directory record lists it as contact-for-pricing, which is now out of date.
Pricing: Starter $129 a year for 30 gradings, Pro $249 a year for 80 gradings, with a free demo but no free trial (site), checked September 18, 2026.
Safety: the vendor states it complies with major regulations and prioritizes data security, which is a claim rather than a detail. Our directory record lists EU hosting. It is silent on whether student work trains its models, so ask.
What the research actually says about AI feedback
Two 2026 studies are worth more to this decision than any vendor page.
On quality, a study of 979 authentic student responses across six STEM courses, with 472 students evaluating both versions, found that "the pedagogical quality of AI-generated feedback is, in practice, comparable to that of human-authored feedback," with only small effect sizes separating them. Human feedback carried more strategy guidance; AI feedback carried more metacognitive and motivational content. Neither was clean: roughly 13 percent of human feedback and 10 percent of AI feedback contained errors or ambiguities (Computers and Education: Artificial Intelligence, June 2026). The finding teachers should sit with is a different one from the same study: students' judgments "are largely influenced by their perceptions of the feedback provider's credibility rather than the actual quality of the feedback itself."
On risk, an experiment with 214 university teachers gave four groups the same medium-quality paper with different fabricated AI-detection reports. Where the report showed a high detection rate with red highlighting, mean quality ratings fell to 4.26 out of 10 against 6.83 for the same paper with a low detection rate and no highlighting; percentage scores fell from 73.30 to 51.61 (Frontiers in Psychology, July 7, 2026). The paper had not changed. Only the number next to it had.
That is the strongest practical argument on this page for a rule your department can adopt this term: keep detector output away from the person assigning the grade. None of the six tools above is an AI detector, which is part of why they belong in a feedback workflow and detectors do not.
It is also worth being straight about the survey figure this article used to lead with. In the Walton Family Foundation and Gallup study of 2,232 US K-12 teachers, fielded March and April 2025, the share saying AI improved the quality of their work ranged from 57 percent to 74 percent by task, and grading and feedback sat at the bottom of that range, at 57 percent (Gallup, June 24, 2025). Teachers rate AI's help with feedback lower than its help with anything else they use it for. Weekly users in the same study saved an estimated 5.9 hours a week across nine tasks, about six weeks a school year.
A worked example: grade 9 argumentative essay, two weeks
Here is how these tools fit a real unit, written as a plan rather than a promise. Thirty students, grade 9 English language arts, one argumentative essay of five paragraphs on a local issue.
Day 1 to 3, drafting. Students write in Google Docs. Brisk's student extension is available for glow-and-grow feedback on demand, and students are told plainly what it is: a first reader, not a grader. No teacher time is spent here.
Day 4, a single checkpoint. You run rubric-criteria feedback across the class set from inside the documents. You read the feedback, not the essays, and you are looking for one thing: the three or four criteria the class is failing together. That is tomorrow's mini-lesson and it is worth more than thirty individual comments.
Day 5, teach the gap. Whole-class instruction on whatever the checkpoint surfaced, usually evidence-to-claim linkage at this level.
Day 6 to 8, revision. Students revise with the feedback tool still available. The revision, not the first draft, is what you will grade.
Day 9 to 10, grading. Final drafts go through the tool once for a proposed score against your rubric. You review every one before release, and you overrule it freely: the 10 percent error rate in the research above is roughly three essays in this class. Your comments go on the two or three pieces where a human reader changes what a student does next.
The unit spends the AI on the mechanical pass and spends you on the judgment. That is the division worth defending. If your criteria are not written down yet, start with an AI rubric generator that applies your own standards, then bring the rubric here.
What about the big names?
Grammarly is a writing assistant rather than a feedback tool for teachers, and its Authorship feature is the part worth knowing: it shows an instructor whether text was typed, pasted or generated elsewhere, and Grammarly is explicit that this is not detection, because detectors are "in an arms race with an ever-evolving technology" (Grammarly).
Turnitin is an integrity and similarity platform that many districts already license. It is not a formative feedback tool, and buying it expecting one leads to disappointment.
MagicSchool includes class writing feedback on its free plan at $0, with Plus at $12.99 a month or $8.33 a month billed annually (pricing). If your school already runs MagicSchool, try its feedback tool before adding a sixth subscription.
ChatGPT will produce competent feedback from a pasted rubric and is free to try, but it keeps no class roster, no rubric library and no record of who has been told what, which is exactly the administrative layer the tools above exist to provide.
Where this leaves you
Start free and start where your students already write. For most teachers reading this that means Brisk Teaching this week, at no cost, on one class. If your department grades against a shared rubric, the calibration EnlightenAI is built around is the upgrade that pays. If you are marking a hundred handwritten mock papers, Olex.AI is the one built for that specific misery.
And keep the detector score out of the room where grades are decided. The research above found the same paper losing two and a half points out of ten because of a number printed beside it. Your judgment is the thing worth protecting, and it is the thing these tools are supposed to be freeing up.
If what you actually need is grading at scale rather than feedback on writing, that is a different shortlist: see AI grading tools for teachers. English teachers looking more broadly should start with AI tools for English teachers.
Dan Fitzpatrick is a Forbes contributor, bestselling author (The AI Classroom, AI for School Leaders, The AI Educator), and founder of The AI Educator. He has trained over 150,000 educators across 100+ countries and advises schools, districts, and government bodies on AI strategy. Prices and free-tier limits in this article were checked against each vendor's own page on September 18, 2026.
