Marking a pile of essays is several jobs done in one sitting: turning the brief into rules, judging each paper against them, checking the form, looking up the references, and deciding whether a matching passage needs a second look. Tools for marking essays can take some of those jobs off the desk. None of them should take the mark.
This guide, published by EdCitation, sets out each of its tools by the job it does, in marking order, with plan and price. It is written for instructors, professors, tutors and writing centres, and sets each job beside what universities' own guidance says, read on 23 September 2026, including what must stay with the marker.
The brief and the references cost nothing to check: Check your paper and Verify references need no account.
What tools for marking essays does EdCitation offer?
EdCitation has a tool for each checking job in marking, from the brief to the reference list, and none of them gives a grade. The table runs in marking order, then a course's jobs, then a department's licence.
| Marking job | EdCitation tool | Plan | Price |
|---|---|---|---|
| The brief or rubric, read into a checklist | Check your paper | Free | No account needed |
| The paper read against its brief, with quotations | Theoretics QA | Max | $24 a month |
| Formatting, and what was left in the file | Mechanics QA | Pro | $8 a month, or three checks a month with a free account |
| Every reference looked up, retractions flagged | Verify references | Free | No account needed |
| Every reference in an uploaded paper, every citation matched | References from a file | Pro | $8 a month, or three checks a month with a free account |
| Passages matching published or web text | Check plagiarism | Pro | On credits: 240 a month with Pro, 720 with Max, more from $10 |
| Passages labelled AI-written, AI-assisted or human | Check AI | Pro | On the same credits |
| A reading list or reference list shared with a class | The Team page and reference lists | Free account | Nothing |
| Books by discipline, to read or borrow | Library | Max | $24 a month |
| Every student of a programme covered | Institution licence | Institution | One invoice |
The checks on credits show their cost first, use nothing if they fail, and report only to whoever ran them. The plans are compared on the pricing page.
What must a marker never hand to a tool?
The first reading of the paper, the mark, and any decision that a student has plagiarised or cheated. Read together, the guidance of LSE, UNSW and Reading keeps these with the marker.
The London School of Economics says it most plainly. Its principles for AI-assisted marking, approved in November 2025 for a pilot year, forbid marks or feedback generated by AI with little or no human involvement, apart from long-established automatic marking of formats such as multiple-choice quizzes (Gordon & Walker, 2025). AI tools, they add, cannot decide marks, replace the first marker's judgement, stand in for second markers or moderators, or replace calibration with colleagues, and educators do the primary reading themselves.
UNSW (2025) tells staff that, as a rule, "markers must not use AI platforms for marking or grading", and that an AI-writing flag only prompts an investigation, in which markers rely on their own judgement. The University of Reading's guide to similarity reports agrees: whether matched text is plagiarism "remains your academic judgment" (Moll, 2026).
EdCitation is built to that line. None of its tools gives a grade or writes a word of a student's paper; each returns places for the marker to read, not a verdict.
In what order should I check a submission?
Rules first, then your own reading, then form and sources, and the plagiarism and AI reports last, so no percentage frames the first reading.
- Read the brief into a checklist, once for the whole pile. Paste the brief or rubric into Check your paper, free, and untick anything misread.
- Read the paper against it yourself. This is the primary reading LSE keeps with the educator. With Max, Theoretics QA then quotes the passage for each thing the brief asks.
- Check the form. Mechanics QA, in Pro, measures formatting and finds what was left in the file.
- Check every source. Verify references does it free; References from a file, in Pro, works from the uploaded paper.
- Look at matching text and AI estimates, where your rules allow. Check plagiarism and Check AI, in Pro, show where to look again. Neither is proof.
- Decide the mark and write the feedback. Keep the checklist and the passages you relied on together, for the moderator and any appeal.
How do I check a paper against its own brief?
Split the brief into rules a paper meets or misses and criteria that need judgement, tick the rules quickly, and spend the reading on the criteria. Checking submissions against the assignment brief works this through on a real MIT brief.
Check your paper: the rules, free
Give Check your paper the task sheet, the marking grid, or the two together, and it lists what they require: section word limits, headings, the number and age of sources, the referencing style, and the page set-up. Each requirement carries the words of the brief it was taken from, and a line you untick is left out of the check. One list then serves every script and every marker on a team.
Theoretics QA: the judgement, quoted
Theoretics QA, part of Max at $24 a month and free for the first 14 days, takes the instructions and the paper as two uploads and returns one report. The confirmed checklist comes back rule by rule, met or not met. Then, for each task the brief sets, a reader judges whether the paper carries it out, quotes the passage behind that view, and names what is absent. It never writes, rewrites or paraphrases a word, and gives no grade.
For a marker, that report is a list of passages to read first, never a conclusion. No Theoretics QA result appears here, because none could be run for this guide. The student's side of the same check is does my paper answer the assignment question.
How do I check formatting and what was left in the file?
By rule, and with no more reading time than the rubric gives it marks. Mechanics QA does by rule the read a marker would otherwise do by eye: layout, headings and reference list against the APA 7 student paper; comments, tracked changes and template text left in the file; what the file records about itself; citations against the list; figures that disagree; and slips a spell-checker misses. No model reads the paper, and each finding cites the rule broken and the fix.
Universities already treat mechanics apart from judgement: LSE lets staff use spelling, grammar and proofreading tools freely when preparing feedback, and does not count them as AI-assisted marking, because they concern mechanical accuracy rather than academic judgement (Gordon & Walker, 2025).
The formatting rules are APA 7's, so on an MLA, Harvard or IEEE paper use Mechanics QA for the leftovers, citations and numbers. The student's view is in the Mechanics QA guide.
What is the best citation checker for teachers?
The best citation checker for teachers checks each reference against the publisher's record and never writes one, and EdCitation's Verify references does exactly that, free and with no account. It matters because a reference claims that a source exists and was used: Northeastern University (n.d.) defines fabrication as the falsification, misrepresentation or invention of any information, data or citation, inventing sources included.
UNSW (2025) tells markers to check that references exist, because AI-written ones are "likely to be plausible but fake". Plausible is the whole difficulty: a chatbot asked about a reference writes more plausible text, with no record to consult, while this checker can report only what the record holds. A pasted list returns one verdict per entry: verified, "check this" when something is doubtful, or not found. Retractions are flagged, and an index that did not answer is reported as such, never as not found. Upload the whole paper and citations are matched to entries, and entries back to citations.
References from a file, in Pro, reads the list out of a Word file or PDF, says where each mismatch happens, and suggests published sources in place of a reference that cannot stand, which gives feedback something concrete to point to.
"Not found" opens a question; it does not settle one. A report, a thesis or an older book may simply be missing from the indexes, so ask the student for the source. The fair way to handle that conversation is in how to spot fabricated references in student work, and a routine for a whole pile in marking reference lists faster.
A worked example: one paper, checked in order
On 23 September 2026 we ran Mechanics QA and the reference check on EdCitation's own sample paper, a four-page APA 7 nursing essay on staffing with 221 words of body text and five references, uploaded as a Word file with no brief.
| Area | What came back |
|---|---|
| Formatting | Nothing to fix: Times New Roman, 12 point, double-spaced |
| Left in the file | No comments, tracked changes or template text |
| File record | A note of the author's name, six revisions and 212 minutes of editing |
| References | Four verified, one not found, none retracted, none left unchecked |
| Citations | Five in-text citations, five entries, none orphaned or uncited |
| Numbers | Nothing found |
| Language | One 46-word sentence, noted as long but readable |
Mechanics QA placed the paper in its "start with the must-fix items" band for one reason: the fifth reference, a 2021 workforce analysis, has no record with any publisher, registry or library.
What a marker takes from it
The missing reference carries the essay's opening claim, that the nursing shortage will deepen, so that is where the reading starts and where the question to the student goes. The report counts references it could not check separately, zero here, so an index that failed to answer never swells the not-found column. And the file record is a fact about the file, not about how the essay was written: like a detector score, a reason to ask and never a finding.
How should I read plagiarism and AI reports?
As a reason to look again, never as proof. A similarity figure measures matching text, not plagiarism, and the guidance read here tells staff not to rely on AI scores as evidence at all.
Matching text
The University of Reading's staff guide says there is no ideal similarity percentage, that even a 1% score could hide plagiarism, and that reports miss images and diagrams, print books and journals, translated work and password-protected pages (Moll, 2026). A low figure is not clearance, and a high one may be cited quotations and a reference list. Check plagiarism, in Pro, gives the share of matching text, marks each matched passage where it sits and links its source, so the question narrows to whether that passage is quoted and cited properly. How plagiarism checkers work explains the figure.
AI estimates
The University of Cambridge (n.d.) advises staff not to rely on AI detection software, as unproven and offering "no evidence" to support an investigation, and LSE tells staff not to use AI tools to detect students' AI use at all. The research agrees: testing fourteen detectors, two of them commercial, Weber-Wulff et al. (2023) concluded that they were neither accurate nor reliable, leaned towards calling text human-written, and did worse once text was disguised.
Check AI, in Pro, sorts a paper's passages into three labels, AI-written, AI-assisted and human, each with its score, and its own page warns that another detector may read the text differently and that no score is proof of anything. For a marker, it shows which passages to reread and ask about, and must never be used on its own as evidence of misconduct. Academic integrity policy wording has the clauses and a fair process.
How do I share a reading list and books with a class?
Build the list in EdCitation and share it from the Team page. Gather readings in Find sources, whose index holds about 300 million published works, and let Cite a source set each in the course's style from the publisher's record. Both lists live on your free account, where a reference can be tagged by week or topic, filtered by tag, given its own style, and copied out with its italics.
On the Team page you invite the class by email; each student creates a free account to open what you shared. A given list helps at marking time too: UNSW (2025) notes that markers readily notice references that match neither the reading list nor the text. For the class session on checking sources, a 40-minute lesson plan is ready to teach.
For set books, the Library narrows a catalogue of textbooks and scholarly books to one of 31 disciplines, then by kind, language and year. Max opens more than 5 million books to read or borrow, a collection EdCitation compiled, and any title can also be bought, traced to a nearby library, or kept in My Books.
Can I upload student papers to a checking tool?
Only as your institution's rules allow: ask your data protection or integrity office first, for EdCitation as for any other service. The rules differ. Cambridge does not permit examiners to upload, copy or share student work with generative AI tools without the student's permission. UNSW authorises one platform for flagging AI writing and tells staff to upload student work to no other. LSE requires approved tools, and explicit permission before student work goes into AI tools.
A brief, a rubric and a reading list hold no student's writing, so Check your paper and the Team page raise none of these questions.
What the Institution licence changes
For a department, the Institution licence puts the decision where these policies place it, with the institution rather than each marker. It covers every student of a programme, with integration into the LMS and sign-in, white label under the university's own name, training and one invoice. Students can then check their own drafts with the same tools before submitting, when a missed rule or a missing source costs least.
Quick questions
Will EdCitation grade student papers?
No. Its findings are places to read, rules met or missed and references found or not, never a grade; LSE, for one, keeps marking decisions with people.
What is the best free citation checker for teachers?
EdCitation's Verify references, with no account and no charge. It checks a student's whole list against the publisher's record in one pass and flags any retracted paper.
Is an AI detector score proof that a student used AI?
No. Cambridge advises staff not to rely on AI detection software, and a 2023 test of 14 detectors found them neither accurate nor reliable.
Can I upload student papers to an outside checking tool?
Only as your institution allows. Cambridge does not permit examiners to share student work with generative AI tools without the student's permission, so ask before uploading.
Which EdCitation plan does an instructor need?
For the brief and the references, none: both are free. The four checks that read an uploaded paper come with Pro, $8 a month, and Theoretics QA and the Library with Max, $24 a month.
References
- Gordon, C., & Walker, S. (2025). LSE principles and guidance for AI-assisted marking and feedback [PDF]. Eden Centre, London School of Economics and Political Science. https://info.lse.ac.uk/staff/divisions/Eden-Centre/Assets-EC/Documents/AI-assisted-marking-Nov-2025/AI-assisted-marking-and-feedback-revised-post-EC-Nov-2025.pdf
- Moll, P. (2026, July 13). Turnitin: A staff guide to interpreting the Similarity Report. Technology Enhanced Learning Support for Staff, University of Reading. https://sites.reading.ac.uk/TEL-Support-Staff/turnitin-a-staff-guide-to-interpreting-the-similarity-report/
- Northeastern University. (n.d.). Academic integrity policy. Office of Student Conduct and Conflict Resolution. https://osccr.sites.northeastern.edu/academic-integrity-policy/
- University of Cambridge. (n.d.). Generative AI and assessment. Blended Learning Service. https://blendedlearning.cam.ac.uk/artificial-intelligence-and-education/generative-ai-and-assessment
- UNSW Sydney. (2025, November 12). Guidance on AI in assessment. UNSW Staff Teaching Gateway. https://www.teaching.unsw.edu.au/ai/ai-assessment-guidance
- Weber-Wulff, D., Anohina-Naumeca, A., Bjelobaba, S., Foltýnek, T., Guerrero-Dib, J., Popoola, O., Šigut, P., & Waddington, L. (2023). Testing of detection tools for AI-generated text. International Journal for Educational Integrity, 19, Article 26. https://doi.org/10.1007/s40979-023-00146-z