Spotting Copying Between Learners: A Practical Guide to Internal Similarity Checking for UK Schools and Training Providers

By Tomas Lerpiniere · Updated 2026-10-21

Unauthorised copying between learners, known in UK assessment guidance as collusion, is distinct from web plagiarism and generative AI use. An organisation-only similarity bank lets schools, MATs, colleges, and Independent Training Providers (ITPs) compare submissions inside their own institution after marking, without sending work to a global commercial database.

In this guide

Feedback Flows tools in this guide

Similarity Scoring Writing fingerprints AI writing detector Reference checker

Every experienced teacher, curriculum lead, and Internal Quality Assurer (IQA) recognises the feeling of déjà vu during marking. You read a distinctively phrased paragraph in a learner’s essay or vocational portfolio, pause, and realise you read that exact sentence earlier in the week. If both scripts sit in the same pile on your desk, you can ruffle back through the pages to find the match.

In modern UK education settings, assessment is rarely confined to a single desk. When three teachers mark a Year 10 History cohort across six teaching groups, when a Multi-Academy Trust (MAT) moderates common assessments across four secondary schools, or when an ITP assesses apprentices on rolling monthly enrolment dates, human memory cannot bridge the gap. One learner shares a completed draft in a group chat to “help a mate see how to structure it,” and the duplicate submission slips through because two different assessors mark the scripts three weeks apart.

For years, sector conversations about digital integrity tools have focused on large web-crawling plagiarism suites designed for universities, or standalone AI detectors. Yet for schools, colleges, and training providers, peer-to-peer copying inside the institution remains one of the most frequent, tangible threats to fair assessment. This guide explains how internal collusion differs from external plagiarism and AI generation, why similarity scores must never be treated as automatic proof of cheating, how to eliminate false positives, and how to run a fair, UK GDPR-compliant investigation workflow across your organisation.

1. Defining the boundaries: plagiarism, collusion, AI generation, and self-plagiarism

Clear policy starts with precise language. In UK secondary, further education, and work-based learning contexts shaped by the Joint Council for Qualifications (JCQ) and the Quality Assurance Agency for Higher Education (QAA), different forms of text overlap require different diagnostic tools and pastoral responses.

Integrity concept Core definition Origin of the text Primary detection method
Plagiarism Presenting another person’s published or external work as one’s own without appropriate acknowledgement External websites, textbooks, revision guides, journals, essay mills External citation checking and subject expertise
Collusion Unauthorised co-operation between two or more learners on work submitted as an individual effort Another learner inside the same school, college, MAT, or training provider Organisation-only similarity bank and side-by-side script comparison
AI generation Submitting text produced wholly or partially by a large language model without authorisation or declaration where required Generative AI tools prompted by the learner AI writing indicators paired with longitudinal writing fingerprints
Self-plagiarism Resubmitting one’s own previously assessed work for a new unit without permission or citation The same learner’s earlier submission in the institution’s archive Internal similarity bank matching against the learner’s prior submissions

Unlike external plagiarism, peer collusion often begins with positive intentions. Schools and training providers encourage collaborative learning: students revise together, discuss case studies in pairs, and share peer feedback. Collusion occurs when learners cross the line from shared preparation into shared production on an individual assessment.

Research into student academic integrity consistently shows that many learners, particularly in Key Stage 4, Key Stage 5, and entry-level apprenticeships, struggle to distinguish between discussing how to answer a question and sharing the written answer itself (Park, 2003; Bretag et al., 2019). JCQ guidance on Suspected Malpractice notes that a learner who knowingly allows their work to be copied by another candidate may also be party to malpractice, even if they wrote the original script themselves (JCQ, 2024). Educating learners on this distinction is just as important as detecting the overlap.

2. Why organisations need internal comparison after marking

Teachers are exceptional at noticing sudden shifts in an individual pupil’s voice, but structural blind spots make manual peer comparison impossible at scale:

  • Team-taught cohorts and split classes. Five teachers may mark the same Year 11 mock or Non-Exam Assessment (NEA) component. If Pupil A in Set 1 shares an essay with Pupil B in Set 4, neither teacher ever sees both scripts.
  • Cross-site MAT assessments. When a Multi-Academy Trust runs common summative assessments across academies, friends at neighbouring schools frequently share resources via social media.
  • Rolling cohorts in apprenticeships and FE. ITPs and colleges rarely run a single cohort deadline. Learner A completes a portfolio in March; Learner B at the same employer submits an almost identical rationale in October to a different assessor.
  • Resubmissions and retakes. Assessors need visibility over whether a revised submission borrows heavily from a classmate who passed on the first attempt.

Running an automated internal similarity check after or alongside marking closes these blind spots. An organisation-only similarity scoring engine indexes submissions against other work inside your institution’s boundary, surfacing cross-class and cross-cohort overlaps without requiring teachers to manually cross-reference hundreds of documents. Want to test two individual scripts right now without logging in? Try our free collusion and text similarity checker.

3. What a similarity percentage is, and what it is not

Every teacher, assessor, and quality lead must understand one foundational rule: a similarity percentage is a measurement of text overlap, not an automated finding of academic misconduct.

At a conceptual level, text comparison systems typically:

  1. Normalise formatting so layout changes cannot disguise identical wording.
  2. Tokenise and shingle the text into overlapping word sequences (n-grams) and compare them against other scripts in your organisation’s internal bank.
  3. Align passages where sequences match, even if paragraphs are reordered or minor words swapped, then calculate the percentage of the document covered by overlapping text.
  4. Apply exclusion filtering for expected matches such as embedded question prompts, quotation marks, and standard reference lists.

Consider two scripts that both return a 35% similarity score:

  • Script 1 (35% overlap): The match consists of assessment questions pasted at the top of each page, three required statutory definitions, and a standard bibliography. Verdict: zero integrity concern.
  • Script 2 (35% overlap): The middle 450 words of personal evaluation and data analysis are word-for-word identical to a classmate’s submission, with three adjectives changed. Verdict: serious collusion concern requiring investigation.

Academic literature on text-reuse tools repeatedly warns against “management by number,” where educators treat an arbitrary score as proof of guilt without inspecting the underlying text (Weber-Wulff, 2019). A match simply says: these two pieces of work share wording; a professional should look at them side by side.

4. Five false-positive traps (and how smart exclusions help)

When teachers lose trust in similarity tools, it is almost always because poorly configured systems flood them with “false highs” caused by legitimate assessment design:

  1. Short-answer questions and restricted responses. There are only so many accurate ways to define osmosis or state a formula. Short factual responses naturally overlap across high-attaining learners.
  2. Prescribed quotations and statutory texts. Twenty students writing on Macbeth will all quote the same lines. Early Years or Health and Social Care learners must cite exact wording from statutory frameworks.
  3. Question stems, rubrics, and mark scheme phrasing. When digital workbooks include tasks and grading grids inside the document, two modest-length submissions can show 40% to 60% similarity before the learner has typed a single original sentence.
  4. Shared scaffolds, writing frames, and vocational templates. If a class of thirty students uses the same teacher-issued PEE/PEEL frame or laboratory report template, raw text matching will flag every student against the rest of the class.
  5. Reference lists, bibliographies, and standard declarations. When a whole cohort uses the same teacher-recommended reading list, bibliography sections match almost entirely. Pair internal similarity checking with a dedicated citation and reference checker so bibliography verification is handled separately from body-text comparison.

To protect staff workload, internal similarity systems should apply layered exclusions, either automatically or via simple toggles during review:

  • Template and question exclusion (base worksheet or assignment brief)
  • Quoted material exclusion (quotation marks or block quotes)
  • Bibliography exclusion (References, Bibliography, Works Cited)
  • Minimum word-run thresholds (ignore isolated three- or four-word phrases)

5. Policy design: using thresholds (such as 50%+) as prompts, not verdicts

There is no universal mathematical threshold that separates honest work from collusion. However, organisations need practical triage rules so teachers are not overwhelmed during heavy marking windows. Many UK schools and training providers adopt a review threshold, such as 50% or above after standard exclusions, as an administrative prompt to inspect the scripts side by side.

When writing your institution’s policy, frame thresholds around three safeguards:

  • A threshold triggers a look, never a sanction. A score of 50% or above means a teacher or assessor will spend two minutes reviewing the highlighted passages side by side. It must never trigger an automatic zero mark, an automated disciplinary letter, or an instant malpractice log.
  • Calibrate by task type. A 50% review prompt works well for structured assignments and modular portfolios. For open-ended 2,000-word coursework essays with no shared template, departments may inspect contiguous matching blocks at a lower overall percentage (for example, 25% to 30% if concentrated in single paragraphs).
  • Empower teacher judgement below the threshold. Even if an organisational flag is set at 50% to manage workload, any teacher who notices an identical, unusual paragraph in a 20%-match script can still open the side-by-side comparison and investigate.

Spot copying across classes and cohorts, without sending learner work outside your organisation

Our organisation-only similarity bank compares scripts strictly within your school, trust, or training provider after marking. Teachers see clear side-by-side comparisons for the classes they teach, while quality leads get a calm governance view. No global data sharing, and never an automatic misconduct verdict.

Explore Similarity Scoring · Read our data commitments

6. Real-world practice: two UK sector vignettes

Vignette 1: Secondary coursework across teaching groups

At a large mixed comprehensive, 90 Year 12 students submit their first full draft of the A-Level History NEA across four classes taught by three teachers, managed in the department’s class management workspace. After initial marking, the internal similarity bank flags a 54% match between Maya (Group A) and Liam (Group C).

The Head of History and Ms Patel open the side-by-side comparison. Toggling exclusions for quoted primary sources and the bibliography drops the score from 54% to 34%. The remaining highlighted text shows two entire paragraphs evaluating historians’ interpretations are identical in sentence structure, differing only where synonyms have been swapped.

Because the drafts were caught during formative review rather than after final JCQ authentication declarations were signed, the Head of Department meets Maya and Liam separately in a calm, non-accusatory setting. Liam explains that he was struggling and Maya sent him a photo of her draft to show the layout. Both misunderstood the boundary between peer support and collusion. The teacher records the flag as Concern, provides guidance on independent redrafting, and briefs all four classes on why sharing draft photos breaches NEA rules.

Vignette 2: Apprenticeship rolling cohorts in an ITP

A national Independent Training Provider delivers the Level 3 Business Administrator apprenticeship across multiple employers. Callum uploads an 1,800-word reflective account. The organisation-only similarity bank flags a 62% match against Priya, an apprentice at the same employer who completed four months earlier under a different skills coach.

Callum’s skills coach and the Lead IQA review both scripts side by side. About 30% of the overlap comes from the employer’s standard project charter template and company background. The remaining 32% sits inside the “Personal Reflection and Lessons Learned” section, where Callum describes overcoming a stakeholder communication breakdown using Priya’s exact words and sequence of events.

In a 1-to-1 progress review, Callum admits that his workplace line manager gave him Priya’s old portfolio folder as an “example of what the training provider wants.” The IQA logs the outcome as Concern, requires a fresh reflective account verified by a professional discussion recording, and contacts the employer mentor to stop past learner portfolios circulating as templates. For provider-specific cohort workflows, see Feedback Flows for Independent Training Providers.

7. The fair investigation playbook: a four-step workflow

When an internal similarity check surfaces a high overlap, staff need a consistent, calm procedure that protects both the integrity of the qualification and the wellbeing of the learner.

1Automated flag and exclusionsVerify that question text, templates, quotations, and reference lists have been excluded before escalating.
2Side-by-side human reviewCheck location of matches, structural mirroring, shared anomalies, and writing fingerprints against prior classwork.
3Exploratory conversationLead with curiosity, test comprehension verbally, and check draft history where appropriate.
4Auditable outcome logClose every flag as Dismissed, No concern, or Concern in the reporting hub.

In side-by-side review, look specifically for:

  • Location of matches: definitions and methodology versus personal analysis, evaluation, and reflection
  • Structural mirroring: the same order of arguments and paragraph transitions even where words have been altered
  • Shared anomalies: the same unusual spelling mistake, misquoted statistic, or formatting quirk
  • Submission timestamps and authorship baselines: who submitted first, and whose natural style matches the shared passage

Align investigation practice with QAA academic integrity guidance: distinguish developmental academic practice errors from deliberate misconduct, and keep investigation workflows fair and transparent.

8. Role design: what classroom teachers need vs what leaders and quality teams need

User role What they need to see Primary action
Classroom teacher / skills coach / assessor Similarity flags for the classes or caseload they mark; immediate side-by-side view; exclusion toggles and outcome logging Inspect flagged scripts, speak with learners, apply formative feedback or escalate
Head of department / curriculum lead Department-wide patterns across teaching sets; unresolved flags before internal moderation Coordinate cross-class conversations; spot tasks where over-scaffolding caused false positives
MAT quality director / Lead IQA / exams officer High-level governance view via the reporting hub; trend data by cohort, site, or qualification; audit logs Ensure policy consistency across Multi-Academy Trusts or Independent Training Providers; verify no learner receives a sanction without documented human review

9. Privacy, governance, and UK GDPR in organisation-only banks

Historically, many UK schools and training providers have hesitated to use similarity checkers because legacy university platforms required uploading students’ intellectual property into a permanent global database. An organisation-only similarity bank operates on a different privacy model. Review our full data commitments for technical governance details.

  • Strict institutional boundary. Scripts are compared only against other scripts within your school, college, MAT, or training provider. Learner work is never pooled into a commercial cross-customer repository, never sold, and never used to train external generative AI models.
  • Lawful basis under UK GDPR. Centres typically rely on Public Task (Article 6(1)(e) for state-funded schools and colleges) or Legitimate Interests (Article 6(1)(f) for independent schools and private training providers) to maintain qualification standards.
  • Compliance with UK GDPR Article 22. ICO guidance on automated decision-making restricts solely automated decisions that produce legal or similarly significant effects. By designing your workflow so a similarity percentage is strictly a triage flag for human review, your institution ensures meaningful human involvement at every stage.
  • Learner visibility and transparency. Learners and parents should be informed via your assessment policy and privacy notice that work is checked internally against cohort submissions. Learners should never browse other students’ scripts directly; side-by-side views stay with authorised staff.
  • Proportionate retention. Retain scripts only for as long as needed to protect cohort integrity and meet awarding body retention schedules, then follow your standard data deletion schedule.

10. A layered integrity approach: similarity, writing fingerprints, and AI indicators

No single metric can capture every aspect of academic integrity. Forward-thinking schools and training providers combine three distinct signals:

  • Internal Similarity Scoring: identifies text overlap between learners inside your organisation (peer collusion and template overuse) or against a learner’s own prior modules (self-plagiarism).
  • Longitudinal Writing Fingerprints: compares a new submission against the same learner’s historical baseline. If two students share a paragraph, fingerprints help teachers see which student’s natural voice matches the disputed text.
  • AI Writing Detector indicators: evaluate linguistic predictability and structural patterns associated with large language models. When five students paste the same assignment prompt into ChatGPT without modifying the output, essays frequently flag for both AI generation patterns and moderate peer-to-peer phrase similarity.

As we explore in our companion guide on evidence over AI detection scores, combining these layers prevents false accusations and equips teachers with rounded, factual context before speaking with a learner.

11. Pastoral care and recording for moderation

Discovering that two learners have copied from one another can feel frustrating, but how the institution responds defines its culture. Never shame learners publicly. Research demonstrates that last-minute collusion frequently stems from anxiety, poor time management, fear of failure, or peer pressure to help a struggling friend (McCabe et al., 2012). Involving pastoral leads or apprenticeship welfare officers alongside curriculum staff ensures underlying pressures are addressed.

When copying occurs in routine homework, class essays, or early coursework drafts, treat it as a teachable curriculum moment while logging the internal Concern so repeat patterns are visible. When unauthorised collusion occurs in final authenticated NEA submissions or end-point portfolio gateway evidence, follow your centre’s formal JCQ or awarding organisation malpractice reporting procedures, supported by your documented side-by-side audit trail.

Frequently asked questions

Plagiarism occurs when a learner submits work taken from an external source such as a website, book, or journal without proper acknowledgement. Collusion happens when two or more learners work together on an assessment intended to be completed independently, resulting in shared passages, identical structures, or copied answers across their submissions.

No. A similarity percentage only measures the proportion of overlapping text between two documents; it never proves intent or misconduct. High scores frequently occur when learners include the same assessment prompts, prescribed quotations, reference lists, or teacher-provided writing frames. Every flag requires human side-by-side review before drawing conclusions.

Many schools and training providers use a 50% similarity threshold as an administrative trigger to open a side-by-side review, rather than an automatic sanction. Whether 50% is concerning depends on the task: a short worksheet with embedded questions naturally generates high overlap, whereas 50% unquoted overlap in an open essay warrants a conversation.

Yes. An organisation-only similarity bank compares scripts strictly against other submissions inside your own school, college, Multi-Academy Trust, or training provider. This identifies cross-class and cross-cohort peer copying while keeping learner data inside your organisation's governance boundary under UK GDPR.

A web plagiarism checker crawls external internet pages, academic journals, and global student repositories to see if a learner copied from outside sources. An organisation similarity bank compares submissions exclusively against other learners' work inside your own school, college, MAT, or training provider to spot peer-to-peer collusion and cross-cohort copying without exposing student data externally.

Not necessarily. First, the overlap could be caused by shared question prompts, set quotations, or a teacher-provided writing frame. Second, even if genuine copying occurred, one learner may have written the work independently and shared it in good faith, or had their work accessed without permission. Only a side-by-side human review and separate conversations with both learners can establish what happened.

Use smart exclusion settings to strip out embedded assignment questions, standard cover sheets, quoted material, and bibliographies before reviewing the similarity score. If your department uses a heavily scaffolded writing frame, compare the length of contiguous matching learner-written sentences rather than relying on the raw overall percentage.

Yes, within your organisation's assessment boundary. Peer copying frequently happens across friendship groups taught by different teachers or assessed by different skills coaches. Surfacing cross-class matches allows the two marking teachers or the curriculum lead to compare the scripts side by side and coordinate a fair response.

Yes, provided learner work remains inside your organisation's boundary, retention aligns with your data protection policy, learners are informed via your privacy notice, and similarity scores are never used to make solely automated disciplinary decisions without human review (in line with UK GDPR Article 22).

Internal similarity scoring identifies peer collusion and template overuse. Writing fingerprints compare a new submission against the same learner's historical baseline. AI writing indicators evaluate patterns associated with large language models. Combining these layers prevents false accusations and gives teachers rounded context before speaking with a learner.

Conclusion: fairness through visibility, not surveillance

Spotting copying between learners should never feel like running a courtroom inside your staffroom. Most peer collusion in schools, colleges, and training providers is opportunistic or born from confusion over where group revision ends and independent writing begins.

By pairing an organisation-only similarity bank with smart exclusions, sensible policy thresholds, and a structured four-step review playbook, educators gain the visibility they need across classes and cohorts. When teachers can see overlapping work side by side after marking, and leaders have a clear governance view of how flags are resolved, academic integrity becomes calmer, fairer, and consistently defensible for every learner.

Continue exploring

Feedback Flows tools for this workflow

Integrity Similarity Scoring Every marked rubric and automarking submission can enter your organisation's private similarity bank. See a similarity %, flag work at 50% o...
Integrity Writing fingerprints Build a style profile from previous authentic samples, then compare a new submission. See how vocabulary, structure and habits align — evide...
Integrity AI writing detector A hybrid check with writer context for EAL and younger writers, highlighted passages you can quote exactly, and a place to record what you d...
Integrity Reference checker Check the whole submission — not just the reference list. Match in-text citations to the bibliography, verify bibliographic details, and opt...

Build a fairer, layered approach to academic integrity

Give your staff the visibility to review overlapping work side by side, exclude false positives, and log clear outcomes (No concern, Concern, or Dismissed) for moderation and IQA. Combine internal similarity checking with longitudinal writing fingerprints and AI indicators inside one UK GDPR-compliant workspace.

For Multi-Academy Trusts · For Training Providers · Try the live demo

Selected references