Can SafeAssign Detect Chegg, Course Hero and PDFs? (2026)

Reviewed and updated August 2026 against Blackboard’s SafeAssign documentation and its Originality Report reference (both updated July 2026).

An earlier version of this page ended with advice about using paraphrasing tools to get around a similarity check. That has been removed. It was both outside what this site is for and, as of 2026, no longer accurate — the tooling in this category has moved on considerably.

What SafeAssign actually is

SafeAssign is Blackboard’s built-in text-matching service, now documented under Anthology. Blackboard’s own description is careful and worth taking literally: SafeAssign “compares submitted assignments against a set of academic papers to identify areas of overlap between the submitted assignment and existing works.”

Note what that sentence does not say. It does not say SafeAssign detects plagiarism, identifies cheating, or reaches a verdict. It finds overlap. Blackboard frames the purpose explicitly as being “effective as both a deterrent and an educational tool” — one that helps students “identify how to properly attribute sources rather than paraphrase.”

This is the same position every serious vendor in this category takes. Turnitin says it “does not check for plagiarism in writing” and only checks submissions against its database. Proctorio says its data “informs human judgment rather than replacing it.” The tool records; a person judges.

What SafeAssign checks against

Matching is only as good as the index behind it. SafeAssign draws on several source collections:

The practical consequence is straightforward: SafeAssign is strongest against material that is public, published, or already submitted somewhere it can see. It is weakest against material sitting behind a login it has no access to.

Can SafeAssign detect Chegg or Course Hero?

This is where most articles overstate in one direction or the other, so here is the careful version.

Where it can. A great deal of Chegg and Course Hero material ends up publicly reachable — indexed pages, reposted answers, content that has circulated onto forums and file-sharing sites. Text-matching does not care where a passage originally came from; it cares whether an indistinguishable string exists somewhere it can see. A widely-used answer to a standard problem is very likely to be somewhere in that space.

Where it cannot. SafeAssign is not logging into a paywalled study platform to read your account. If a solution exists only behind a subscription wall and has never been reproduced anywhere public, there is nothing in the index to match it against.

The part people forget. The most common way copied homework surfaces has nothing to do with the internet index at all — it is the institutional archive. If anyone on your course submitted the same sourced answer before you, that submission is sitting in your own institution’s SafeAssign archive, and it matches perfectly.

And detection is not the only route. Course Hero’s own Honor Code states that a user found to have violated their school’s academic integrity policy in a disciplinary action for misusing the site “will be terminated from our services.” Chegg operates a formal honor-code investigation process that verified school representatives can open. Neither platform is a sealed room.

Can SafeAssign check a PDF?

Yes, for normal text-based documents. SafeAssign processes standard submitted file types and its Originality Report explicitly covers “each piece of your submission, including questions and attachments.”

The genuine limitation is not the file extension, it is whether the file contains selectable text. A PDF that is really a photograph of a page — a scan with no text layer — has no text to extract, so there is nothing to compare. That is a property of the file, not a feature of the tool, and it applies equally to an image pasted into a document.

Reading the Originality Report

SafeAssign’s report is structured more prescriptively than Turnitin’s, and one difference is worth knowing: SafeAssign gives you a risk band, where Turnitin refuses to publish a threshold at all.

The report opens with Overall Risk — low, medium or high — which Blackboard says is “determined based on the average text matching range your attempt falls in.” Alongside it, Overall Text Matching shows the average and highest match across the pieces of your submission.

The full report is then divided into Report Summary, Citations, Document Properties and Submission. Matching sources are listed in the right-hand panel, each numbered and colour-coded against the highlighted text — the report can display up to 30 colours for 30 different sources.

Blackboard’s instruction to both parties is the important line: “The report identifies all matching blocks of text. Instructors and students need to determine if the matching text is properly referenced. Investigating each match prevents detection errors due to differences in citing standards.” A high match on a properly quoted and referenced passage is not misconduct; it is a correctly-functioning matcher.

Can you see your own report?

Only if your instructor allows it, and this is a genuine setting rather than a rumour. Blackboard states that when instructors create an assessment and choose to use SafeAssign, “they determine if students can see the results of the report.”

Where access is granted, you can open the Originality Report shortly after submitting — even before grades are posted. Where it is not, you will not see a score at all. Same tool, two completely different student experiences, decided locally. It is the same pattern as Canvas quiz-log auditing being a course-level opt-in and Turnitin’s AI indicator being an admin toggle.

What about paraphrasing tools?

This section used to say that paraphrasing would keep content out of a similarity check. That was written before AI-paraphrasing detection existed, and it is now misleading in a way that could get someone into real trouble.

Two things have changed. First, rewording does reduce string matching — that much was always true, and it is why a similarity score alone was never strong evidence. Second, and decisively, similarity matching is no longer the only check running. Turnitin now ships AI writing detection built to flag “an AI writing tool such as ChatGPT, text spinners, or AI bypassers,” and reports likely AI-paraphrased text as its own separate category.

So “it will not match” and “it will not be noticed” have come apart. Whether your institution runs SafeAssign, Turnitin, both, or something else is a local decision — which is exactly why reasoning about what a specific checker might miss is a poor basis for deciding what to submit.

What SafeAssign is not built to do

Being clear about the gaps is more useful than implying total coverage, because overstating a tool is how both students and instructors end up drawing wrong conclusions from a number.

• It does not decide misconduct. It surfaces overlap and hands the judgement to a human — Blackboard tells instructors and students alike to investigate each match
• It does not read paywalled accounts. There is no login into a study platform to see what you looked at
• It does not classify AI writing. Matching text against sources is a different capability from predicting whether a model wrote a sentence
• It does not see intent. A quoted, cited passage and a lifted one look identical to a text matcher, which is precisely why the report lists sources rather than issuing a verdict
• It does not run everywhere. Whether SafeAssign is enabled on a given assignment is an instructor choice, made assignment by assignment

Frequently asked questions

Does SafeAssign check previously submitted work?

Yes — the institutional document archive holds papers previously submitted to SafeAssign at your institution, and a global reference database holds papers contributed across institutions. Resubmitting your own earlier work can therefore match against itself, which is worth raising with your instructor before it looks like something it is not.

Does SafeAssign detect AI writing?

SafeAssign is documented as a text-matching and originality tool — it compares text against sources. That is a different capability from the AI-writing classifiers that predict whether a language model produced a passage. Institutions that want AI detection typically run a tool that offers it explicitly.

What is a bad SafeAssign score?

Blackboard does not publish a pass/fail percentage; it publishes risk bands and tells both instructor and student to investigate each match. A score inflated by a correctly-cited quotation and a reference list means something very different from the same number drawn from one uncited source.

Can SafeAssign see images or handwriting?

Text matching needs extractable text. An image of text, a photographed page, or a scan without a text layer offers nothing to compare — which is a limitation of what the tool does, not a loophole worth building a plan around, since a human still reads the work.