Flagged for AI? The 6-Step Proof Stack

#ad. This guide is sponsored by Paperpal. The 6 steps work with any detector. Step 2 leans on Paperpal because it's the one that gives you sentence-level evidence instead of a single number. The discount code at the bottom is TECHBRO30.

What you'll walk away with: A 6-step playbook (the Proof Stack) that takes you from "my essay got flagged" to "I have evidence and a plan", without rewriting a single sentence first.

Prerequisites: A flagged assignment, your earlier drafts, and 30 minutes before you reply to your professor.

Time to complete: 25-40 minutes per assignment.

Why this guide exists

False positives are at an all-time high. Honest students are getting AI scores in the 50s and 70s on essays they hand-wrote over three weeks. Professors see the score and assume the worst. Most students panic, rewrite the whole thing, and accidentally make it look like a cover-up.

The Proof Stack flips that. You build evidence first, you triage the flagged sentences second, and you only rewrite what actually needs rewriting. By the time you walk into that conversation with your professor, you have receipts.

Work through the 6 steps in order. Don't skip step 2.

Step 1: Don't panic, and don't touch the file

The first 10 minutes after you see a high AI score are the most dangerous. If you start editing the document right now, you nuke the version history that becomes your strongest piece of evidence in step 4. Close the file.

Two things to internalise before you do anything else:

1. A high AI score does not mean you cheated. AI detectors flag patterns, not intent. Even your own writing can trigger them if you edit too cleanly, write in a formal academic register, or naturally use uniform sentence structures. Non-native English speakers, students with ADHD or dyslexia, and anyone trained to write in IELTS-style structures get false-positived more often. This is documented behaviour, not a conspiracy.
2. The score is the start of a conversation, not the verdict. Most universities require investigation before any sanction. You have time to build a case.

Take a breath. Open a new doc to take notes in. Now move to step 2.

Done when: The original assignment file is closed and untouched. Version history is preserved.

Step 2: Self-check first with Paperpal (this is where most students get it wrong)

Most students immediately rewrite "the AI-sounding bits". The problem is they don't know which bits actually triggered the flag. They guess, rewrite the wrong sentences, and the score goes up because they introduced new uniform patterns.

You need a sentence-level breakdown before you change anything. That's what this step gives you.

Why Paperpal for this step

Most free detectors give you a single percentage. That's useless. You can't act on a number. What you need is a heat map of which sentences look AI to a detector and which sentences look human, so you know exactly where the risk lives.

Paperpal's AI Detector highlights your essay sentence by sentence in three bands:

Band	What it means	What to do
AI-written	Sentence reads like it came from a model	Candidate for rewrite in step 3
Human-AI blend	Sentence has both signals	Read aloud test in step 3
Human-written	Reads like a person wrote it	Leave alone

That breakdown is the difference between targeted edits and a panic rewrite. You only touch the AI-flagged sentences. The blends get a read-aloud check. The human ones stay exactly as they are.

How to run it

1. Go to paperpal.com.
2. Open the AI Detector tool from the dashboard.
3. Paste your essay (or upload a copy of the file, not the original).
4. Run the check.
5. Screenshot the result with the highlighting visible. This becomes part of your evidence pile in step 5.

The output you want to focus on is the per-sentence highlighting, not the headline percentage. If 4 sentences in a 50-sentence essay are red, you have a small surgical job in step 3. If 30 sentences are red, you have a bigger conversation about your writing voice, but you still know exactly where to start.

Discount code

If you pick up Paperpal Prime to run the unlimited check (and to use the AI Disclosure templates in step 6), use code TECHBRO30 at checkout. Same link: paperpal.com.

Done when: You have a screenshot of the sentence-level breakdown and a list of which specific sentences got flagged red.

Step 3: Rewrite only the flagged sentences

Now, and only now, you touch your essay. Open a copy. The original stays preserved as version-history evidence.

Pull up the list from step 2. Work through one flagged sentence at a time. For each one, run the read-aloud test:

Read the sentence out loud, slowly. If it sounds robotic, too uniform, or too clean, rewrite it. If it sounds like something you'd actually say, leave it. The detector got it wrong, and you can defend it.

When you do rewrite, the move that defeats pattern detection is rhythm variation. Stop trying to write "smarter". Start writing with cadence.

The rhythm move

Real human writing has violent length shifts. Short sentence. Then a longer one that wanders through a couple of subordinate clauses, picks up a qualifier or two, and lands the point with some breathing room. Then another short one for emphasis.

Aim for a paragraph that looks like this:

Short sentence. (5 words)
Medium one carrying a single idea. (10 words)
Then a long one that develops the argument across multiple beats, introduces a contrast, and closes on the key claim with a little weight behind it. (28 words)
Punchy close. (2 words)

If your flagged sentences are all 15-22 words long with similar structures, you've been writing on autopilot or editing too uniformly. Rhythm shifts solve it.

What else to add when rewriting

While you're rewriting a flagged sentence, drop in at least one of:

• A specific date, month, or course code
• A specific number that isn't a round estimate
• A specific name (a person, a paper, a place, a class)
• A specific moment or example from your own research

Generic abstract sentences are what detectors flag hardest. Concrete specifics are the strongest counter-signal. "Students often struggle with time management" reads as AI. "I lost two weekends to one ECO211 problem set in March" reads as a person.

Done when: Every red-band sentence has been read aloud, and the ones that sounded fake have been rewritten with rhythm and at least one concrete specific.

Step 4: Pull your writing trail

This is your evidence layer. Whatever tool you wrote in, the timestamps and version history are your proof that the work happened over time, in your hands.

Google Docs

1. Open the original document.
2. Click File > Version history > See version history.
3. You'll see a panel on the right with every save event timestamped. Each one is a snapshot of the doc at that moment.
4. Click through a few of the early versions to confirm they show messy drafting (typos, deleted paragraphs, half-formed sentences). That mess is gold. It proves a human was thinking, not pasting.
5. Take screenshots of: the version history panel showing the timeline, and 2-3 early drafts that show real evolution.

Microsoft Word

1. If you saved to OneDrive: File > Info > Version History. Same idea as Google Docs.
2. If you used Track Changes during writing (Review tab > Track Changes), every edit is logged. Open it and screenshot the markup.
3. If you saved local-only without Track Changes, look in File > Info > Manage Document > Recover Unsaved Documents and check your AutoRecover folder for autosaves.

Notes app / Apple Notes

1. Each note has a Created and Modified timestamp at the top (View > Show Note Info, or right-click on macOS).
2. If you brainstormed or outlined in Notes before drafting in Word, those timestamps prove the work started earlier than the assignment file did.

Whatever you actually used

The principle is the same regardless of tool. Find the timestamps. Capture the message thread to yourself, the iCloud sync log, the Notion edit history, the Obsidian backlinks, the WhatsApp message where you sent yourself a paragraph at 2am. All of it counts.

Done when: You have at least 3 timestamped artefacts showing the work happened over multiple sessions, in your own messy hand.

Step 5: Document your process

Step 4 was about the file. Step 5 is about the thinking.

Pull together everything that proves you understood and engaged with the topic. The point is to show that you can talk about your essay because you wrote it, not because you re-read what ChatGPT gave you.

Build an evidence folder with:

• Research screenshots. The journal articles you read, browser history showing your search trail, library checkout records, PDF annotations.
• Outlines and brainstorms. Handwritten notes, photos of whiteboards, mind maps, the messy first attempt.
• Timestamped messages. The WhatsApp chats with classmates where you debated the argument, the Discord screenshots, the email to a tutor with a question.
• Voice memos. If you talked through ideas, the audio file timestamp is evidence.
• Textbook references. Open to the page you cited and photograph it next to your draft.

The phrase to remember: evidence beats arguments. When the conversation with your professor happens, you don't want to be defending your character. You want to be opening a folder.

Done when: You have a single folder (Drive, Dropbox, or local) containing the version-history screenshots, the Paperpal sentence-level report, your research artefacts, and your outline. Named clearly: [Course code] - [Assignment name] - Evidence Pack.

Step 6: Know your university's AI policy before you walk in

This is the step students skip and regret. Walk into your professor's office without knowing the policy and you'll either over-confess (admitting things you didn't do) or under-prepare (missing the protections you actually have).

Find the policy

The policy lives in one of three places:

1. Student handbook. Usually a PDF on the university website. Search it for "AI", "artificial intelligence", "academic integrity", "plagiarism".
2. Academic integrity portal. Most universities have a dedicated page. Search "[your university name] academic integrity AI".
3. Course-specific syllabus. Your lecturer may have written their own AI rules into the course outline. Re-read the first week's docs.

What you're looking for

Check the policy for these specific things:

• Does the university require disclosure of AI use, even for grammar or research?
• Does the policy distinguish between generation (AI wrote it) and assistance (AI helped you research)?
• Does it explicitly ban AI humanizer tools or paraphrasing tools?
• Is there a formal appeal process with multiple stages?
• Does the policy state that AI detector output alone is sufficient evidence for misconduct? (At many SA universities, it isn't. See the SA examples below.)
• What sanctions exist for first vs repeat offences?

South African universities: where you stand

If you're at an SA university, the rules vary widely:

University	Stance	What you need to know
NWU	Mandatory disclosure on every assignment	Use without disclosure = academic misconduct
Stellenbosch	Postgrads sign an AI declaration. Turnitin AI detection active	Confirm whether your specific assignment required disclosure
UJ	Undisclosed AI use = "fraudulent authorship"	Sanctions scale by intent. First offence is typically a warning
UP	53 confirmed AI cases 2024-25. Triangulates with drafts + version history + oral defence	Humanizer tools explicitly banned. Have your drafts ready
UNISA	4 detection tools running together (Turnitin, Invigilator App, Moodle analytics, IRIS). Zero tolerance	Integrity course is mandatory before registration. Re-read it
UCT	Dropped AI detection in October 2025	If your detection happened post-Oct-2025, ask which tool was used and why
Wits	"Promote ethical use, not detect cheating"	Lighter touch overall, but disclosure still expected
Stadio	Traffic-light system. Some assignments require AI use	Check the colour code on your specific assignment

If you're at a university not on this list, the same checklist applies. Find the handbook, find the AI section, read it twice.

Use Paperpal's AI disclosure templates

If your university requires a disclosure statement and you legitimately used AI for grammar, citation formatting, or research summaries, you need to declare that cleanly. Paperpal Prime ships with AI Disclosure templates: pre-written, university-friendly statements you can adapt. The framing matters. Undisclosed AI looks like cover-up. Properly disclosed AI looks like academic honesty.

If you didn't use AI at all, you don't need a disclosure. But you do need to know that's the policy-compliant answer.

Done when: You can quote the relevant policy clause in conversation, you know which sanctions apply, and you have your disclosure statement ready (or you've confirmed you don't need one).

The full Proof Stack at a glance

Print this. Screenshot it. Whatever. Use it every time someone gets flagged.

1. Don't panic, don't touch the file. Preserve version history.
2. Self-check with Paperpal. Get the sentence-level breakdown (code TECHBRO30).
3. Rewrite only the flagged sentences. Read aloud, vary rhythm, add specifics.
4. Pull your writing trail. Version history screenshots, timestamps.
5. Document your process. Research, outlines, voice memos, messages.
6. Know your university's AI policy. Disclosure statement ready if needed.

flowchart TD
    A[Got flagged] --> B[Step 1: Close the file]
    B --> C[Step 2: Run Paperpal AI Detector]
    C --> D{Which sentences flagged?}
    D --> E[Step 3: Read aloud + rewrite reds only]
    E --> F[Step 4: Screenshot version history]
    F --> G[Step 5: Build evidence folder]
    G --> H[Step 6: Read AI policy + draft disclosure]
    H --> I[Walk into the conversation with receipts]

What this guide doesn't do

Three honest limits:

1. It doesn't help if you actually generated the essay with AI. None of these moves will save a defence if you can't talk about your own argument. Every step assumes the work is yours and the flag is wrong.
2. It doesn't replace your university's appeal process. If a formal hearing has been scheduled, you also need how to appeal an AI detection flag. That one walks through the appeal mechanics specifically.
3. It can't guarantee your professor will accept the evidence. Some lecturers are entrenched. The Proof Stack maximises your odds by preparing for every counter-argument, but you may still need to escalate.

Pre-conversation checklist

Before you reply to your professor, walk through this:

• Original file untouched (version history preserved)
• Paperpal sentence-level report screenshotted
• Flagged sentences rewritten (in a copy, with rhythm and specifics)
• Version history exported / screenshotted (3+ snapshots)
• Research, outlines, and process artefacts in one named folder
• University AI policy read, relevant clauses bookmarked
• Disclosure statement drafted (or confirmed not needed)
• You can verbally explain your argument without the document open

Only when all eight are checked do you open the email reply.

CTA: get the tool that step 2 runs on

The whole Proof Stack works without paid tools. But step 2 is the one where a generic detector won't cut it. You need sentence-level highlighting, not a percentage.

Use Paperpal: paperpal.com. Code TECHBRO30 for the discount.

Disclosure: Paperpal sponsored this guide. The 6-step framework works with any sentence-level detector. I don't know any other one that does the three-band highlighting at the same depth, so I'm pointing you at the one I tested.

Next steps

• If you're not flagged yet but want to write flag-proof, read the flag-proof writing checklist
• If you're already in a formal hearing, read how to appeal an AI detection flag
• If you want to understand what each AI score percentage actually means, read Turnitin's AI score explained
• For the broader picture of which SA universities take which stance, read University AI Policies in 2026

This guide powers the "Comment FLAGGED" CTA on ATB-148. You got it because you asked for it.