A flag on your AP Seminar paper rarely arrives with an explanation. The notice says the work may show unacceptable use of AI, and it never names the system that raised the concern, which is why students and teachers end up searching for what AI detector does college board use.
College Board does not publish a single detector. Its exam policies describe investigating submissions where there is evidence of inappropriate generative AI use, and the screening usually runs through tools your school already licenses, most often Turnitin, with Originality and Copyleaks in the same mix.
This checklist covers which systems sit behind a flag, how accurate each one is, where false positives come from, and what to do when your own writing gets caught. When you want a second read before resubmitting, AI Busted puts your draft through a detector and a humanizer, and the free 5-day trial lets you test it on the exact paragraphs that were flagged.
What AI detector does College Board use?
College Board runs no detector of its own, so the direct answer is Turnitin: schools screen submitted work through Turnitin's AI writing detection. College Board investigates submissions where evidence suggests inappropriate use of generative AI, and most districts already license Turnitin for plagiarism checks, so the screen is in place before you submit. If your school licenses Copyleaks or Originality instead, that becomes your screen.
When a paper is flagged, College Board's guidance explains what options a student has.
AI detectors compared: what each one actually flags
Five screening paths turn up in College Board integrity reviews, compared on three criteria: which text pattern the system measures, who reads the output, and how it behaves on genuine student writing.
| Option | What it measures | Best for | Limitation |
|---|---|---|---|
| Turnitin AI writing detection | Sentence-level perplexity and burstiness, meaning how predictable and how varied each sentence is | Schools whose teachers already submit essays through Turnitin | A flag reaches your teacher, not College Board, and formulaic human prose can still trip it |
| GPTZero | Perplexity and burstiness with a paragraph-level highlight of the sentences that scored highest | A teacher spot-checking one submission before a conference | Short AP-style paragraphs give it little signal |
| Originality.AI | Whole-document AI probability alongside plagiarism matching | Editors reviewing long drafts outside a classroom rubric | Tuned for publishing, so it ignores the process evidence AP Seminar scores |
| Copyleaks | AI probability across multiple languages | Multilingual submissions or translated drafts | Confidence drops on languages with thinner training data |
| Teacher review plus an AP reader | Your evidence, revision history, and citation trail | Any flagged performance task or written exam response | Slow, and it only works if you kept drafts |
Read it by asking who acts on the output. A Turnitin or GPTZero flag goes to your teacher, who decides whether to escalate; College Board's published guidance says it reserves the right to investigate submissions itself, and that review weighs your drafts and citations rather than a probability number.
First checks that decide whether a College Board AI flag holds
A College Board integrity review usually starts with a number, not a close reading. Your teacher runs the submitted file through the school's writing screen and spends real attention on whatever it underlined. So the first thing an AI flag tests is not your style. It is whether you can show how the paper came to exist.
Teachers describe original student work coming back flagged often enough that a bare denial is the weakest reply. When students who used AI to sharpen one conclusion get questioned, the conversation expands to the whole paper, because the reviewer cannot isolate the part you admit to.
- [] A dated earlier draft of the paper exists as its own file, not only as the final version.
- [] Every source quoted in the body appears in your bibliography with matching spelling and year.
- [] Any paragraph you drafted with AI help is one you can summarize out loud, in your own words.
- [] Quoted text matches the original word for word, including punctuation inside the quotation marks.
- [] Your teacher saw your research question and outline before the final draft arrived.
Skipping this group leaves you arguing about a percentage with nothing in hand. Check your own draft before anyone else does: AI Busted pairs a detector with a humanizer, rewriter, and AI remover, and its free 5-day trial lets you see which sentences read as machine text.
Evidence checklist: what to collect before you respond
A flag you can answer is a flag you can document: College Board treats a flagged submission as something the student responds to, so the version of your work you can produce matters more than the version you remember writing.
| Task | Why it matters | Done |
|---|---|---|
| Ask your teacher which system produced the flag and what percentage it reported | Turnitin's AI writing detection reports one overall figure for a submission, and that figure frames the conversation | |
| Export the flagged file to PDF with its last-saved date visible, before you edit anything | The copy you discuss should be the copy that was screened | |
| Pull revision history from Google Docs or Word, plus any drafts you emailed or printed | Timestamps showing the text changing across several days are the strongest authorship evidence you own | |
| Mark every sentence in the flagged sections that traces back to a source, a class note, or an outline | A reviewer reads one claim at a time, not a blanket denial | |
| Run the same final draft through an independent detector, such as a free online AI content detector, and keep that reading | A second number shows whether the flag is an outlier or a pattern | |
| Write a one-page chronology of dates, draft names, who read it, and what you changed after feedback | A dated page is something a coordinator can pass along without asking you to explain it out loud |
Work the table from top to bottom and finish it before you discuss the flag with anyone. The first two rows take minutes and protect everything that follows. If the notification lands mid-week, aim to have the whole page assembled by Friday, because conversations about a flag tend to start fast.
A row counts as done only when someone else can look at the artifact. A saved PDF with a visible date counts. The sentence "I wrote this myself" does not, because there is nothing on the page to inspect. If you cannot point to the file, the date, or the note, the row is still open.
Ask for the report rather than a summary of it. A report usually pairs one overall figure with the sections it flagged, and knowing which passages were highlighted tells you what to address instead of defending the whole paper. If the school will not share the report, ask whether the figure came from the entire submission or from a single matched passage.
Editing before you freeze the copy is the most common self-inflicted problem. Rewrite the flagged paragraphs and the file a reviewer can open no longer matches the file the school screened, which is harder to explain than the original flag. Fix grammar after the export, not before.
Do not run the flagged paragraphs through a rewriting tool to make them look different. Rephrased sentences read unlike the rest of your coursework, and anyone comparing the flagged text against your earlier submissions will notice the shift. That kind of change looks like a correction you were never asked to make.
A school screen and a College Board integrity review are separate steps. A teacher's report is what usually starts the conversation, while College Board reserves the right to investigate a submission itself when it sees evidence of inappropriate generative AI use. A teacher may accept an explanation that a formal review would not.
Keep the chronology to one page and the dates in order. Send the frozen PDF, the revision history export, and the reading from the detector you ran as separate attachments, so each item can be checked on its own. A short email that names each attachment in a line is easier to act on than three paragraphs of explanation.
If the flag holds anyway, the next step is procedural rather than rhetorical. Students who get a notification about unacceptable AI use have a defined set of options, and those options carry deadlines. The write-up on what happens when College Board detects AI walks through that sequence, including where a written response fits into it.
Print the page, mark the Done column, and keep the packet together with the date you received the notification written on the front. That packet is what you reach for if the review moves to a second stage. Building it now takes an afternoon; rebuilding it later means guessing at dates you no longer have.
What AI Detector College Board Uses on Your Submitted Work
College Board does not run its own AI detector. Screening comes from tools your school already licenses, most often Turnitin, whose AI writing detection reports a percentage of text flagged as likely machine-generated. Copyleaks, Originality, and GPTZero also show up in school workflows. Our detector comparison runs the same essay through several of them.
That distinction shapes what you can verify. There is no public College Board score to read; the flag reaches you through a teacher or AP coordinator. College Board's own guidance covers what happens once a student is notified their work has been flagged, and the response route runs through your school.
The checks below split into two moments: before you submit, and after a flag appears. Each one is observable, so you or a teacher can point at a file, a date, or a document and agree on pass or fail.
- [] Your draft history shows dated versions, whether in Google Docs version history, a dated Word file, or one saved copy per draft.
- [] You ran the final text through two detectors and kept both results, including any that flagged it.
- [] Every cited source opens, and every quotation matches the original wording.
- [] You can explain out loud why each paragraph sits where it does.
- [] After a flag, you filed a written response through your teacher within the window your school set.
Skip this and you have no dated evidence and no detector record, which turns a false positive into an argument you cannot win.
What the College Board screen actually flags
The College Board does not run a detector of its own. Screening comes from tools the school or the AP program licenses, and the one most often named in AP Seminar threads is Turnitin's AI writing detection. A flag is not a finding. It opens a review, and the review is where a student's options after being flagged actually live.
Ask which screening tool your school licenses before you submit, since that answer decides how much a flag means in your case.
The part that catches people out is how the score is produced. A detector reads the whole submission as one signal, so a paper you drafted by hand and finished with AI help on the closing paragraph can come back mixed, reading worse than your process deserves. The usual advice, write it yourself and you have nothing to worry about, holds right up until that happens to you.
When it does, the useful move is documentation rather than rewriting: dated drafts, your source notes, and a short written account of which sentences the tool touched and which it did not. In practice, the number matters far less than what you can show about how the document was built.
Frequently asked questions about College Board AI detection
Does College Board actually use AI detectors?
College Board does not build or operate its own AI detector. It screens student submissions using third-party detection software, most often Turnitin, because many schools already license Turnitin for plagiarism checks.
Official AI guidance from College Board states that it reserves the right to investigate submissions where there is evidence of inappropriate generative AI use. In practice, a flag usually starts with your teacher reading a screening report, not with a national scoring office.
How does College Board check for AI-generated student papers?
College Board checks papers by combining a software screen with human review of the writing itself. Turnitin's AI writing indicator reports the share of a submission it believes was machine-written, and that percentage reaches your teacher alongside the similarity report.
A reviewer then compares the flagged sections against your drafting trail: outlines, earlier versions, citations, and reading notes. Flags that cannot survive that comparison get dropped.
What happens if your work is flagged for unacceptable AI use?
Your school notifies you of the flag, and you get the chance to respond before a score decision is made. The first useful step is asking your teacher which sections triggered the screen and what evidence of process would answer it.
Students who kept dated drafts, reading notes, and revision history have a far easier conversation than students holding only a final file.
Does AP Classroom detect AI on assignments?
AI detection on AP Classroom assignments depends on the screening your school has set up, which is usually Turnitin. Multiple-choice progress checks are scored against answer keys, so they produce no AI signal at all.
The submissions that get screened are written work such as AP Seminar performance tasks and AP Research papers.
How accurate are AI detectors at catching AI writing?
Accuracy is high on unedited machine output and much lower on text a student has reworked sentence by sentence. Copyleaks, GPTZero, and Originality frequently disagree on the same passage, which tells you how much weight a single percentage deserves.
Turnitin's own documentation treats its AI writing indicator as a signal for human review rather than a verdict. Teachers also report false positives on formal, polished prose, so keep drafts that show how the text came together.
Final verdict: what to do next after a College Board flag
Your next move depends on which side of the flag you are on. If you wrote the draft and want to keep it, the practical fix is making the prose read as yours before you hand it in. AI Busted is where to start: its Humanizer and Rewriter work on sentences you already wrote rather than generating new ones, and the free 5-day trial means you can see what it does to one paragraph before you decide.
Choose differently when the case is already open. Editing a flagged file after a teacher has raised it proves nothing and can read as an admission. Dated drafts, class notes, and a straight answer about which sentences came from ChatGPT carry the weight instead.
The core recommendation holds either way. A College Board screen measures text patterns, not intent, so keeping your own sentences recognisably yours from the first draft onward is what prevents the next flag.
