Discussion Board Post Flagged as AI: Find Out What Flagged It First

Turnitin's published file requirements say a submission needs at least 300 words of prose text in a long-form writing format before an AI Writing Report is produced. Its definition of qualifying text is "individual sentences contained in paragraphs that make up a longer piece of written work, such as an essay, a dissertation, or an article." A 220-word weekly post is under the floor, and its accepted file types are .docx, .pdf, .txt and .rtf, which a text box in a course site is not. So when a number lands on a discussion post, something happened before it got there. Finding out what is the first move, because the three usual answers call for three completely different responses.

The rest of this is about that question, and then about the thing that only happens on a weekly board: by week nine you are writing against eight of your own earlier posts.

From Turnitin's guide to the AI Writing Report, under the heading "File requirements", the whole list:

• File size must be less than 100 MB
• File must have at least 300 words of prose text in a long-form writing format
• File must not exceed 30,000 words
• File must be written in a supported language: English, Spanish, Japanese, Arabic
• Accepted file types: .docx, .pdf, .txt, .rtf

Two rows matter here and both are usually skipped.

The 300-word row is a floor on whether a report exists, not a warning about accuracy. The same guide lists the states the indicator can display, and one of them is "File didn't meet the requirements", described as: "The submission does not meet the file requirements needed to successfully process it for AI writing detection." That is a different outcome from a low score. It is not a percentage at all. There is a second official sentence about the 300-word line, and it is worth reading with its date on it. Turnitin's release notes for 15 December 2023 record a bug fix: the bug had "allowed submissions that did not meet the file requirements (less than 300 words) to be processed and given an AI writing detection score", and the entry says "this issue has now been resolved". About the reports that came out between 6 and 15 December while the bug was live, the same entry adds that "submissions that contain less than 300 words may result in an AI writing score that is likely less accurate". Two official lines, two different situations: under 300 words there is normally no report, and where one appeared anyway the vendor rated it less accurate. How accurate a Turnitin AI score is on short documents takes the accuracy half further.

The file-type row is the one that tells you something about your own course. A discussion post typed into a course site is not a .docx or a .pdf. For it to be scored by this particular report, it has to become a file first. Somebody or something did that, and how it was done decides what the resulting number is even describing.

The supported-language row prints four: English, Spanish, Japanese, Arabic. Arabic was added on 18 August 2026; before that the list was three. Text outside those four is not something this report handles badly. It is something this report does not handle.

That row does more work on a discussion board than on an essay. If your programme is taught in another language, or the board for a language module runs in the language being studied, a post written there sits outside what the AI Writing Report covers. So a percentage attached to it did not come from this report, which drops you straight into the second row of the table in the next section. which languages Turnitin's AI detection covers and can Turnitin detect AI in non-English submissions go through what sits inside that list and what does not.

The same guide adds one more line about reach: "Please note that only our English AI detector includes AI paraphrasing and AI bypasser detection capabilities." That is a vendor describing the scope of its own product, and it is worth reading as exactly that and nothing more. We draw no conclusion here about how well any tool works, and this is not a sentence to plan around. Its use is that you know what the vendor says it looks for.

One more route worth naming, because it has nothing to do with generated text: drafting the post in your first language and translating it into English before you paste it in. Whether that reads differently to a detector is its own question, worked through in does machine translation get flagged as AI. And the broader claim, that writing in a second language draws more flags, is held up against the documentation in non-native speaker writing flagged as AI.

The first row is the one people miss, and it changes everything about the response. If eleven weeks were bundled into one file, then the score is a statement about eleven weeks of your writing pooled together, and arguing about week nine's second paragraph is arguing about the wrong object.

There is a clean way to find out which row you are in. Turnitin's FAQ says "the AI writing detection indicator and report are not visible to students" and separately that "only instructors and administrators are able to see the indicator." It also says: "However, with the PDF download feature, instructors can download and share the AI report with students." So asking for the PDF is a reasonable request with a documented mechanism behind it, and the answer sorts you into a row. If a PDF of a Turnitin AI Writing Report exists, you are in row one or you had a longer submission than you thought. If nobody can produce one, you are in row two or three.

If it turns out to be row two, why the same text scores differently on every detector covers why two tools give two answers on the same paragraph.

Here is the part that has nothing to do with length.

A course discussion asks you the same shape of question every week for twelve weeks. Read the chapter, respond to the prompt, reply to two classmates. You are a consistent person with a consistent way of approaching a reading, so by week nine your posts share an opening move, a paragraph count, a way of getting into the text and a way of getting out of it.

Turnitin's FAQ names the profile:

"Sometimes false positives (incorrectly flagging human-written text as AI-generated), can include content without a lot of structural variation, text that literally repeats itself, or text that has been paraphrased without developing new ideas."

And the sentence after it, which instructors are entitled to act on:

"If our indicator shows a higher amount of AI writing in such text, we advise you to take that into consideration when looking at the percentage indicated."

An essay is written once. A weekly discussion board is the one assignment in a degree where you are asked to produce structurally similar text on a schedule, and where the similarity is a feature of the assignment design rather than anything you decided.

The fix is small and it is not a writing fix. Keep a one-line note of how you opened each week. Not what the post was about, the move you used to get in. A short list of moves that are actually different from each other:

• a question you could not answer from the reading
• a specific line or number from the assigned text, with the page
• a place where two of the assigned readings disagree
• something from your job, placement or lab that cuts against the reading
• a term you think the author uses two different ways

Then do not repeat a move inside four weeks. It costs about a minute, it is entirely under your control, and the side effect is that the posts get better, which is the part your instructor actually cares about.

People conflate these two and then rewrite the wrong thing.

If you reused a paragraph from your week three post in your week nine post, that is matching text. Turnitin is explicit that the two numbers do not touch:

"The Similarity score and the AI writing detection percentage are completely independent and do not influence each other."

So self-repetition across weeks is a question for the similarity side, and whether your own earlier post is even in the comparison depends on whether your institution adds submissions to the repository, which is an institutional setting rather than a fact about Turnitin. Does Turnitin store your paper covers the repository question, and the AI report and the similarity report sets out what each of the two measures.

The practical version: rewording a repeated paragraph to move an AI number is aiming at the wrong report.

Now the good news, and it is genuinely good.

The discussion board is the only assignment format that builds a writing sample of you, week by week, without you doing anything. By week nine there are eight dated pieces of your prose sitting in a system you do not control, written before anybody was arguing about anything, on a schedule set by somebody else.

That is a better record than most students can produce for a final essay. Microsoft's own documentation says version history "only works for files stored in OneDrive or SharePoint in Microsoft 365", so a .docx that has lived on your laptop for six weeks has no history at all. We went through where the record lives for each tool in how to keep version history in Word, Google Docs and Overleaf. A discussion board sidesteps that whole problem by accident.

One thing to do about it, and do it now rather than in week eleven. Copy each week's post into one document as you go, with the date on it. Ten seconds a week. If your institution closes course access at the end of term, that corpus goes with it, and the moment you actually want it is usually after the course has ended.

Both things are true at once, which is the odd shape of this format: the same run of posts that makes your writing look repetitive is also the best evidence of how you write that you will produce all year.

Two honest caveats before the tool part.

First, a 200-word post does not have much in it to scope. The thing worth having when you are working on a single short post is a pair of eyes and ten minutes, not software.

Second, nothing here promises a number in either direction. Turnitin says its own model "may not always be accurate (it may misidentify human-written, AI-generated, and AI-paraphrased text), so it should not be used as the sole basis for adverse actions against a student."

Where software starts to earn its keep is the bundled case: eleven weeks in one document, most of it fine, four paragraphs marked. HumanPen is built for that shape. Nothing gets touched unless it is in scope, and scope comes from you, either by marking passages by hand or by handing it a Turnitin or iThenticate AI report so the flagged ones fill the list. Week two and week five stay exactly as you typed them. Paragraphs are the unit, so a half-covered paragraph gets pulled in whole and shown back to you first. You are charged for the words rewritten, not the file. DOCX, PPTX and TXT all go in, and TXT holds its line breaks and indentation, which is the shape a discussion post is already in. Anything that comes back still flagged can go through again at no cost when the conditions are met.

Its FAQ tells you to look over complex documents after downloading them. On a discussion board that check has one specific target. A post is a turn in a conversation. If the revised version no longer answers the prompt, or no longer connects to the classmate you were replying to, no number is worth what you just lost.

My post is 180 words. Can Turnitin even give it an AI score? The published requirement is at least 300 words of prose text in a long-form writing format. Below that, the report guide describes a "File didn't meet the requirements" state rather than a percentage. What your institution's system does with a discussion post before it reaches any checker is a question for your instructor, not something a general guide can answer.

What about the two replies I have to post each week? A fifty-word reply is a long way under the floor on its own. Whether it was included in anything at all depends on how the submission was assembled, which is the same question as the one above. It is worth asking, because a reply that quotes a classmate is carrying somebody else's words into whatever got checked.

Does the same 300-word floor apply to the similarity report? No. The requirement quoted here is from the file requirements for an AI Writing Report specifically. The similarity side is a separate report with separate behaviour, laid out in the AI report and the similarity report.

Should I stop sounding like myself across weeks? No, and you could not if you tried. Vary the way you get into the reading, not your voice. The thing described in Turnitin's false-positive list is structural sameness and literal repetition, not having a recognisable style. Whether it is worth changing how you write at all is argued out in should you change how you write to avoid being flagged.

My score came back as an asterisk with no number. That is a display rule, not a result. Scores above 0% and below the 20% threshold are shown as an asterisk with no percentage and no highlights attributed, which what the asterisk means on a Turnitin AI score explains.

A detector gave my post a low score. Does that settle it? No, in either direction. The vendor's own line is that the model may misidentify human-written, AI-generated and AI-paraphrased text and should not be the sole basis for adverse action. No tool establishes who wrote a document.