Direct answer
An AI detector cannot score an image directly. Turnitin's AI writing detection analyses the words inside a submitted document, not the pixels of a screenshot or photo, so a .png, .jpg, or scanned page must first be converted into machine-readable text before any AI score can be produced [1]. Once the text is extracted and saved as a proper document, the detector evaluates the prose segments and returns a percentage-based indicator of how much qualifying text looks AI-generated [2]. The extraction step is therefore not a formality — it is the step that determines whether your score is meaningful at all.
Can an AI Detector Read Text Directly From an Image, or Do You Have to Extract It First?
No — you have to extract it first. Turnitin's AI writing detection is built to analyse long-form prose inside submitted documents, and it works by examining the writing itself rather than the visual layout of a page [1]. An image file carries no analysable text layer unless something has already performed optical character recognition on it, which is why a screenshot of a ChatGPT reply cannot be dropped into a detector and scored.
This matters because the detector reports on segments of prose, not on whole files. The system identifies passages of qualifying writing and calculates the proportion of those passages that its model classifies as AI-generated [2]. If the input is an image, there are no segments to classify, so the detector has nothing to return — you get no score rather than a low score. Those are very different outcomes, and confusing them is one of the most common reasons students misjudge their own risk.
It is also worth being precise about what the number means even in the best case. An AI score is a probability indicator, not a verdict, and it is designed to be read alongside human judgement rather than in place of it [1]. That distinction becomes even sharper once you have passed text through an extraction step, because extraction can quietly alter the very patterns the detector is measuring. Treat any score as a signal about the submitted text — not as an inherent property of whatever you originally photographed.
The practical takeaway is simple: the image is a container, and the detector only ever sees what comes out of that container. Get the text out cleanly, and the detector can do its job. Leave it inside the image, and no detector — Turnitin included — has anything to work with [2].
How Do You Convert Text From an Image Into Something an AI Detector Can Score Accurately?
The goal is a clean, text-bearing document in a format the submission system actually accepts. Turnitin's submission rules revolve around document and text formats with minimum length requirements, and image-only files are not treated as AI-analysable submissions [3]. In practice that means your conversion target is a .docx, .pdf, or .txt file containing real, selectable characters — not a PDF that is really just a picture wrapped in a page.
Start with the highest-fidelity source you can get. If the text originated in a chat interface or a word processor, export or copy it rather than photographing a screen; re-typing or copy-pasting preserves punctuation, sentence boundaries, and paragraph structure that OCR tends to smear. If a photo or scan is genuinely your only source, run OCR at the highest resolution available, then proofread against the original line by line. OCR reliably mangles em dashes, curly quotes, in-text citations, and superscripts — and citation-dense academic prose is exactly where those errors cluster.
Formatting loss is not cosmetic. Because detection works on prose segments, a paragraph that OCR has chopped into fragments or merged with a heading may no longer register as a qualifying segment at all [2]. That can push your reported percentage down for reasons that have nothing to do with how the text was written, which is a false comfort rather than a genuine result. Before you trust any score, confirm the extracted document reads the way the original did: same paragraph breaks, same sentence punctuation, same citation markers.
Finally, check the boring constraints before you upload. Length thresholds and accepted file types are enforced at submission, and a file that fails those checks produces an error rather than a report [3]. A short screenshot of two paragraphs is often below the minimum length a detector needs to return anything useful, so extracting a single image is frequently not enough — you need the surrounding document too. Build the full text first, then check it.
After You Extract Text From an Image, How Do You Check Its Turnitin AI Score Before You Submit?
You check it against the same report your instructor will see, on a document that is not added to your institution's paper database. The value of a pre-submission check comes from seeing the actual output — the flagged passages, the percentage, and the similarity matches — while you still have time to revise, rather than discovering them after the deadline has closed [4]. Turnitin's own guidance is clear that AI scores require human interpretation and that formulaic, heavily templated, or non-native-sounding prose can attract flags that a reader would not consider plagiarism [4]. Seeing your own score early is how you find out whether that applies to you.
Read the report in a specific order. First look at which passages are flagged, not just the headline number, because a single flagged paragraph in a 4,000-word essay means something very different from a flagged abstract and conclusion. Then check whether the flagged text is the part you actually wrote with AI assistance — a section you drafted in ChatGPT and lightly edited will usually be the one that lights up, and that is the section to rewrite in your own voice. Finally, cross-read the similarity report alongside the AI report, since a high similarity match and a high AI score are two separate problems with two separate fixes.
Be careful with the low-confidence band. When detection confidence falls below Turnitin's threshold, the report displays *% instead of an exact figure, which signals a weak or uncertain result rather than a clean pass [2]. Students routinely misread *% as "zero" and submit unchanged. It is not zero; it is "the detector is not confident," and an instructor reading the same report sees the same asterisk.
If the flagged text came out of an image — a screenshot of a draft, a photo of handwritten-then-typed notes — remember that the extraction itself may have changed the result. Re-check the cleaned-up document rather than the raw OCR output, and if the score is still high, revise the prose before you resubmit. The point of checking early is to convert an uncertain outcome into a decision you can still act on [4].
Extracting text from an image is the easy half; knowing whether the resulting document will clear your institution's AI check is the half that actually keeps you out of trouble. turnitin0 lets you upload that cleaned-up .docx, .pdf, or .txt and see the real Turnitin AI detection report and similarity report — the same two PDFs your professor sees — in most cases within 5–15 minutes, without adding your file to the student paper database. You get the flagged passages and the score in front of you while there is still time to rewrite, which is precisely the advantage a last-minute submission takes away.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
Get Real Turnitin AI & [Similarity Report](https://www.turnitin0.com/guides/us/ai-prices)
FAQ
Can I upload a screenshot or photo directly to an AI detector?
No. AI writing detection operates on text inside submitted documents, so image files carry nothing for the model to classify [1]. You must run OCR or copy the text into a .docx, .pdf, or .txt file first, then submit that document instead.
Does OCR change my AI score?
It can. Because detection works on prose segments, OCR that breaks sentences, drops punctuation, or merges paragraphs can change which segments qualify for analysis and therefore shift the reported percentage [2]. Always proofread extracted text against the original before you rely on the result.
Why does my report show *% instead of a number?
Turnitin displays *% when its detection confidence falls below the threshold required to report an exact figure [2]. It indicates a low-confidence signal, not a confirmed zero, and instructors reading the same report see the same asterisk.
Is there a minimum length for an AI check?
Yes — submission systems enforce length and file-type requirements, and very short extracts often fall below them [3]. A single screenshot of two paragraphs is usually too short to return a useful result, so build the full document before checking.
Will a pre-submission check add my work to my university's database?
Not with a non-repository check. turnitin0 runs the file without adding it to Turnitin's student paper database and does not share reports with third-party databases, so your draft is not stored as a future match against other students. You can also delete files from your account afterwards.