Direct answer
A ChatGPT checker cannot analyze an image directly: AI-writing detectors such as Turnitin read the text inside a submitted document, not pixels, so a screenshot or photo of your essay returns no meaningful AI score on its own [1]. To check image-based work you must first convert the image into editable text — ideally by recovering the original .docx, .pdf, or .txt file — and then run the check on that file [2]. Detection also only measures patterns in written language, so it says nothing about whether an image itself was AI-generated [3].
Can a ChatGPT Checker Actually Read an Image?
Short answer: no checker of AI-written prose reads an image as an image. Turnitin's AI writing report is produced from the text of a submitted document, and the indicator is calculated over the prose the system processes — which is why the report is tied to a file, not to a picture of a file [1]. A screenshot of a paragraph is, to the detector, just a bitmap; there is no text layer to analyze unless something extracts it first.
This is a structural limit rather than a missing feature. Detection models look at token-level and sentence-level patterns in language, so the input has to be machine-readable text in a supported document format [1]. That is also why the guidance around the AI writing report focuses on document submissions and on how the score is displayed — Turnitin shows *% instead of an exact percentage when its confidence falls below the 20% threshold, a signal that the result is low-confidence rather than a clean pass [1].
The practical consequence for students is simple: if the only copy of your draft you can find is a screenshot, a photo of a printed page, or an image exported from a chat window, you do not yet have something a checker can score. You have to reconstruct the text first, and the quality of that reconstruction determines how trustworthy the resulting report is.
What Can AI Detectors Tell From Text Versus an Image?
From text, a detector can report a probability-style indicator of AI involvement plus sentence-level flags showing which passages drove the score [3]. What it cannot do is establish authorship: the indicator describes statistical patterns in the writing, not who typed it, and it does not authenticate provenance or intent [3]. Heavily edited prose, formulaic academic phrasing, and text run through rewriting tools can all produce signals that need human interpretation rather than a verdict [3].
From an image, a prose detector can tell you essentially nothing about AI writing, because the language signal it depends on is absent until the image is converted to text [2]. Even after conversion, something is usually lost: extracted text can drop formatting, footnotes, table structure, and sometimes citation markers, and those are exactly the elements a similarity report uses to match sources [2]. A check run on OCR output is therefore an approximation of a check run on the real file.
It also helps to separate two different questions students often merge. "Was this essay written by AI?" is a text-analysis question answered by an AI writing indicator [3]. "Was this image generated by AI?" is an image-forensics question that prose detectors do not address at all, and no Turnitin AI writing report will answer it [1]. Keeping those two questions apart prevents a lot of wasted effort.
How Do You Get an Image-Based Draft Checked Before You Submit?
The reliable route is to recover the original file. Institutions expect the editable document you actually wrote in, and integrity guidance consistently treats the original .docx as the authoritative artifact — formatting, headings, and reference lists survive intact, so the report reflects what your marker will see [4]. Check your cloud drive, email attachments, and the "recent files" list of your word processor before doing anything with the image.
If the original genuinely no longer exists, convert the image to text deliberately rather than casually. Use a proper OCR tool on a high-resolution scan, then rebuild the document structure by hand: restore headings, re-link citations, and re-check every number against the source, because OCR silently corrupts digits and symbols [4]. Then compare the rebuilt text against the image line by line — a step most students skip and later regret.
Once you have a clean text file, run the check on that file rather than on the image, and read the output as diagnostic evidence rather than a grade [2]. Look at which sentences were flagged, whether the flagged passages are the ones you drafted with a model, and whether the similarity report shows matches you did not intend [2]. That reading is what lets you revise before submission instead of after [4].
If rebuilding the file and reading the flags is the part you would rather not do alone, turnitin0 gives you the same Turnitin AI and similarity reports your institution sees, generated from a normal .docx, .pdf, or .txt upload — so the image problem disappears the moment your draft is a document again. Turnitin0 is an independent service and is not affiliated with Turnitin, LLC.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Can I upload a screenshot to Turnitin and get an AI score?
No. The AI writing report is generated from the text of a supported document, so an image with no text layer produces nothing to analyze [1]. Convert the image to a .docx, .pdf, or .txt first.
Does an AI detector tell me if an image was AI-generated?
No. AI writing detection analyzes language patterns in prose and does not evaluate images, charts, or visual provenance [3]. Image forensics is a separate discipline with its own tools and its own uncertainty.
What does the *% in a Turnitin AI report mean?
It means the detected signal fell below Turnitin's 20% confidence threshold, so the exact percentage is withheld and the result should be read as a low-confidence indicator rather than a definitive score [1].
Will OCR text give the same report as my original file?
Usually not exactly. Extracted text can lose formatting, footnote structure, and citation markers, which weakens similarity matching — so the original editable document is always the better input [2].
What if my draft is only in a chat window as an image export?
Rebuild it as a document, restore headings and citations by hand, verify numbers against the image, and then check the rebuilt file so the flags you see correspond to real sentences you can revise [4].