Why your AI writing score is missing, inconsistent, or will not download

Turnitin publishes known-issues pages, and several live entries concern the AI writing indicator directly. Among them: the score shown in the assignment inbox can differ slightly from the score inside the full report because of rounding at different stages; during processing the interface can display an "unavailable" message giving reasons that do not apply; the indicator can be hidden entirely by a browser issue affecting Safari; and downloading the AI report can fail depending on the user's location. None of these are statements about your writing.

There are three of them, all public and none behind a login: a general known-issues page, one for add-on products (which is where AI writing detection lives, under the Originality add-on), and one for LMS integrations.

They are dated. The general page carried "Similarity Report issues as of: 27 July 2026" when we read it; the add-on page said "All Originality add-on issues as of: 1 July 2026"; the integrations page said it had been updated 20 days earlier. That dating is the important part, and it cuts both ways — anything below may be fixed by the time you read this, and new entries may have appeared. The pages are the source, not this article.

One more number worth noticing: the general page had been marked helpful by 60 out of 408 people who voted. Make of that what you like about how it reads to someone arriving in a panic.

This one explains a specific and maddening experience, and it is documented in one paragraph:

"We're aware of an issue where users may notice a small discrepancy between the AI writing detection score shown in the assignment inbox and the score displayed within the full report. This difference is caused by the way scores are rounded at different stages of the calculation and display process."

Turnitin adds that "Although the discrepancy is minor, it may cause confusion when comparing scores across the interface", and then that "We are working to implement a resolution as soon as possible." That last sentence is also the shelf life of this entry: a vendor that says it is fixing something is a vendor whose known-issues page will stop saying it.

So if you have been told one number and then seen a different one, the difference may be entirely presentational. That is worth knowing before anyone builds an argument on a one-point gap, in either direction.

The entry to know if you have ever been told your submission was excluded:

"We're aware of an issue where, during the AI writing detection processing period, users may briefly see an error message stating that 'AI writing detection is unavailable for this submission.' This message includes reasons such as 'excluded by instructor,' 'unsupported file type,' or 'resubmission excluded,' even though these do not apply to the submission in question."

And the resolution: "Once processing is complete, the expected AI score and report are displayed as normal."

Three specific, plausible, alarming reasons — all of which the vendor says can appear while nothing of the kind is true. If someone has quoted one of those phrases to you, it is worth asking when they looked.

Two documented causes, neither of them about your document.

A browser issue. "There's a known issue that is causing the AI writing administrator setting and the AI writing indicator to be hidden from users. This is a browser issue primarily impacting users on the Safari browser." The published workaround is to clear the cache, and to make sure pop-ups from Turnitin are allowed.

A route through the Authorship report. A separate entry describes non-US Feedback Studio with Originality users being unable to access AI writing detection when launching the document viewer from the Authorship Report.

If an instructor tells you the AI report "isn't available", these are worth knowing about, because both are fixed by doing something different rather than by concluding anything.

The general page carries an entry titled "AI download issue based on geographical location":

"We are aware that a user's geographical location, specifically mainland China, may prevent the download of the AI report. Users of the Classic Standard assignment may also encounter an issue where the score does not appear in the Classic Similarity Report. Our engineering team is aware and actively exploring a solution."

This matters for a large group of people, and it is easy to misread as a problem with the submission. It is a documented access problem with a named location and an open resolution status.

Two neighbouring download entries are worth a glance for the same reason. A period in the filename — not counting the extension — can cause a file to be left out of a bulk download. And downloads of the new Similarity Report view can produce character errors in non-Latin languages, with a separate entry for formatting errors in Arabic.

Slightly outside the AI indicator but in the same report, and it answers a question people ask:

"Submitted documents must contain at least 500 characters to enable the Flags feature in the classic and new Similarity Report experience."

The known issue attached is that submissions below that threshold can incorrectly display a Flag icon in the assignment inbox, and opening it produces an error. The published workaround is to check whether the submission meets the 500-character requirement first, and if it does not, "the flag report can be disregarded".

1. Check the pages before you argue about a number. They are public, dated, and take two minutes. If your exact symptom is listed, you have a much better opening than a theory.
2. Quote the entry, not the conclusion. "There is a known issue where the inbox score and the report score differ because of rounding" is checkable. "The tool is unreliable" is not, and invites an argument you cannot win.
3. Note the date. These entries get resolved — there is a Resolved tab on the pages. An entry that was live when you were flagged may be gone when someone else looks.
4. Do not stretch it. A rounding discrepancy between two displays says nothing about whether the underlying score is right, and a download failure says nothing at all about your text. Claiming more than the entry says is how a good point gets discarded.
5. Separate this from reading the report itself. What each part of the report means is a different question, and we go through it in how to read a Turnitin AI writing report. The two numbers on a submission are also independent of each other, which we set out in the AI and Similarity reports side by side.

Most of the time it will be, and then the useful object is the set of highlighted passages rather than the total.

HumanPen works from exactly that: give it the document and the report, and the passages carrying highlights are the ones it edits, while everything unmarked keeps the wording you wrote.

None of that helps with a missing score or a failed download. For those, the pages above are the whole answer we have.

The score in the inbox and the score in the report are different. Which is right? Turnitin documents this as a known issue caused by rounding at different stages of calculation and display, and describes the discrepancy as minor. It is not a sign that one of them is wrong about your writing.

My report says AI detection is unavailable because of an unsupported file type. Is that real? It may not be. Turnitin documents a known issue where that message, and reasons including "excluded by instructor" and "resubmission excluded", appear during processing even when they do not apply to the submission.

Why can't the AI report be downloaded? One published entry attributes download failures to the user's geographical location, naming mainland China specifically, with a resolution described as being explored.

Are these issues permanent? No. The pages are dated, carry a Resolved tab, and change. Check the current page rather than relying on any summary of it, including this one.