GPT-5 Pro detector
What the GPT-5 Pro detector does to a long draft
Reasoning-tier output has a particular problem: it is long, it is organised, and it is wrong in ways that are hard to see because the organisation is doing so much work. Paste it here and you get back a list of the places the writing reads as generated, with the reason attached to each one.
That makes it a triage pass rather than a verdict. The reasons tell you which sections to cut, which to rewrite and which were fine all along. On a 2,000-word synthesised answer, knowing where to start is most of the job.
Workflow
How to run the GPT-5 Pro detector on a draft in three steps
Under a minute per pass, and you can re-run after editing if you have runs left.
- Step 1
- Paste the draft, or the section you are least confident about. Up to 12,000 characters per run.
- Step 2
- Run the check. The text goes to our server, a model reads it, the findings come back as a list.
- Step 3
- Work down the list. Act on the flags you agree with, ignore the ones that misread your draft, and re-run only if you changed a lot.
Background
What GPT-5 Pro is, and why its drafts look different
GPT-5 Pro is the top compute tier of the GPT-5 family, above the standard model and above Thinking mode. The explainers that rank for the term describe it running several reasoning threads in parallel instead of one linear pass, taking noticeably longer, making fewer major errors on hard problems, and fanning out multiple searches or tool calls to assemble a synthesised report.
Here is the part worth carrying into your editing. None of those explainers describes a distinct Pro writing style, and that silence is informative: the tier changes the reasoning, not the prose. What you actually see in a draft is scale, length, section depth and completeness, sitting on top of the same sentence-level habits GPT-5 already had.
Triage
What to do with each GPT-5 Pro flag
Reasoning-tier drafts throw a particular set of flags, and each one has an obvious edit behind it.
- Completeness flags
- A section covering an angle nobody asked about. Delete it. Length added by a model is not length your reader agreed to.
- Structure flags
- Headings and nested lists on something that is not a report. Flatten it into prose and the machine shape goes with it.
- Even-weighting flags
- Every section the same length. Let the important one run long and cut the rest hard.
- Hedge flags
- A qualifier in front of a claim that was not in doubt. Delete the qualifier, keep the claim.
- Rhythm flags
- Sentences all landing at the same length. Merge two, then cut one to a single line.
- Flags you disagree with
- Read the stated reason. If it does not describe your sentence, leave the sentence as it is.
Straight answer
Can this tell GPT-5 Pro from GPT-5 or Thinking mode?
No. A tier-specific detector would need the tier to leave a signature in the prose, and nothing published says it does. Detection research also reports that classifiers trained on one model transfer poorly to another, which is the harder version of the same problem.
The model running this check is instructed never to claim which system produced a text. In practice that costs you nothing: if you drafted with a reasoning model, the findings will describe your draft accurately, and the model name was never the thing you needed to know.
Why flags happen
Why a human draft gets flagged as GPT-5 Pro
There is a shape that draws flags regardless of who produced it: short, general and essay-structured. A detector that publishes its own false-positive guidance names those three factors, and they follow directly from how the signals work. Less text means less to read, and a standard structure is standard for everyone.
Long, thorough, evenly organised writing is also exactly what a good technical writer produces deliberately, and what many second-language English writers produce because a level register is the one they trust. So a flag is a statement about prose. If you are on the receiving end of one, the answer is process, drafts, notes and edit history, not an argument about scores.
Allowance
Checks per day and the 12,000 character limit
You get 5 checks per rolling 24 hours, tracked per browser with no account and no e-mail. The window rolls continuously rather than resetting overnight, so runs come back as the oldest ones expire.
Each run takes 1 to 12,000 characters, which is generous next to several free checkers, one of which caps free queries at 2,000. Even so, reasoning-tier drafts run long. For anything over the limit, check the opening and the conclusion first, since structural patterns concentrate at both ends.
Who it is for
Who runs the GPT-5 Pro detector, and when
Analysts turning a long synthesised answer into something a colleague will read. Content leads reviewing research-heavy drafts. Students who used a reasoning model for a literature summary. Anyone who has looked at a 2,000-word answer and thought it covers everything and says nothing.
The moment to run it is after the argument is settled and before anyone else sees the draft. Run it earlier and you spend checks on sections you were going to cut.