What is Grammarly Authorship?

A feature that categorises text by where it came from as you write — typed by you, pasted from an online source, or generated by AI — so you can show how a document was produced. Grammarly describes it as part of Grammarly Pro. It records provenance rather than inferring it after the fact.

Is Authorship the same as an AI detector?

No, and the difference is the whole point. A detector reads finished text and estimates how machine-like it is. Authorship watches the document being built and logs where each part came from. One is a guess about the product; the other is a record of the process.

Does Authorship prove I wrote something?

It is considerably better evidence than a detection score, and it is not proof. It records what happened in one editor — text typed elsewhere and pasted in appears as pasted, and work done outside Grammarly is not covered at all. Treat it as one strong item in a drafting record rather than as a certificate.

What does Grammarly claim for its AI detector?

99% detection accuracy and first place on the RAID independent benchmark, both quoted from their own page. Alongside those, and unusually, the same site states repeatedly that no AI detector is 100% accurate and that no detector can conclusively determine whether AI was used.

Should I turn Authorship on?

It depends on whether you would rather have the record or not have it. If your work may be questioned, a provenance log is the strongest evidence available short of full version history. If it may be demanded of you, you have created something that can be asked for. Both are true and only you know which matters more.

Do you compete with Grammarly?

Yes, on the writing side, which is a conflict worth stating before the rest of the page. Every figure here is theirs and linked, and our assessment of Authorship is more favourable than of anything else in this cluster.

Grammarly Authorship, and why it is not a detector

Authorship records where your text came from while you write it. Every detector on this site reads the finished document and guesses. That is the difference between a record and an inference, and it decides what either one can actually establish about your work.

Last reviewed 16 August 2026 · The HumanFlow team

What it does

Grammarly's own description, read from their AI detector page on 16 August 2026: Authorship “automatically categorizes your text based on where it came from (AI, an online database, typed by you, etc.) so that you can easily show your work”. It is part of Grammarly Pro.

That is provenance. It is the same category of evidence as version history in a document editor, and a different category entirely from perplexity, which is what a detector measures. One observes the writing happening. The other examines what is left afterwards and reasons backwards.

Why the distinction decides everything

This site argues, at length, that a detection score cannot establish authorship because it infers process from product — and that the evidence which actually helps is a record of the work being done. Authorship is a commercial product built on that same reasoning, which is unusual enough to be worth pointing at.

The comparison is not close on evidential weight. A detector produces a probability nobody can inspect, from a model that updates without notice, calibrated by a vendor. A provenance log produces a description of what happened in the editor, which can be shown to a person and argued about specifically.

What it still cannot do

It records one editor. Text drafted in a notebook, thought through on paper, or written in a different application and pasted in appears as pasted — indistinguishable, in the log, from text pasted out of a chatbot. Anyone who plans on paper and types up a fair copy produces a record that looks worse than their process was.

It also only covers work done while it is switched on and inside Grammarly. That is not a criticism of the feature; it is the boundary of what any single tool can witness, and the reason a drafting record built from several sources is stronger than any one of them.

And it is not proof. It is a vendor's account of what its own software observed, which is better than an inference and short of a certificate.

The part worth thinking about before switching it on

A record that helps you is a record that exists. Once a provenance log is available, it can be asked for — by an institution, an employer, or a process you did not anticipate when you enabled it.

For most people facing an accusation, having the record is far better than not having it, and we would say so plainly. But the choice is not costless and pages that recommend provenance tools rarely mention the second half. If your use of AI assistance is permitted and disclosed, the log corroborates you. If it is not, the log is not on your side.

What Grammarly claims for its detector

Separately from Authorship, Grammarly ships a free AI detector and claims 99% detection accuracy and first place on RAID's independent benchmark. Both are quoted from their page and neither is verified here.

What is more notable is what sits beside those numbers. The same site states that no AI detector is 100% accurate, that no detector can conclusively or definitively determine whether AI was used to produce text, and that the best detector is “one that's accurate, transparent, and designed for responsible use, not just one that promises perfect results”. A vendor publishing a 99% figure and a disclaimer that strong on the same page is doing something most of this category does not.

Where we stand, since we compete with them

We sell a rewriting tool and Grammarly sells a writing suite, so this is a competitor page and should be read as one. Our assessment of Authorship is nonetheless more favourable than of anything else in this cluster, for a reason that costs us nothing to admit: it is evidence of the kind we spend the rest of the site telling people to keep, and we do not offer an equivalent.

Compared against

Turnitin vs Grammarly Authorship — Grammarly Authorship against Turnitin on what independent testing found, who each is sold to, and what happens to the text you submit.

Claims read from Grammarly's AI detector page on 16 August 2026: grammarly.com/ai-detector. Quoted as their claims, not independently verified.

Common questions

What is Grammarly Authorship?: A feature that categorises text by where it came from as you write — typed by you, pasted from an online source, or generated by AI — so you can show how a document was produced. Grammarly describes it as part of Grammarly Pro. It records provenance rather than inferring it after the fact.
Is Authorship the same as an AI detector?: No, and the difference is the whole point. A detector reads finished text and estimates how machine-like it is. Authorship watches the document being built and logs where each part came from. One is a guess about the product; the other is a record of the process.
Does Authorship prove I wrote something?: It is considerably better evidence than a detection score, and it is not proof. It records what happened in one editor — text typed elsewhere and pasted in appears as pasted, and work done outside Grammarly is not covered at all. Treat it as one strong item in a drafting record rather than as a certificate.
What does Grammarly claim for its AI detector?: 99% detection accuracy and first place on the RAID independent benchmark, both quoted from their own page. Alongside those, and unusually, the same site states repeatedly that no AI detector is 100% accurate and that no detector can conclusively determine whether AI was used.
Should I turn Authorship on?: It depends on whether you would rather have the record or not have it. If your work may be questioned, a provenance log is the strongest evidence available short of full version history. If it may be demanded of you, you have created something that can be asked for. Both are true and only you know which matters more.
Do you compete with Grammarly?: Yes, on the writing side, which is a conflict worth stating before the rest of the page. Every figure here is theirs and linked, and our assessment of Authorship is more favourable than of anything else in this cluster.