Responsible use

How to use QED and Sic well. These points are also part of our terms.

1Writing with QED

1. The content is yours. QED writes from your notes. You are responsible for every claim in the result, as you would be for a paragraph a co-author drafted.
2. Check every output. QED can drop a number, add an aside or firm up a hedge. Run compare, then read the passage against your notes.
3. Use your own voice. Supply passages you wrote, or have the right to use. Do not supply another person's writing in order to pass text off as theirs.
4. Say that you used it. Text written with QED is AI-assisted. Follow your journal's policy; the API will draft the statement.
5. A model is not an author. It cannot take responsibility for a paper, so it does not go on the byline. Credit it where contributors are credited.

2Reading with Sic

1. A score is evidence, not a verdict. Sic does not know who wrote a text. At its strict threshold it still flags about 1 in 2,000 human research papers and 1 in 100 human commentaries.
2. Never act on a score alone. Do not accuse, grade, reject or report a person on the strength of Sic's output. Use it to decide what to read more closely and whom to ask.
3. Stay inside its scope. Sic is calibrated on English scientific prose. We have not measured it on student essays, on other genres or on other languages, and in those settings its error rates are unknown.
4. A low score proves nothing either. Light edits, short insertions and text from models Sic has not seen are often missed.
5. Quote the error rates. If you report a Sic score to someone else, report the threshold and its false-positive rate with it. Every response carries them for that purpose.

3What we will not help with

We do not offer a way to make text score lower on a detector, ours or anyone's, and the tools tell agents not to attempt it.

Accounts used to impersonate a named person, or to make accusations against people on the basis of scores alone, may be suspended.