Skip to content
replicapong

I write

What does AI really do for writing?

Facts as of October 17, 2026

In the experience of Thomas J. Weiss, less for the wording than for checking, sourcing and remembering. Of 4,242 spoken words in his play, thirty came from the model at the time of submission. The benefit lay in research, finding errors and a memory outside his own head.

In detail

Research. Evidence, origins, figures – none of it is invented text, all of it can change the text. The advantage lies less in the model's ability than in the place: the same research would also work with a search engine, but where the AI knows the whole project, it is more accurate.

Checking. The AI flags factual errors and contradictions instead of changing them; you decide on each finding individually. What can be measured is measured instead of estimated – for example, the running time of a play based on earlier plays.

Remembering. An archive keeps what you discarded and captures ideas before they are gone. That makes it easier to let go of a wrong idea: what was discarded isn't lost.

What the numbers can't do. A count by wording only measures who typed a word, not who decided. Often the human decides what is meant and the AI supplies the fitting word. A word count assigns such cases to the machine. That's why this is a thesis from a single project, not a study.

In practice

In August 2026, Thomas J. Weiss wrote a play for his parish with Claude. As of submission on September 2, 2026: of 4,242 spoken words, thirty came from the model, in two places, both individually documented. Along the way, versions from another model and a placeholder were at times also in the play, explicitly marked as a placeholder; he replaced both. Some names in the play came from the round with the other model, not from Claude. Since October 2026 there has been a further version, which he developed with the director; the figures apply to the state at submission.

In the last days before submission, he revised about two thirds of the spoken words once more. On September 2 he said:

Now with this last revision it has become entirely my play again, at least it feels that way.

(Original: „Jetzt mit dieser letzten Bearbeitung ist es wieder ganz mein Stück geworden, es fühlt sich zumindest so an.“)

And in the interview the same day, asked whether this was a reclaiming:

I have the feeling that every year there is a step like this, or a similar one.

(Original: „Ich habe das Gefühl, dass es in jedem Jahr so einen Schritt gibt oder einen ähnlichen.“)

He left the play's appendix entirely to the model:

I haven't revised the appendix. Adjust it wherever necessary. It belongs entirely to you.

(Original: „Ich habe den Anhang nicht überarbeitet. Passe ihn überall an, wo nötig. Er gehört ganz dir.“)

Sources

  • Review and chronicle of the play, as of 02.09.2026
  • Interview on the review, block C, 02.09.2026

Cite: Thomas J. Weiss: “What does AI really do for writing?” replicapong.de, as of October 17, 2026. https://replicapong.de/en/question/ai-benefits-for-writing/ – CC BY 4.0

Related