Amazing: AI Can CLean-Up the "APPENDIX S" AI Policy for AMA CPT Meeting

Several years ago, AMA CPT introduced "Appendix S," a policy statement about language and coding for software-intensive services.   From May 2025 to the upcoming May 2026 CPT meeting, AMA has released multiple very extensive revisions of "Appendix S," one of which will probably be voted on, at the May 1 Boston AMA CPT meeting.

•  See a 2021 Linked In announcement here. See a 2022 publication about the work and its goals, here.  See a 2025 online article by Dr. Frank here.

One recurring problem: AMA releases each new version of Appendix S as an extremely heavy markup of the prior draft. Frankly, I find it all but unreadable. It looks like an ordinary MS Word redline, but it isn’t. You cannot simply click “accept all changes” and produce a clean copy.

Over the last several months, I’ve given successive versions of Appendix S to ChatGPT for assessment and comparison, which has been very helpful. But today I tried something different. I gave it the heavily marked-up mid-April version of Appendix S and asked it to return a clean version. It did so, producing a new, downloadable Word .docx file.

I can’t share the document itself, because access is controlled by AMA CPT and subject to confidentiality restrictions. However, I can show you how ChatGPT interacted with me during the cleanup process, as it turned an almost unreadable markeup into a usable clean copy.

##

Here's what the AMA review document looks like (ouch!);

Here's my dialog with Chat GPT:

If possible, produce a clean copy of this redlined document with changes accepted. Note it is not a normal Word document where changes can be quickly accepted within Word. Producing a clean copy of it will require interpreting the redline markup as it is and then producing a clean copy, if you can. let me know if dividing it in parts (e.g. 3 equal parts) would help.

##

I’m going to inspect the uploaded .docx directly and reconstruct a clean version from the visible redline text, then I’ll give you a revised file to download. If the markup turns out too tangled in a few spots, I’ll still produce the cleanest defensible version and flag any lines that may need a human check.

I’ve confirmed the file is readable as a DOCX, but the visible “redline” is largely baked into the body text rather than behaving like ordinary Track Changes. I’m now reconstructing the accepted version paragraph by paragraph and then I’ll render the result to make sure the layout stayed clean.

I found one useful wrinkle: the document includes an older comparison table that is entirely struck through, so it should disappear in the accepted copy. I’m removing that deleted table, tightening a few places where accepted/deleted fragments left awkward phrasing, and then I’ll QA the rendered pages.

[Thought for 5 minutes 47 seconds]

I produced a clean accepted copy from your uploaded redline document

[copy of upload link here]

Download the clean DOCX [this is also a cloud link]

A couple of notes:

• I treated the visible markup as acceptance of the inserted text and removal of struck text.
• I also removed the fully struck comparison table that appeared to be deleted content.
• One table row still flows onto a final short page; the content is intact, just split across pages.

If you want, I can next do a second-pass version in 3 sections to polish any judgment-call wording line by line.