What it does
One PDF, one AI, the whole note
I finished a Sunday bike ride, sat on the back patio with a beer, and looked at the porch swing I've been meaning to fix. The paint is peeling. A bolt is working loose. I took four photos, wrote a short note, and recorded the ride.
Then I shared the note as a PDF and handed that one file to an AI. Then another, and another. On the web, in an iPhone app, and running locally. Different models, different places. They all got the porch swing, the stories I wrote on the photos, and the ride. I didn't paste the photos in one by one, or explain which words went with which bolt.
Two layers
The pages are a normal field report. A map of the ride, the stats, the note, the place, the weather, and the four photos. Anyone can open it and read it. The line under each photo is the time it was taken.
The words I wrote about each photo live in the data underneath, with the location, the weather, and the full ride track. That data is the note's GeoJSON, embedded in the same file, and the original photos ride along as attachments. A person reads the pages. A tool that opens the attachments gets the rest.
Pull them out on your own computer with pdfdetach -saveall "note.pdf". Then the .geojson and the photos are ordinary files. Share is a PDF is the longer version of how that works.
Why they could read it
I didn't teach them anything, and I didn't change how a PDF works. Embedding a file inside a PDF is an old part of the format, a little pocket that has been there for about twenty years. The share puts the GeoJSON and the photos in that pocket. The models already knew how to open it.
What this export left out
This was the share-sized PDF, the one meant for Messages and email. Live Photo motion did not come along. Neither does RAW, or a photo big enough to blow past the size limit. The file says so. manifest.json inside it names whatever was left out. The archive export is the one that keeps those originals, and the file gets big when they do. That is a choice about what to pack, not about whether a model can read the file.
Why I made it this way
Usually this would be a folder of photos, a text file, and a note from me explaining what goes with what. The PDF was already the package. I collected it on the phone and handed it over later.
I didn't build the PDF share so I could talk to AIs. I built it so a note could leave my phone and still be complete — readable by anyone, importable by the app, and owned by me.
That completeness is what made this work.