| Capability (book chapter) | ChatGPT | Claude | Gemini |
|---|---|---|---|
| Photo of a page → runs the prompt (everywhere) | Yes, free | Yes, free | Yes, free |
| Projects / saved instructions (Ch 2 Context File) | Yes | Yes | Saved Info |
| Voice input for dictation (Ch 4 Ten-Minute Database) | Yes | Yes | Yes |
| Inbox / calendar connectors (Ch 5 Route C) | Paid | Yes (read/draft) | Native if Gmail |
| Deep research mode (Ch 10 quarterly deep-dive) | Paid | Paid | Paid |
| Scheduled / recurring tasks (Ch 13 rhythm) | Paid, limited | Paid | Paid, limited |
| Teach-by-screen-recording (Ch 2, Teach It Once) | Developer tool | Paid, desktop app | Not yet |
| Browser drives websites, e.g. your MLS (Ch 2 advanced lane) | Paid, agent mode | Paid, Chrome ext. | In flux |
| Designed PDFs / documents (Ch 7 packet, pricing story) | Varies | Yes (artifacts) | Varies |
| Image generation for marketing graphics (Ch 9) | Yes | No | Yes |
Nothing in the book breaks. Every capability has a manual lane: paste instead of connect, copy instead of schedule, ask for a structured table instead of a designed PDF. The standing rule from the Machine Pages applies: the assistant should say what it can't do, produce the structured output anyway, and tell you where to keep it.
Ask your assistant these exact questions — it knows its own current menus better than any screenshot:
"Where do I save standing instructions you'll always follow?" · "How do I connect my email?" · "Can you run a task on a schedule?" · "Can I teach you a workflow by recording my screen?" · "Where are your data controls, and what's the strongest privacy setting on my plan?"