Working with AI
A short guide for using ChatGPT, Claude, and similar tools with this site.
This site is built to be readable by humans first, then legible to machines. Every post has a clean Markdown companion, the site has an LLM index, and post pages include a Copy docs menu so you can bring the writing into your own tool without scraping the page.
This page is the human part of that system: what the affordances do, how to prompt well, and where I do not want the machine to blur the line between my framing and yours.
Affordances at a Glance
/read/<slug>.mdexists for every post as the clean source for AI tools.- The Copy docs button on each post copies that Markdown companion.
- Open in ChatGPT and Open in Claude deep-links send a pre-framed prompt with the post URL.
/llms.txtis the site-wide index for AI crawlers and tools that honor the llmstxt convention.
Recommended Prompts
Treat these as starting points, not formulas. The point is not to make the model sound certain; the point is to make it useful enough for you to think against.
Summarize the argument in this post, then list 3 places the author may be over-claiming or under-defining terms.Pull out the actionable steps. Rewrite them as a checklist I can adapt, then mark which steps depend on context.What in this would not generalize to <my context>? Name the assumptions I should test before applying it.What Works, What Does Not
Works well:
- Synthesizing across posts and finding recurring patterns.
- Asking clarifying questions when a paragraph feels dense or too compressed.
- Comparing arguments: where two posts reinforce, contradict, or complicate each other.
- Surfacing implicit assumptions, missing definitions, and practical next steps.
Does not work yet:
- Citing specific paragraphs reliably. Models still invent anchors, quote marks, and line numbers.
- Preserving my voice on rewrite. Useful for rough translation; bad for sounding like me.
- Making factual claims about my life that are not in the text.
- Treating a generated summary as a replacement for reading the original.
Attribution and Ethics
If you publish something derived from this material, link back to the original page. Do not present my framing, examples, or lived context as if they started with you. Build from it; do not launder it.
I write longhand. These AI affordances are reader-side collaboration tools, not a claim that the posts were produced by a model. Use the tools to question, translate, summarize, and adapt the work. Keep the source attached.
Coming
None of this is a promise. These are the directions I am considering as the site grows from a bridge into a more deliberate archive.
- Per-post embeddings downloads for local search and comparison.
- An MCP endpoint for tools that want structured access instead of scraped pages.
- Structured citation files so AI tools can point back to stable sections more honestly.
- Conversation-replay exports for readers who want to show how an interpretation was formed.
- A fine-tune-ready corpus dump, if I can make the consent and attribution boundaries clean.
Feedback
If an AI workflow helps you understand, challenge, or responsibly apply something here, I want to know. If it misleads you, I especially want to know.