Using the AIL (AI Influence Level) on the blog

I’m experimenting with being more transparent about AI usage on this site. Not in a performative “look how ethical I am” way, but because I use AI tools differently across different posts and you deserve to know what you’re reading.

I’m trying out the AI Influence Level (AIL) framework created by Daniel Miessler. It’s a simple 0-5 scale that rates how much AI touched a piece of content:

• AIL 0: No AI involved at all – handwritten, typed from scratch, pure human
• AIL 1: Minor AI help – grammar fixes, sentence cleanup
• AIL 2: Major AI augmentation – significant edits or expansions
• AIL 3: AI created from full human structure and outline
• AIL 4: AI created from basic human idea
• AIL 5: Almost entirely AI-generated with minimal human input

What This Looks Like Here (Right Now)

Most posts on this site are AIL 0 or AIL 1. I write them myself, sometimes run them through Claude for grammar cleanup or to make sure my explanations are clear. That’s it.

This post? AIL 2. I outlined what I wanted to say, Claude helped fill in sections and smooth out the writing, and I edited it to match my voice. The structure and ideas are mine, but AI did meaningful work on the prose.

When I write technical posts that include code snippets, I often use AI to add verbose comments – stuff I wouldn’t naturally write but that makes the code more understandable for readers. That’s usually AIL 1.

I’ll update this section as my patterns change. Check the lastModified date at the top to see when I last revised my thinking.

Why I’m Trying This

Because I think readers deserve to know. Not in a hand-wringing “AI is scary” way, but in a “here’s what you’re actually reading” way.

I might change my mind about this framework. Maybe the ratings will feel too granular, or not granular enough. Maybe I’ll find better ways to communicate this. That’s fine – this is an experiment, and I’ll update this post as I learn.

I use Claude Code to build my deployment system and CLI tools. I use regular Claude to help explain technical concepts or add documentation comments. GitHub Copilot autocompletes code I’m already typing about 25% of the time. These are useful tools that make me more productive.

But there’s a difference between “Keith wrote this and AI cleaned up the grammar” and “Keith gave AI an outline and it wrote most of the prose.” Both are fine! Both are legitimate uses of AI! But they’re different, and I want to be clear about which is which.

How This Might Change

We’re in this weird moment where AI tools are genuinely useful for specific tasks, but also being oversold as magic solutions to everything. The social safety net work I’m doing at Propel uses AI for parsing messy government websites and identifying error patterns – practical, specific applications that make people’s lives better.

My relationship with these tools will evolve. Maybe I’ll use AI more. Maybe less. Maybe the framework itself will feel wrong after a few months of practice. I’ll update this post as my thinking changes.

For now: when you see an AIL badge on a post, you’ll know AI touched it. When you don’t, that’s me writing from scratch. Simple as that.

Updates & Revisions

November 22, 2025: Initial version. Added AIL framework to the site. Most posts are AIL 0 or 1. This post is AIL 2 – I outlined it, Claude helped with prose, I edited to match my voice.

I’ll add notes here as my approach changes or as I learn what these ratings actually mean in practice.