What is this site?
We downloaded public writing from candidates in the Seattle 2026 General Election and ran it through Pangram, a high-quality AI detector. We found that many candidates likely used AI tools, and that they may have been not consistently candid about it.
This site presents the Pangram results, along with anything we found a candidate saying about their use of AI.
How reliable is Pangram?
Many AI detectors are snake oil. Pangram is not - in particular, it is very rare that it flags human text as AI.
We could quote you some difficult-to-interpret numbers here, but we personally find simple demos more compelling. Compare these two Pangram results:
- 2020 General Election Voters' Pamphlet (State, King County Edition)
- 2026 edition of the same thing (i.e. the current one)
The 2020 guide was written before ChatGPT came out in 2022. You can see it scores exactly 0% AI, despite being pretty long. Meanwhile, the 2026 edition has a handful of flagged segments, all of them campaign-supplied text.
Generally we recommend just trying Pangram on whatever text you have that you know to be human- or AI-written; the results are easier to understand than technical data. But if you want rigor:
Another detail: Pangram does a sliding "window" over many words in a text, so it often doesn't get the boundary between human and AI text segments quite right, and shows lower confidence scores around the edge. This is especially visible in the questionnaires returned by highly-AI-scored candidates. The Urbanist's human-written questions are often partially tagged as low-confidence AI because AI text surrounds them. We think this visible interleaving both shows that Pangram is credible and demonstrates its limitations.
Did you use a generative artificial intelligence (AI) platform to craft some or all of your responses? What was the extent of your use?
This site was made with extensive AI help under close human supervision. It took about 25 hours of human labor.
The texts were collected by an AI. They were spot-checked by the humans many times, though at no point were they exhaustively reviewed by humans. Some simple programmatic checks were used to guard against hallucinations.
The site code is pretty much 100% vibecoded (i.e. written by AI) under instruction from human software engineers.
All nontrivial site text is human-authored. Run it through Pangram to check!
Obviously Pangram itself is an AI tool, though not generative AI.
Do you object to candidates using AI?
The short answer is: no, not in general. We use AI extensively and, we like to think, appropriately.
We do have some concerns about AI use by campaigns, but we won't say much more about them here. We'd prefer this site to neutrally document the fact of AI use and how candid campaigns are about it. Our goal is mostly to help people have an informed discussion; we are all feeling our way through this weird future together.
Does a high AI score mean a candidate didn't write it?
It's evidence. Briefly, a high AI score means the text strongly resembles AI-generated text, a style humans rarely write on their own.
Even when a candidate did use AI, it's tricky to tell the difference between using AI for editing, letting AI do the writing in a close back-and-forth with a human to get the message just right, and letting AI churn out the whole thing unsupervised. Some AI editing tools like Grammarly also make text look more AI-like.
That being said, usually AI-edited text shows up as a mixture of human and AI, or "AI-assisted", not purely AI.
See this article from Pangram for more.
Who made this?
Technically adept Seattle voters unaffiliated with Pangram or any candidate.
Are you tracking me?
We use Google and Cloudflare analytics to understand how many people are seeing the site, how they interact with it, where our traffic is coming from, what kinds of devices they're viewing with, and similar. We don't see who you are as individuals, nor do we provide data to third parties (beyond whatever Google and Cloudflare are doing with it internally).
We do, of course, use cookies for this.
Here's Google's explanation of what they do with your data.