Why every summary links to its sources
AI summaries save time, but they can be wrong. Here's how we keep them honest and how to read them.
17 Oct 2026
ReadKernel's summaries are written by AI. That saves you hours, but AI can make mistakes, and we'd rather tell you that up front than pretend otherwise.
The rules the AI works under
- Facts only from the articles. The model sees the articles' own text and nothing else. It is told never to add dates, names, job titles or background from its own memory. Memory goes out of date; the articles are today's.
- No source-speak. Summaries don't say "according to X". The source list sits right under every story instead.
- Short on purpose. Two or three sentences leave less room for invention than a long essay.
- Nothing to summarise, no summary. When an article gives us only a headline (Hacker News links often do), we show the headline alone rather than let the AI pad it.
What we do when it goes wrong
We read the brief ourselves, and when we find a mistake we fix the prompt that caused it, not just that one story. Our first week caught a summary that called a 2026 election "2024" (the model filled a gap from memory), and a political story that slipped into a tech brief. Both rules above exist because of those.
How to read a ReadKernel story
- Read the headline and summary. For most stories that's all you need.
- Check "why it matters". It's our one-line take, so treat it as an opinion.
- Open the coverage before you act. If you're going to make a decision, quote a number, or share it at work, click through to the original articles. On every story page they are one tap away, newest first.
If you spot something wrong, tell us. We'd rather be corrected than quietly wrong.