How we work

How we use AI

PrepHabit is researched and written with the help of AI models. AI can make mistakes, so every claim carries its source and we correct errors on the page where they appeared. This page says exactly what the models do, what a person reviews, and what AI is never allowed to publish.

What AI does here

Three jobs, kept apart on purpose, so the model that writes a claim is never the one that checks it.

  1. Step 1

    Watch official feeds

    Software and AI models watch National Weather Service alerts, National Hurricane Center and USGS feeds, FAA statements, CISA advisories, the status pages of affected companies, and Reuters and AP. A new notice starts a draft. It never confirms anything by itself.

  2. Step 2

    Draft cards and guides

    A model writes the first version of a tracker card or a guide, and has to attach a source to every claim it makes.

  3. Step 3

    Check each claim, separately

    A second model, in a separate step, traces each material claim back to its original evidence and checks the scope, the date, the place and how certain the source actually is.

  • Two outlets repeating one wire story, or three articles repeating one company statement, count as one source.
  • The tracker summarizes named official updates and verified reporting. It does not claim to verify events independently in real time.

Safety instructions never come from a model

Every safety instruction has a named authority behind it

What to do in a gas leak, how much water to store, how to run a generator safely: each instruction comes from a named authority such as Ready.gov, the CDC, the American Red Cross, the National Weather Service or your utility, and the page links to it with the date we read it.

A model may put that instruction into plain words. It may not add one of its own.

No AI product testing

Nothing on this site was tested hands-on, by a person or by AI. Product picks come from published specifications and official guidance, and we never present AI-written text as a hands-on review.

What a person reviews

Verified facts from official sources can publish without waiting. These cannot. A person approves each one by what it says, whatever label it carries.

Before it publishes

  • Any statement that a named person, company, contractor or state is responsible for an event, even when it is called preliminary.
  • A Disputed cause.
  • A Contradicted status on a claim about a named person or organization.
  • Official instructions that conflict with each other.
  • New medical or evacuation advice.

After it publishes

A person reads every error report sent through the contact form. The site's rules, including the ones on this page, are set by people and are not something a model can change.

Checks reduce mistakes. They do not remove them. If a page matters to a decision you are making, open the source it links to.

The corrections page explains how to report an error.

What AI is never allowed to do

AI never publishes these by itself

  • Casualty estimatesNumbers of dead or injured are not published automatically.
  • Safety instructions without an official sourceIf no authority has issued the instruction, the card leaves it out.
  • Nationwide threats sourced from social postsA social post is never a source on its own.
  • Who is responsibleAny attribution of responsibility to a named person, company or state waits for a person to review it.

What we say about AI as a cause

No public outage so far has been traced to an AI acting on its own.

We treat that sentence as a fact to keep checking. The tracker has an "AI and automation" category for events where AI is involved or blamed. If a named source that examined the evidence traces an outage to an AI acting on its own, the tracker will say so and this sentence will change.

See what's happening

The same rule applies to us

Our tagline, "Emergency prep for the AI era", names the period we live in and makes no claim about what causes outages. We never write that AI caused an outage or an attack unless a named source confirms it, however often the claim is repeated online.