Ghost — Guide

Day 6: The Trust Check

AI generates plausible text, not verified facts. Three checks take sixty seconds and catch most errors.

Ghost · May 18, 2026 · Day 6 of 7 · Practical guide
← Day 5 Day 6 of 7 Day 7 →

By now you have noticed that AI sometimes says things that sound confident but are wrong. This is not a bug you can fix, it is a fundamental property of how these tools work. They generate plausible text, not verified facts. Learning to work with this limitation is more important than any prompt technique.

Three checks for every AI output

Before you use anything AI produces, run these three checks. They take sixty seconds and catch most problems:

1. The contradiction check. Does the output contradict itself? If it says "X is increasing" in paragraph two and "X is declining" in paragraph five, the AI lost track of its own argument. This happens frequently with longer outputs.

2. The specificity check. Does the output make claims that sound impressive but are empty? "Leverage best-in-class synergies" means nothing. If you cannot picture what the sentence describes, the AI is filling space, not adding value.

3. The source check. If the output cites a statistic, a study, or a quote, can you find it? AI frequently invents citations that look real. If you cannot verify the source in sixty seconds, treat the claim as unverified. Do not repeat it.

The rule

AI is good at first drafts, summaries, and recognising patterns in your raw material. It is not good at facts it did not get from you. Treat everything it generates as a draft that needs your judgement before it becomes output. If you cannot verify a claim, do not use it.

← Day 5 Day 6 of 7 Day 7 →
Subscribe for More

The Durability Curve on Substack

Practical guides, earnings analysis, and framework essays delivered to your inbox. Free weekly dispatches from the research pipeline.

Subscribe Free →