Plimsoll

The problem

AI output can go wrong in four common ways:

• Hallucination — it invents a fact, a quote, or a source.
• Sycophancy — it tells you what you want to hear.
• Conflation — it mixes two true things into one false thing.
• Drift — later lines wander off what the start got right.

Those are the risks of using AI without checking the output. The model does not know it is doing this. You cannot ask it, and trust the answer.

Two short examples. Deloitte’s 2025 health-staffing report for Newfoundland and Labrador cited papers that were not real. In 2023 a New York lawyer filed ChatGPT case citations that did not exist (Mata v. Avianca).

Consequences

• Loss of credibility.
• Loss of a customer.
• Loss of revenue.

The solution

Plimsoll. Paste the text. Get a list of what we could verify and what we could not, with clickable citations so you can check it yourself.

How it works

1. We pull claims from your paste (quotes, patterns, and a model that lists claims only).
2. We check those claims against pages you linked, then against Brave Search.
3. The model does not grade the claims. Overlap with a real page does.
4. You get green / yellow / red, plus links.

What you get

• Verified items — we found support on a linked page or in a search result.
• Unverified items — we could not confirm them. Look before you approve.
• Clickable citations on both, so you can open the source and get out.

Experimental. Not a substitute for a lawyer or an expert. Green is support, not a stamp.

How to run it

1. Paste your document, or drop a file: PDF (including scans), Word (.docx), TXT, MD, JPG, PNG. A public Google Doc link works too.
2. Optional: drop extra source text.
3. Press CHECK THIS.
4. Read the list. Click the citations. Decide for yourself.