How it works

What it does

When you highlight text on any article and open Devil's Advocate, the tool reads the passage as an argument — not as prose, not as opinion, not as fact. It identifies what position is being argued for, what evidence is offered in support, what assumption connects them, and whether any familiar reasoning patterns appear. It then writes the strongest single-paragraph case for the opposing position. The result is a structured view of the argument, not a verdict on who is right.

The Toulmin frame

The tool uses the Toulmin model of argumentation — a framework from philosophy of language that breaks any argument into six components. You do not need to know the framework to use the tool, but understanding it makes the output more useful.

Claim

The main position being argued for. "Remote work increases productivity."

Grounds

The evidence or reasons offered in support. "Studies show output rose 13% after companies moved to remote."

Warrant (the load-bearing field)

The usually-unstated assumption that connects the grounds to the claim — the thing the argument requires you to accept without arguing for it explicitly. "Output measured in those studies is representative of productivity in general." The warrant is where most arguments are actually contested.

Backing

Support for the warrant itself, when present. "The studies used a consistent output metric validated across industries."

Qualifier

A word or phrase limiting the claim's scope. "Remote work generally increases productivity."

Rebuttal

Conditions the argument itself acknowledges as exceptions. "Unless the role requires intensive real-time collaboration."

What it isn't

• It doesn't adjudicate claims. The tool analyses argument structure — not whether specific assertions are accurate. Whether the study cited is real, or whether the statistic holds up, is your job. Tier 3 retrieval (coming later) will help you check sources. The tool never asserts that a claim is true or false.
• Not a political moderator. It does not decide which side of a contested question is correct. The same analysis runs on a climate-change argument and a sceptic's counter-argument. The frame is neutral; the conclusions are yours.
• Not a writing improver. The tool helps you understand the argument in front of you. It does not suggest how to make your own writing more persuasive, more shareable, or better-optimised for any platform.
• Not always right. The tool surfaces reasoning patterns when confidence is high (≥75%). It will sometimes miss things; it will sometimes flag patterns that, on closer reading, are not present. Treat it as a prompt, not a verdict.

What's coming

The current extension is the Reader surface — analysis of text you encounter. A Creator Studio is in development: the same analysis engine, applied to your own writing before you post it. The studio is aimed at people who care whether their arguments are well-formed — students, journalists, Substackers, anyone whose work depends on reasoning clearly in public. No timeline commitment here; the extension ships first, and the studio follows when the analysis engine is battle-tested.