All-Seeing AI

Read arguments closely. Write them better.

All-Seeing AI reads a text closely and marks where the reasoning deserves a second look, with the exact words quoted. It can check whether a conclusion follows, review your own drafts, and find passages in your own papers to cite. You see the evidence and make the call.

Sign in with Google or an email link, no password. Your analyses and citation library stay private to your account.

What an analysis looks like

An example report on a short opinion piece. Scroll inside the window, and point at a finding to see the words it quotes.

Opinion piece · 6 sentences

5 passages deserve a second look

  • 4 fallacies
  • 1 technique
  • 1 claim to verify

Keep Market Street open

City Hall wants to replace the Market Street car lanes with bike lanes. Everyone I talk to thinks this is a terrible idea. The council member pushing it doesn’t even own a car, so of course she wants to punish drivers. If we give up these lanes, next they will close the whole centre to cars and the shops will die. Either we keep the lanes or we let downtown turn into a ghost town. Real residents deserve better than this radical experiment.

  1. FallacyAppeal to majority0.86

    Everyone I talk to thinks this is a terrible idea.

    How many people agree is offered as the reason, not anything about the lanes themselves.

    Classifier · Span tagger

  2. FallacyAd hominem0.91

    doesn’t even own a car, so of course she wants to punish drivers.

    Attacks the council member’s circumstances and motive instead of the proposal.

    Classifier · Span tagger

  3. FallacySlippery slope0.88

    next they will close the whole centre to cars and the shops will die.

    A chain of worse outcomes is asserted, with no step shown to follow from the one before.

    Classifier

  4. FallacyFalse dilemma0.84

    Either we keep the lanes or we let downtown turn into a ghost town.

    Only two outcomes are offered; shared lanes, a trial period or other changes are left out.

    Classifier · Wording rule

  5. TechniqueLoaded language0.72

    radical experiment

    The label carries the judgement, so the reader is asked to reject the plan before weighing it.

    Span tagger

Logic check

The conclusion does not follow

It needs a premise the text never states: losing car lanes drives shoppers away.

Claims to verify

1 claim, no source given

“The shops will die” is a prediction stated as a fact.

Findings are candidates for you to review, not verdicts.

Tools

One workspace for examining other people's arguments and building your own.

Analyse

Fallacies and propaganda techniques

Paste an article, speech or post. Every finding quotes the exact wording, shows a probability, and links to a definition of the technique so you can judge it yourself.

Logic

Does the conclusion follow?

The argument's premises and conclusion are written out in formal logic and a theorem prover decides whether the conclusion follows. If it does not, it looks for the premise that is missing.

Review

A critic for your draft

Get a reviewer's objections, the claims you should verify and a whole-text logic check. Accept or dismiss each point as you work through it.

Editor

Argument editor

Write in a structured editor, run the check when you choose to, and save drafts only when you ask.

Sources

Bring a link or a file

Open an article by its link, or upload a PDF, Word, EPUB, LaTeX, HTML or subtitle file. Audio and video links can be transcribed in the Transcript tool.

Watch

Follow feeds and pages

Watch a feed or a page of links. New items are analysed every 30 minutes and flagged when the density of findings crosses your threshold.

Batch

Many texts at once

Upload a CSV of up to 200 texts, run the same checks on each, and download the results as a CSV.

Citations

Cite from your own library

Upload your PDFs, find where each claim is supported, and get LaTeX or plain-text citations with a BibTeX export. Unsupported claims are reported as such, never given a made-up source.

Academy

A course in argumentation

Fifteen levels of lessons and exercises, with a searchable reference of fallacies and propaganda techniques.

Notebook

Armory and Shield

Keep your own store of arguments and counters, and a record of claims you consider propaganda with what counters them. Kept in your account; a connected AI assistant can add to them and link you straight to what it saved.

How it works

Every finding quotes the exact words and says which check raised it.

  1. You supply the text.Paste it, open a link, upload a file or pick a draft from your history.
  2. Several checks read it.Wording rules, a trained classifier and a language model, each on its own.
  3. You decide.Accept, dismiss or rewrite. Nothing is scored as true or false, and nothing is changed without you.

Plans

Start free. Upgrade when you need more analyses, batch runs, feed watches, API access, or the argument arena.

Free

€0

  • 50 analyses a month
  • Editor, citations and the first three Academy lessons
  • History synced to your account
  • No batch runs, feed watches or API keys
Start free
Max

€80 / month

or €800 a year (two months free)

  • 3,000 analyses a month
  • Everything in Pro
  • Argument arena: test argument A against B in a simulated audience
Choose Max

Teams: €15 per seat a month, with shared drafts, approval lanes and posting to company accounts. Enterprise: priced for your organisation. Write to us.

Prices include VAT. Payments are handled by Stripe, and you can cancel any time from your account; a paid plan then stays active until the end of the period you paid for. Terms

What it does not do

We would rather you trust it a little less than a lot. So:

  • Findings are candidates for review, not verdicts. Probabilities are model estimates, and some techniques are caught far more reliably than others.
  • It does not fact-check claims, rate truthfulness or judge intent.
  • The logic check tests the argument as it was formalised. A bad translation of the text into logic can give a wrong answer, so read the premises it extracted.
  • Citations come only from papers you upload or import, and each one is checked against the passage it points to.
  • The app explains how far to trust a finding and what can fool it.

Use it from your own tools

Call the analysis from your own scripts through the API, or connect it to an AI assistant that supports remote connectors so you can analyse and cite without leaving the conversation.

Learn to spot them

Plain-language guides to the fallacies and propaganda techniques the app looks for, each with examples, how to spot it and how to respond.

Questions

What does All-Seeing AI do?

It reads a text closely and marks where a reasoning fallacy or propaganda technique may be at work, quoting the exact words. It can also check whether a conclusion follows from its premises, review your own drafts, and find passages to cite in papers you upload.

Is it free?

Yes, the Free plan includes 50 analyses a month, the editor, citations, two transcript trials and the first three Academy lessons. Pro costs €9 a month or €90 a year and adds 1,000 analyses a month, the whole Academy and libraries, batch runs, feed watches and API keys.

Does it fact-check claims?

No. It examines how an argument is built, not whether its facts are true, and it does not judge anyone's intent. Findings are candidates for you to review, not verdicts.

How accurate is it?

Some techniques are caught far more reliably than others, so every finding is a candidate for you to review, not a verdict. The app explains how far to trust a finding and what can fool it.

What happens to the text I analyse?

Text is sent to the All-Seeing AI service only when you run an analysis. It is stored on the server only if you opt in to server history, and then for no more than 30 days.

About

All-Seeing AI began in 2023 as a project of researchers in computational social science and political communication. It has since narrowed to one question: how can software help people examine reasoning, theirs included, without pretending to settle it for them?

Questions, partnerships or bug reports: info@all-seeing.ai