AI & your work
Last updated: 28 August 2026
The short version
- You think first. The AI arrives only as far up the dial as you turn it.
- In deep mode no model is asked to write anything — the AI stays out of your draft entirely.
- When you ask for help, we send the passage or draft section in question, not your whole library.
- Outputs are suggestions, and they can be wrong. Check anything that matters.
- The AI never rewrites unless you ask, and it keeps your voice when it does.
The principle
The AI in AIReader.xyz is a companion that is present without being in the way. You always think first. The dial turns the AI up only as far as you need, and never past the point where it would do your thinking for you.
- You think first. Every surface asks for your own words before the AI's. Silence is a feature, not a gap.
- Turn the dial up as needed. Higher settings give sharper insight, never more authorship.
- Authorship stays with you. The AI comments, questions, restructures, and points; you decide and you write.
- Every intervention is grounded in craft. Advice cites a rule rather than taste. If no rule applies, the AI says so instead of inventing one.
- Brief, with room left over. Short notes in the margin beat essays. What the AI says should leave you with something to do.
The three positions
| Mode | Presence | What the AI does | What it never does |
|---|---|---|---|
| Deep | Not there | No model is called. The interface offers a prompt in your own words — state your claim, turn this passage into a question. | Generate text, summarise, or answer. |
| Dialogue | Beside you | Socratic tutor. Asks one good question, offers a nudge, reflects your words back. Suggestions appear inline, marked as suggestions. | Answer its own questions, hand over conclusions, or rewrite your passages. |
| Diabolical | Across the table | Expert critic. Names the exact weakness, the rule it breaks, and one way to fix it. Rewrites a sentence or passage when you ask, preserving voice and meaning. | Write the piece, replace your argument, or flatter you. |
What we send to AI providers
When you ask for assistance in dialogue or diabolical mode, we send the text the request is about — the passage you selected, the section of the draft you're working on, or the question you typed — together with the instructions for that mode. That's it.
- We don't send your whole library or your other documents.
- We don't send your email address, your name, or your account details.
- Requests go to OpenAI, and to Anthropic as a fallback, under their API terms. Under those terms your content is not used to train their models.
- We don't train any model on your work.
There is one thing we send that you don't explicitly ask for. So that your library is searchable and related material can be linked, saving or importing sends a short excerpt — the title plus roughly the first 400 characters — to OpenAI's embeddings API, and stores the resulting vector in our database. That indexing runs at every dial position, deep included. It never produces text and never touches your draft; it only builds the index.
The Privacy Policy covers the rest: what we store, where it lives, and how to get it back or delete it.
Outputs are suggestions
Language models produce plausible text, which is not the same as correct text. Quotes can be misattributed, summaries can flatten an argument, critiques can be confidently wrong about what a passage says. Treat every output as a suggestion to weigh against the source, not as a finding. What you publish or submit is yours, and you're responsible for it — see the Terms of Service.
It doesn't rewrite unless you ask
In dialogue mode the AI never rewrites: it questions, and any wording it offers is marked as a suggestion for you to take or ignore. In diabolical mode it will rewrite a sentence or passage, but only the one you asked it to, and it works to keep your voice, your vocabulary, and your meaning. There is no ghost text — nothing is written into your document without an explicit request from you.
How to keep the AI out of your writing
Set the dial to deep. In deep mode no model is asked to generate, suggest, summarise, or rewrite anything, and the interface falls back to prompts in your own words. It is the default position, and you can work in it indefinitely. Everything in AIReader.xyz — importing, annotating, drafting, revising, exporting — works with the AI switched off.
Deep mode governs AI assistance, not the search index: as described above, saving still sends a short excerpt to the embeddings provider. There is currently no setting that turns that off — if it matters to you, tell us at [email protected] and we'll say where it stands.
Questions
If something the AI did doesn't match what's described here, we want to know: [email protected]. The same principles apply on AIWriter.xyz.