What research is Wretify built on?
Wretify is a clean, minimalist writing app for nonfiction writers, built for focus — with an AI that has read everything you've written and refuses to write it for you.
The short answer
Wretify is a clean, minimalist writing app for nonfiction writers, built for focus — with an AI that has read everything you've written and refuses to write it for you. Its main design decisions each answer a research finding about what happens to writers, and their books, when AI does the writing or when attention is spread too thin. The AI refuses to write because of the research on cognitive debt. Citations come only from real records because language models invent references. Only three books are active at once because working memory is small. Each chapter keeps a note on where you stopped because unfinished tasks follow you into the next one.
Which design decisions come from which research?
| What Wretify does | What the research found | Read more |
|---|---|---|
| The AI refuses to write your chapters; it asks questions and brings sources instead. | People who wrote with ChatGPT could barely quote their own essays; better AI-assisted essays did not mean more learning; people do not feel they own AI-written text. (Kosmyna et al. 2025 (preprint) · Fan et al. 2025 · Draxler et al. 2024) | Why the AI refuses to write |
| Every citation comes from a real record in OpenAlex, never from the model’s memory. | In one test of 636 citations, 55% of GPT-3.5’s and 18% of GPT-4’s were fabricated; in a 2025 test, 20% of GPT-4o’s were. (Walters & Wilder 2023 · Linardon et al. 2025) | Can AI make up sources? |
| Citations live inside the sentence they support, and the reference list is built from them. | Researchers revising with AI found “citation positions often shift” across revisions. (Nogueira et al. 2025 (preprint)) | References and citations |
| The grammar pass makes the smallest fix, leaves your style alone, and names each mistake. | Even grammar-only AI edits can significantly change meaning; AI polishing narrows the variety of writing by 21–50%. (Abdulhai et al. 2026 (preprint) · Sourati et al. 2026) | Does AI change the meaning of what I write? |
| At most three books and three articles are active at once. | Working memory holds about three to four things, not the seven usually quoted. (Cowan 2010) | How do I finish a nonfiction book? |
| A bookmark with a note in each chapter, and the chapter opens at it. | A brief “ready-to-resume” plan reduces attention residue from an interrupted task. (Leroy & Glomb 2018) | How do I protect deep work? |
| No notifications, and one room open at a time. | Attention left on an unfinished task lowers performance on the next; distraction interferes more when working memory is busy. (Leroy 2009 · de Fockert et al. 2001) | How do I protect deep work? |
| The AI thinks with you on planning — questions, blind spots, logic — rather than planning for you. | Nonfiction and academic writers care most about owning the planning stage. (Reza et al. 2025) | Should I use AI to outline my book? |
How strong is this research?
It varies, and each page says how. Some findings come from large or randomised studies in peer-reviewed journals; some come from preprints that have not yet been peer-reviewed, which the pages name as preprints; one widely repeated finding did not replicate, and the page that cites it says so. Where a popular claim failed a check of its source — that focus builds myelin in minutes, that task-switching costs 40%, that the brain runs on 90-minute cycles — the pages say that too.
Is every feature in Wretify backed by research?
No, and this page does not pretend otherwise. Some decisions are design choices — how the page looks, where a button sits, keeping your chapters as plain Markdown files in your own Google Drive so they are yours whether or not Wretify exists. The table above lists only decisions that answer a specific finding, and every finding was checked in the original study.
Where can I read the research myself?
Every page below answers one question and lists its sources in full:
Sources
- Kosmyna et al. (2025), Your Brain on ChatGPT — preprint — doi.org/10.48550/arXiv.2506.08872
- Fan et al. (2025), British Journal of Educational Technology 56(2) — doi.org/10.1111/bjet.13544
- Draxler et al. (2024), The AI Ghostwriter Effect, ACM TOCHI 31(2) — doi.org/10.1145/3637875
- Walters & Wilder (2023), Scientific Reports 13 — doi.org/10.1038/s41598-023-41032-5
- Linardon et al. (2025), JMIR Mental Health 12 — doi.org/10.2196/80371
- Nogueira et al. (2025) — preprint — arxiv.org/abs/2512.11661
- Abdulhai et al. (2026) — preprint — arxiv.org/abs/2603.18161
- Sourati et al. (2026), Nature Human Behaviour — doi.org/10.1038/s41562-026-02550-0
- Cowan (2010), Current Directions in Psychological Science 19(1) — doi.org/10.1177/0963721409359277
- Leroy & Glomb (2018), Organization Science 29(3) — doi.org/10.1287/orsc.2017.1184
- Leroy (2009), Organizational Behavior and Human Decision Processes 109(2) — doi.org/10.1016/j.obhdp.2009.04.002
- de Fockert et al. (2001), Science 291 — doi.org/10.1126/science.1056496
- Reza et al. (2025), CSCW 2025 — doi.org/10.1145/3757566
Updated .