Introducing personal models: AI models trained for one person, on their own writing.
Frontier models are trained once, for everyone. That's why their text sounds like the model. A personal model is trained on how you write. We benchmarked ours against Claude Opus 5.5, GPT-6 Astra and Gemini 4 Argon on 50 writers, each given 50 texts from that person to learn from.
A blind judge cannot differentiate our models' writing from the person's own, and the AI detector reads everything as human.
It writes like the person
We followed the protocol of PersonalBench, a benchmark for writing as a specific person,1 on a licensed set of 50 writers. Each system wrote new texts as each writer from a one-line summary of something the writer had actually written. A blind judge read three of the writer's real texts and the new one, and decided whether the same person wrote it.
A model designed for authorship verification agrees. This is a model built only for distinguishing different writers. Our models' writing is closest to the person's own, and then the frontier models are all well behind.3
It reads as human
Most people try to detect AI text with a detector. Pangram is the most accurate detector we're aware of. We gave it one text per writer from each system. It flagged almost every text from the frontier models, even when imitating a real human from 50 examples. It didn't flag any of ours.
Why prompting falls short
The common way to do this is to prompt with examples of your writing. In PersonalBench, every prompting method scored below a random stranger's writing. We gave frontier models 3 of a person's texts, then 50, nearly seventeen times as many, and they moved only partway toward the person's style. Our personal model, trained on 25 texts, scored higher than any of them.1
Writing well is as important as sounding right. In our system, each personal model works alongside our models for substance and fact-checking, and a blind rater gave that writing a score of 3.70/5, higher than the participants' own writing at 3.11.5
Results
| Evaluation | Personal model | Opus 5.5 | GPT-6 Astra | Gemini 4 Argon | The person |
|---|---|---|---|---|---|
| Blind judge: same writer | 82% | 60% | 60% | 50% | 82% |
| Authorship model (LUAR) | 0.79 | 0.72 | 0.72 | 0.70 | 0.80 |
| Pangram: human | 100% | 4% | 10% | – | 100% |
| Blind judge, personal model from 25 texts | 78% | 60% | 60% | 50% | 82% |
Frontier models received 50 of the writer's texts. Another person scores 15% on the judge and 0.62 on the authorship model. The margin our personal models have over every frontier model is statistically significant at 95% on both likeness metrics.
- PersonalBench (Sawant, August 2026) measured how well models write as a specific person using LUAR, a published authorship-verification model. Few-shot prompting, profile extraction and contrastive prompting all scored below a random other person's writing.
- 50 writers from a licensed public corpus, 250 new texts per system, written from one-line summaries of held-out texts no system saw. The judge is Claude Sonnet 5.5, shown three of the writer's texts and one candidate, never which system wrote it. Raw rates: the person 82.0%, personal models 82.4% (78.3% from 25 texts), Opus 5.5 59.9% (45.3% from 3 texts), GPT-6 Astra 59.6% (52.4%), Gemini 4 Argon 49.7% (39.2%), a different person 14.7%.
- LUAR similarity to the writer's other texts: the person 0.798, personal models 0.788, Opus 5.5 and GPT-6 Astra 0.718, Gemini 4 Argon 0.698, a different person 0.622.
- Pangram, one text per writer per system. Two of GPT-6 Astra's texts were under the detector's minimum length. Gemini 4 Argon with 50 texts was not scored.
- A blind rater, Claude Sonnet 5.5, scored each text 1 to 5 as a contribution to its discussion; the person's real text was scored the same way. With this system the judge still attributed 75.8% of texts to the writer.