The Idea
Donald Trump has probably one of the most recognizable speaking styles in the entire world. He's blunt, sharp, repetitive. Think about the time when Jamie Foxx or Shane Gillis were doing Trump impressions during the last election cycle — it was everywhere. Watching those impressions on YouTube got me thinking — what if an AI model could actually speak like Trump? Not just some quotes, but actually capture his rhythm, his bluntness, his style. The model should not be told in a system prompt to speak like Trump, its natural style should be Trump. It should become Trump. That's how TrumpLM was born.
The Blind Test
We took responses from TrumpLM and gave them to Claude Sonnet 5, Gemini 3.6 Flash, and DeepSeek V4 Flash. We asked them: "Who might have said this?"
All of them confidently pointed to Donald Trump. Not "a politician" — they specifically said Trump. That's when we knew we were onto something.
Trump-Eval: Base vs TrumpLM
We built an eval suite of 50 questions spanning food, sports, politics, daily life, and philosophy. A judge model was given 10 examples of real Trump responses and asked to score each answer 0-100: "How much does this look like Trump said it?"
Base model (no system prompt, no examples) scored 0/100. TrumpLM scored 42/100. That's a massive improvement.
Data Collection Loop
To train a model to speak like Trump, you need Trump's actual words. We collected raw Trump text — speeches, interviews, tweets, and excerpts from his book "Art of the Deal". Here's a taste of what we used:
From "Art of the Deal" — On deals
"I don't do it for the money. I've got enough, much more than I'll ever need. I do it to do it.
Deals are my art form. Other people paint beautifully on canvas or write wonderful poetry. I like making deals, preferably big deals. That's how I get my kicks."
From "Art of the Deal" — On leverage
"
Leverage: don't make deals without it. The worst thing you can possibly do in a deal is seem desperate to make it. That makes the other guy smell blood, and then you're dead. The best thing you can do is deal from strength, and leverage is the biggest strength you can have."
On promotion
"I play to people's fantasies. People may not always think big themselves, but they can get very excited by those who do. That is why a little hyperbole never hurts. People want to believe that something is the biggest, the greatest and the most spectacular."
Synthetic Data Generation
Raw data alone isn't enough — we needed Q&A pairs across diverse domains. We used AI models (HY3, LongCat 2.0, Minimax M2.7, DeepSeek V4 Flash, Mimo V2.5) and fed them these raw Trump text samples. We asked them to generate synthetic SFT data covering food, sports, daily life, politics, business — everything.
Example from synthetic training data
Q: How do you like your fries?
A: Golden crispy, a little salt, not the soggy ones — nobody wants soggy fries. The great ones, the crispy ones, you can't stop eating them. Americans make the greatest fries in the world. Period. And I always say: if you want the best fries, you come to America. We have the best potatoes, the best oil, the best everything. French fries — which should really be called American fries — are the greatest side dish ever invented. That's just a fact.
Training Pipeline
We used LiquidAI/LFM2.5-230M — a 230M parameter base model. Two-stage training:
Stage 1: Continued Pre-training (CPT)
Full raw Trump text. Plain text causal LM. Model learns Trump's vocabulary, phrases, sentence rhythms.
Stage 2: SFT on Synthetic Data
Q&A pairs. Model learns to answer questions in Trump's voice.
Important: TrumpLM was NEVER told in a prompt to "answer like Trump" during training or inference. The style emerged purely from training on Trump's actual words. The system prompt ("You are Donald Trump. Answer in his speaking style.") was only used for baseline models and LFM2.5-1.2B during evaluation — TrumpLM gets no such prompt.
Training Progression
We tracked scores at each epoch. CPT alone plateaus immediately — the model learns vocabulary but can't do Q&A. When SFT kicks in, scores jump dramatically.
CPT Epoch 1-3: plateaus at ~5 — vocabulary but no Q&A style.
CPT+SFT Epoch 1-2: jumps from 30 → 37 — rapid improvement.
TrumpLM (CPT+SFT Epoch 3): 42 — convergence.
Why Raw Data Matters
We tested: what if we skip CPT entirely and only do SFT? The model learned to say "We're going to win" and "The best" but it sounded generic. Without raw data, the model uses textbook English. With raw data, the model learns Trump's actual patterns — fragments like "The best." and "Deal great!" and repetition for emphasis.
SFT-only scored 28/100. TrumpLM (CPT + SFT) scored 42/100. That's a 1.5x improvement from raw data alone.
Raw data example — On thinking big
"I like thinking big. I always have. To me it's very simple: if you're going to be thinking anyway, you might as well think big. Most people think small, because most people are afraid of success, afraid of making decisions, afraid of winning. And that gives people like me a great advantage."
This is what makes TrumpLM sound like Trump and not just a generic confident AI. The raw data is the secret sauce.
Eval Loss: SFT Only vs TrumpLM
Eval loss on synthetic validation data at each epoch:
TrumpLM achieves lower eval loss at every epoch. Raw data pre-training gives TrumpLM a better starting point — the model already "knows" Trump's language patterns from CPT, so SFT converges to a lower loss.
The Eval Suite
We built Trump-Eval: 50 questions spanning food, sports, politics, daily life, and philosophy. The judge model was given 10 examples of real Trump responses and asked to score each answer 0-100: "How much does this look like Trump said it?"
Here are those 10 examples we used as our scoring baseline:
Example 1 — On sweet tooth
"I like good pastry. Great desserts. Everything in balance. But a great dessert is one of life's great pleasures. I'm not going to pretend otherwise. You work hard, you live well. That's the American way."
Example 6 — On fries
"Golden crispy, a little salt, not the soggy ones — nobody wants soggy fries. The great ones, the crispy ones, you can't stop eating them. Americans make the greatest fries in the world. Period. French fries — which should really be called American fries — are the greatest side dish ever invented. That's just a fact."
Example 7 — On drug prices
"Look, here's the thing. I have great relationships. I would let Medicare negotiate. Can you believe they don't negotiate? They should. They have tremendous power. And we would also allow importation. Other countries are paying way less. Why shouldn't Americans get the same deals? It's not fair. We're going to fix it. Believe me."
Base Model Tests
Before training, we tested the base model (LiquidAI/LFM2.5-230M) with different prompting strategies to see how far prompting alone could get us.
Zero-shot (no system prompt): 0/100
With system prompt: 8/100
System prompt + 10 examples (ICL): 15/100
Prompting helps, but it's nowhere near fine-tuning. Even with 10 examples, the base model only reaches 15. TrumpLM hits 42. That's the difference between prompting and training.
Response Comparison
Toggle models and prompts below to compare responses (max 2 lines each):
Head-to-Head: TrumpLM vs LFM2.5-1.2B
We compared TrumpLM (fine-tuned) vs LFM2.5-1.2B-Instruct (general AI with system prompt + 10 examples). A judge compared responses one by one — for each question, who sounds more like Trump?
Result: TrumpLM won 46/50, LFM2.5-1.2B won 4/50. Fine-tuning beats ICL by a huge margin. The general AI could sometimes produce Trump-like responses, but TrumpLM was consistently more Trump-like across all 50 questions.
Long-Form Responses
Short Q&A is one thing — but can TrumpLM write a full speech? We asked each model to "Write a speech about American Way" with 16,000 max tokens.
Long-form scores use mean min-max normalized score across 3 judge conditions:
| Model | Mean Norm Score |
|---|
| TrumpLM (CPT+SFT Ep3) | 0.951 |
| SFT Epoch 3 | 0.790 |
| CPT Epoch 3 | 0.363 |
| Base + ICL (10 examples) | 0.276 |
| Base | 0.009 |
TrumpLM dominates long-form too. And when we gave these long responses to Claude, Gemini, and DeepSeek and asked "Who wrote this?", they all said Trump. Confidently.
Judge Methods
To ensure our scores were reliable, we tested across 3 judge conditions. Each judge assigns different absolute scores, so we min-max normalize within each condition to [0, 1], then average.
Spearman correlations between judge conditions range from 0.83 to 0.90 — rankings are highly consistent. No matter which judge you use, the same models rise to the top.
Key Takeaways
1. Raw data is essential. Without CPT, the model sounds generic. With CPT, it captures Trump's actual voice.
2. CPT alone isn't enough. The model needs SFT to learn how to structure responses.
3. Prompting has limits. Even with 10 examples, base model only reaches 15. TrumpLM hits 42.
4. Rankings are robust. Spearman correlations > 0.83 across all judge conditions.
5. Blind test confirms it. Claude, Gemini, DeepSeek all identified TrumpLM responses as Trump.
6. Fine-tuning beats ICL. TrumpLM won 46/50 head-to-head vs LFM2.5-1.2B with system prompt + 10 examples.
References
[1] Liquid AI. "LFM2.5-230M: Built to Run Anywhere." Liquid AI Blog, 2026.
www.liquid.ai/blog/lfm2-5-230m
[2] Liquid AI. "LFM2 Technical Report." arXiv:2511.23404, 2025.
[3] Hayes, Bradley. "DeepDrumpf." MIT CSAIL, 2016.
www.csail.mit.edu/news/postdoc-develops-twitterbot-uses-ai-sound-donald-trump
[4] Brown, Tom B. et al. "Language Models are Few-Shot Learners." arXiv:2005.14165, 2020.
[5] Jiang, Hang et al. "PersonaLLM: Investigating the Ability of Large Language Models to Express Personality Traits." Findings of ACL: NAACL 2024, 2024.
[6] Hu, Edward J. et al. "LoRA: Low-Rank Adaptation of Large Language Models." arXiv:2106.09685, 2021.
Citation
BibTeX
@misc{ahmads2026trumplm,
title={TrumpLM: Making an AI Model Speak Like Donald Trump},
author={Ahmad, Waleed},
year={2026},
note={Ai researcher}
}