AI & Model Disclosure
Last updated: August 7, 2026
Presidentializer uses machine learning models to generate its output. This page explains how those models work, what data they were trained on, and how output is marked as artificially generated.
You Are Interacting With an AI System
Every conversion you run is produced by software, not by a person. Nothing this site produces is a real statement by any human being, and no output should be read as a statement of fact about anyone.
Capitalization Model
The core feature, Trump-style capitalization, is powered by a Conditional Random Field (CRF) model trained on publicly available posts from Truth Social, sourced from the CNN Truth Social archive. The model predicts whether each word should be lowercase, Title Case, or ALL CAPS based on linguistic features including part of speech, named entity type, sentiment, and position in the post. It changes the case of your words; it does not write new ones.
The model is retrained periodically as new posts become available. It runs on a dedicated inference server and processes every converter request.
Nickname Generation (Credit-Gated Feature)
When the nickname toggle is enabled, known public figures are substituted using a curated static dictionary. This part is always free and does not involve any AI model. For people the dictionary does not recognise, the service uses the Anthropic Claude APIto analyse sentiment and, when it is negative, generate a nickname. Your input text is sent to Anthropic's API for this purpose and is not used to train Anthropic's models. Anthropic's Privacy Policy applies to that processing.
How the nickname is bounded. Adjectives come from two fixed vocabularies. A mild list (energy, manner, or a plainly subjective character judgment) is available for anyone. A sharper politicallist is reserved for people the model identifies as well known in public life. Anyone it cannot place gets the mild list, so an unfamiliar name still produces a nickname rather than nothing. Choosing the "Personal" topic caps every name in that request at the mild list, whoever they are.
Neither list contains anything implying a criminal, professional, medical, or sexual allegation, and the model is separately instructed never to produce a nickname for a minor or for someone described as the victim of a crime, accident, illness, or abuse. The vocabulary limit is enforced in our own code after the model responds, not only requested in the instructions we send it. A nickname whose adjective falls outside the permitted list is discarded and the name is left unchanged. These are safeguards, not guarantees; see Accuracy & Limitations below.
This step costs 1 credit per generation. New accounts receive 5 free credits; additional credits are purchased in one-time packs. See our Terms of Service for details. Capitalization (the core feature, above) is unaffected by your credit balance and is always free.
Marking of AI-Generated Output (EU AI Act Art. 50)
Output from this service is machine-generated and is marked as such:
- Converted text is labelled in the interface as AI-generated satire, next to where it appears.
- Our API responses carry machine-readable provenance fields (
ai_generated: true, along with the generating system and asynthetic-textmarker) so downstream systems can detect that the text is synthetic. - Pages that produce output declare it in page metadata for automated readers.
If you publish output from this tool, particularly text about public figures, elections, or other matters of public interest, you become the deployer of that content and you are responsible for disclosing that it was artificially generated, unless it has undergone human editorial review and someone holds editorial responsibility for it. Do not strip the labelling and present output as genuine; our Acceptable Use Policy prohibits it.
The capitalization feature alone changes only the case of text you wrote yourself. We take the view that this is an assistive editing function that does not substantially alter your input, but we mark it as machine-processed anyway.
Accuracy & Limitations
- Model outputs are probabilistic and may not match the style in every case
- The model reflects patterns in training data and may reinforce biases present in that corpus
- AI-generated nicknames may be inaccurate, unexpected, or offensive, and the safeguards above can fail
- Outputs are parody and opinion; they should not be interpreted as factual statements about any person
- The system cannot verify whether a person named in your text is genuinely a public figure. It relies on the model recognising them, and defaults to the mild vocabulary when it does not
- A mild nickname is still a nickname; the vocabulary limits how sharp output can be, not whether it is welcome
Human Oversight & Reporting
Nothing this tool produces is published by us: output is returned to you, and the share buttons and downloadable card hand it to you to post or not. We host none of it at a public address, and nothing you convert is stored. Whether it reaches anyone else is your decision. If a generated nickname or other output seems harmful, or concerns someone who should not have been targeted, report it to hello@trulyrandom.co and we will review the dictionary entry or the generation safeguards.
Training Data
The model was trained on publicly available social media posts. We do not claim copyright over third-party content used for training, and no personal data from those posts is stored in the model beyond the per-word capitalization statistics needed to predict letter case. The model itself is an original work trained by the Presidentializer project. Your converter input is never added to the training corpus.