How Bloomberry Learns Your Writing Voice
Most AI tools apply the same logic on day one and day one hundred. Bloomberry doesn't. Here's the six-mechanism system behind voice that actually improves.
By Sadok Hasan
How Bloomberry Learns Your Writing Voice
Most AI writing tools make a quiet assumption: that the model knows how to write, and your job is to give it a topic. The training step, if it exists at all, is a one-time setup — a few example posts fed into a style prompt, then forgotten.
The problem is that voice is not a style setting. It is a pattern that emerges from hundreds of specific decisions — where you put the hook, how long your sentences run before you break them, whether you end on a question or a statement, how often you use data versus observation. No style prompt captures that. And a one-time setup does not track how it changes over time.
Bloomberry takes a different approach. The AI LinkedIn post generator and the broader writing system are built around six specific mechanisms that determine how well the model learns your voice — and how much that learning improves as you use the product. This post maps those six mechanisms. Each one is covered in full in a dedicated post in this series. Start here, then read whichever mechanism most matches the question you have.
The Six Mechanisms
1. Why sample volume changes how the model uses your voice
A system trained on three posts and a system trained on fifty posts are not doing the same thing. With very few examples, the model has to lean on the examples directly — it mirrors what it sees, because there is not enough data to identify patterns. With more examples, it can detect patterns: how you open, how you close, what percentage of your posts use data as a hook. The relationship between volume and model confidence is a spectrum, and a well-designed system behaves differently at each point on it.
Read: How AI uses more samples to write like you →
2. Why your LinkedIn voice and your X voice are stored separately
The same person who writes three-paragraph reflective LinkedIn posts about hard operational decisions also writes twelve-word X takes about the same subjects. If you train a voice AI on both, it averages them — and produces content that is too long for X and too thin for LinkedIn. Platform-specific voice profiles are not a preference; they are the correct design. Bloomberry keeps LinkedIn and X separate, and consults only the relevant profile when generating for each platform.
Read: Why your LinkedIn voice and X voice are different →
3. Why hashtag position matters more than hashtag usage
Every AI tool has a "uses hashtags" boolean. Two people both use hashtags. One puts three at the end as a block; the other weaves six inline throughout the post. A binary cannot distinguish them, so the AI applies its own default — and gets it wrong for both. Voice fidelity at the level of hashtag behavior requires measuring three things: count, position, and consistency. Bloomberry measures all three from your actual writing history and applies your specific pattern.
Read: Why hashtag position matters more than hashtag usage →
4. Why copying and scheduling a post is a learning signal
When you copy a generated post to your clipboard, that is high-confidence approval. When you schedule it, that is equally strong. When you edit a paragraph before posting, that edit — the delta between what the AI wrote and what you actually published — is the most precise voice signal available: a direct measurement of where the model's defaults diverge from your actual preferences. Most tools discard all of this. Bloomberry captures it passively, without requiring any extra action in your workflow.
Read: Every post you publish is a training signal →
5. Why your voice profile recalibrates automatically
Initial calibration captures your voice at a moment in time. Writing evolves — new formats, new audiences, a different relationship to authority or brevity. A static profile diverges from your actual current voice over time, and you would never notice until the output starts feeling subtly off. Bloomberry recalibrates automatically as new samples accumulate, triggered by the data rather than by a user action. The profile tracks who you are writing like now, not six months ago.
Read: Why your AI voice profile updates automatically →
6. Why Bloomberry stopped trying to classify what you meant
When a user pastes something into an AI writing tool, the system has to decide: repurpose this, generate something new, or something else? Most tools build a classifier for this — a heuristic layer that reads the input and routes it. The classifiers fail on exactly the cases that matter: the ambiguous ones. Long articles trigger action-word detection. Bare URLs get misclassified. The classifier is doing work the base model handles better. We deleted ours. Here is what we learned from doing it.
Read: Why we stopped trying to guess what you meant →
Frequently Asked Questions
How does Bloomberry learn my writing style?
Bloomberry analyzes your actual writing samples — LinkedIn posts, X threads, or any content you've published — and builds a voice profile from the structural patterns, vocabulary preferences, sentence rhythms, and formatting habits it finds there. The more samples it has, the more precisely it can distinguish your voice from the model's defaults.
How many posts does Bloomberry need to learn my voice?
A handful of samples is enough to start. The system behaves differently depending on how many samples it has — leaning directly on your examples when the corpus is thin, and detecting statistical patterns when the corpus is richer. There is no minimum required before it's useful; there's just a ceiling on how confident it can be with very few examples.
Does Bloomberry use a different voice profile for LinkedIn vs X?
Yes. LinkedIn and X writing are structurally different — different hook length, different argument depth, different relationship to irony and brevity. Bloomberry maintains separate profiles per platform so that your LinkedIn voice stays LinkedIn and your X voice stays X, rather than averaging them into something that fits neither.
How often does my voice profile update?
Automatically, as you use the product. When you accumulate enough new samples — through publishing, editing, copying, or scheduling — the system recalibrates in the background without requiring any action from you. The profile tracks your current writing, not the version of you from six months ago.
What signals does Bloomberry use to improve my voice profile?
Three types: approval signals (copying or scheduling a post signals that you liked the output), correction signals (editing AI output before publishing tells the system exactly where its defaults diverge from your voice), and accumulation signals (each new post adds to the corpus the system learns from). You don't have to do anything differently — the learning happens through your normal publishing workflow.
Related reading: AI that writes like you — what it actually takes | How AI voice matching works | The complete guide to teaching AI your writing voice
Bloomberry is built on the premise that voice fidelity is an engineering problem, not a prompt problem — and that solving it requires more than a style setting.
Ready to write sharper?
Bloomberry turns your ideas into publish-ready thought leadership.
Try Bloomberry free