When AI writing works (and when it doesn't)

The honest pitch for any AI writing tool has to include where it does not belong. A tool that claims to be the right answer for every writing task is overselling. A tool that names its fit boundaries clearly is more useful than a tool that hedges, because the user can decide confidently whether to use it for a given piece of work.

The fit question for Onaji, and for voice-matched AI drafting tools generally, has a clear answer. The tools work for a specific category of writing and fail on a specific other category. Naming both categories explicitly is more honest than marketing around the failure modes.

合The cases where AI writing is the right tool.

The clearest fit is what this whole article series has been about: cadence-oriented professional commentary. Writing that needs to happen on a rhythm, in the writer's recognizable voice, about events and ideas in the writer's field, where the goal is steady public presence rather than a single career-defining piece.

LinkedIn posts are the canonical example. Newsletter issues, on a weekly or biweekly cadence, are another. Short blog posts responding to industry news. Twitter or Bluesky threads reacting to current events. Commentary in internal company all-hands or Slack. All of these share the same shape: regular production, voice-driven, with the value in the aggregate presence rather than any single piece.

For cadence-oriented writing, the writer's competing options are all worse than a voice-matched tool.

Writing from scratch produces voice fidelity but fails the cadence test; per-post cost is too high to sustain weekly production for a busy professional. Posting less than weekly loses the compounding effects the cadence is supposed to produce.

Writing with a generic AI tool produces cadence capacity but fails voice; drafts are either published as the tool's voice or rewritten at a cost that negates the cadence benefit.

Hiring a human ghostwriter produces voice approximation at a cost most professionals cannot justify (the $500–$5,000/month range), and introduces its own problems (the ghostwriter has to be kept current on the writer's thinking, needs lead time, and can inadvertently flatten the voice toward their own style).

A voice-matched tool solves the cadence-plus-voice problem at a price and friction level that the other options cannot match, specifically for the case where the writing's goal is sustained presence. This is where the tool's fit is strongest.

離The cases where it isn't.

There are at least four categories of writing where a voice-matched AI drafting tool is the wrong choice.

The first is high-stakes signed work. Keynote speeches, book introductions, senate testimony, wedding toasts. Writing where the specific performance is the work and the writer's craft decisions throughout are the value. These are not cadence events; they are single-instance events where the cost of getting the voice wrong, or even getting it slightly off, is high. A writer producing a keynote should be in the craft headspace the whole way through, generating deliberately, editing deliberately. An AI first draft is likely to shape the thinking in ways the writer cannot fully undo during editing.

The second is genuinely original craft writing. Poetry, literary fiction, essays that are themselves the art. The writer of an original personal essay is trying to discover what they think by writing, and the discovery is inseparable from the act of generation. Handing the generation step to a tool forecloses the discovery. A tool that produces a rough draft of a literary essay has, in a real sense, decided what the essay is about, and the writer who tries to edit toward their own voice will find they are editing toward a version of the essay the tool chose.

The third is writing that is the writer's primary professional craft. If the writer is a journalist, a novelist, a professional essayist, a speechwriter by trade, writing is the thing they are trying to get better at, and the practice of generating prose is the skill they need to keep sharpening. Outsourcing the generation step to a tool, even a voice-matched one, undermines the practice. Writers who write for a living usually gain little from voice-matched drafting tools, and can lose something important (the daily repetition that keeps their craft sharp).

The fourth is writing where the writer genuinely has no position yet. A professional sitting down to write a post on a topic they have not thought through will benefit from the blank-page struggle, because the struggle is where the thinking happens. A voice-matched draft on this topic would produce polished-sounding prose that gestures at a position the writer does not actually hold. Publishing that prose is worse than not posting. The writer should either do the thinking first and then use the tool, or skip the post.

境The line between them.

The line between fit and non-fit is surprisingly clean, once the categories are named.

AI drafting tools are fit for writing where: - The cadence matters more than any single piece (volume over virality). - The writer has a position and knows what they want to say (thinking is done; only generation is missing). - The writer is not pursuing writing as a craft (writing is a tool, not the endeavor). - The voice is supposed to be recognizable over time (so voice-matching is the feature).

AI drafting tools are unfit for writing where: - The specific piece matters more than cumulative presence (single career-defining speech). - The writer does not yet have a position (thinking is still happening). - The writer is practicing writing as craft (generation is part of the discipline). - The voice is supposed to be deliberately performed or varied (essays that explore register; fiction that shifts perspective).

Most professionals' LinkedIn writing falls squarely into the fit category. Most professionals' high-stakes career speeches fall into the unfit category. The categories do not overlap much, which is why the choice is clean once named.

選How to tell which side you're on.

The simplest diagnostic, for a writer considering whether to use a voice-matched tool for a specific piece: would you be satisfied if the finished piece reads as recognizably you, but not as your best single performance?

If yes, the tool is the right fit. Cadence-oriented writing is won by consistency and voice presence, not by any individual post's craft ceiling. "Recognizably you, not your peak" is exactly what a voice-matched drafting tool produces.

If no, the tool is the wrong fit. The piece you are writing is trying to be a peak performance, and peak performances benefit from the writer being fully in the craft headspace during generation. Use the tool for the fifty other posts this year that want to be sustained presence; skip it for the one that wants to be the keynote.

Applying this diagnostic honestly often surprises writers. Many assume they want every piece to be a peak, and discover that most of their LinkedIn activity is actually cadence work, and is better served by a tool that enables sustained presence than by a tool that demands peak effort.

The takeaway.

Every writing tool has fit boundaries. Naming the boundaries is not a weakness; it is the condition for a writer to use the tool well. Voice-matched AI drafting tools (Onaji included) have a specific fit zone: cadence-oriented professional commentary where the writer has a position, is not practicing writing as a craft, and wants their voice to be recognizable over time.

For that zone, the tools are the best available option. Outside that zone, they should not be used. The distinction matters more than most tool marketing lets on.

The writer who picks the right tool for each piece of writing ships more, and the shipped work carries their voice reliably. The writer who uses the wrong tool for a piece (either forcing peak performance through a cadence tool, or grinding cadence work by hand) pays a cost that was avoidable.

Clarity about the fit zone is what makes the choice easy.