Artificial Idea | AI careers · practical prompts · no hype Thursday, November 6, 2025 · Issue #28 · Prompt Tutorial
The voice problem
How to use AI as a writing coach (not a ghostwriter) ,the distinction that matters
The fastest way to become invisible as a professional writer is to let AI write for you. The fastest way to become more visible is to use AI to write better than you currently do alone.
Issue #27 made the case that proactive information seeking, consistently translated into action, is the most important professional habit in the current AI transition. Writing is one of the primary mechanisms through which that translated action becomes professionally visible. The analyst who synthesises information well but communicates it poorly is less visible than their capability warrants. The professional who develops a distinctive written voice, grounded in genuine expertise and communicated with clarity and precision, compounds their professional visibility in ways that have always mattered and that the current moment is making more consequential.
The challenge is that AI has made professional writing simultaneously easier and more homogeneous. The tools that help people write more quickly are also producing a convergence of style, structure, and vocabulary that is making professionally written content harder to differentiate. A reader who consumes significant volumes of professional content in 2025 has developed, often without articulating it, a sensitivity to the specific patterns of AI-generated prose: the even sentence length, the hedged conclusions, the structural predictability, the absence of a genuine point of view. That sensitivity is a filter, and content that triggers it is discounted before its substance is evaluated.
The professionals whose writing is cutting through are not writing without AI. They are using AI differently from those whose writing blends into the background. The distinction is the one this issue addresses.
The ghostwriter problem
When a professional uses AI as a ghostwriter, directing it to produce content on a topic and publishing the output with minimal intervention, three things happen that compound over time into a significant professional liability.
The first is voice erosion. Written voice is developed through the practice of writing, specifically through the process of finding the words that accurately express a specific thought in a way that is distinctly yours. When AI finds the words instead, the practice does not occur and the voice does not develop. Professionals who write primarily through AI assistance over an extended period consistently show, in the research on this, a declining ability to write without it, and a declining distinctiveness in the writing they do produce.
A 2025 study by Stanford's Human-Computer Interaction Group tracked 340 professionals who used AI writing assistance over eighteen months, measuring the distinctiveness of their writing at six-month intervals using stylometric analysis. Professionals in the high-AI-assistance group showed a 34% decline in stylometric distinctiveness over the study period. Professionals in the low-AI-assistance group showed a 12% increase. The tools that were supposed to make their writing better were making it less recognisably theirs.
The second consequence is expertise concealment. Distinctive professional writing is not primarily distinguished by its style. It is distinguished by the specific, earned knowledge it communicates in ways that only someone who has thought seriously about a subject can communicate. AI-generated content on a topic reflects the average of what has been written about that topic, not the specific insights of a professional who has spent years working in it. The content sounds competent and reads smoothly, but the reader who knows the subject will notice the absence of the specific, idiosyncratic, experience-grounded observations that signal genuine expertise. And the readers who matter most are usually the ones who know the subject.
The third consequence is credibility fragility. As AI detection tools improve and as the stylistic signatures of AI-generated content become more widely recognised, professionals whose public writing is primarily AI-generated are accumulating a credibility risk. The risk has not fully materialised yet. It is materialising, and the professionals who have built their public writing presence on AI-generated content are building on a foundation that is less stable than it currently appears.
The writing coach model
The alternative is not to avoid AI in the writing process. It is to use it at different points in the process, for different purposes, in ways that develop rather than substitute for the professional's own capability.
The writing coach model treats AI as the interlocutor that helps a professional find and strengthen their own voice rather than as the producer of content that the professional then publishes. It is more effortful than ghostwriting and produces something more durable: a professional writer who is getting better at writing, rather than one who is getting better at prompting a tool that writes for them.
The model has four components. Thinking before writing, where AI is used to pressure-test and develop the ideas before any drafting begins. Structural scaffolding, where AI is used to build an architecture for the piece rather than to write it. Voice editing, where AI is used to evaluate the draft the professional has written and identify where it is weakest. And specificity sharpening, where AI is used to identify where the writing is generic and to help the professional find the specific observation or example that would make it genuinely distinctive.
Each component is distinct and each requires a different prompt.
Prompt 1: The idea pressure-tester
The problem it solves: ensuring that the idea or argument at the centre of a piece of professional writing is worth making before investing in the writing, and that it is genuinely yours rather than a restatement of the obvious.
You are a rigorous intellectual editor
whose job is to evaluate whether an idea
is worth writing about and whether
the proposed treatment of it is distinctive.
The idea I want to write about:
[describe the central argument or insight
you want to communicate, as specifically
as you can]
My professional context and the source
of this idea: [explain where this idea
comes from in your experience,
what specific observation or experience
generated it]
My intended audience: [who will read this
and what they already know about the topic]
Please:
1. Assess whether this idea is genuinely
distinctive or whether it restates
something that is already widely said
about this topic, with specific reference
to what is already widely said
2. Identify the most interesting version
of this idea, the version that would
make a knowledgeable reader think
rather than nod along
3. Identify the specific detail, example,
or observation from my professional
context that would make this argument
something only I could make
4. Tell me the strongest objection
to the central argument and whether
the idea survives it
Do not help me write the piece yet.
Help me decide whether the idea is
ready to be written.
The instruction not to help with the writing yet is the constraint that makes this prompt serve the writing coach model rather than the ghostwriter model. The temptation when using AI in any writing process is to move to production too quickly, before the thinking is done. This prompt is explicitly designed to resist that temptation.
Prompt 2: The structural scaffold builder
The problem it solves: developing a structure for a piece of writing that serves the argument rather than following a generic template, which is what most AI-generated structures do.
You are a structural editor helping me
build an architecture for a piece of
professional writing that serves
its specific argument.
The central argument I am making:
[one clear sentence stating your argument]
The specific evidence or reasoning
I have available to support it:
[list the specific points, examples,
data, or observations you plan to use,
in whatever order they occur to you]
The audience and their prior knowledge:
[who will read this and where they start]
The length I am working to:
[word count or reading time target]
Please:
1. Propose three structurally distinct
ways to organise this material,
each serving the central argument
differently, with the logic of
each structure explained
2. Identify which structure is most
likely to be persuasive to this
specific audience, with your reasoning
3. For the recommended structure,
build a section-by-section outline
where each section is described
by what it needs to accomplish
rather than what it will contain
4. Identify the transition between
sections that is most likely to
lose the reader if it is not
handled carefully, and why
Do not write any prose yet.
Build only the architecture.
The instruction to describe each section by what it needs to accomplish rather than what it will contain is the constraint that produces structures that serve arguments rather than structures that organise information. Content-based outlines tell you what you will say in each section. Function-based outlines tell you what each section needs to do to the reader's understanding. The latter produces better writing because it keeps the writer focused on effect rather than on coverage.
Prompt 3: The voice evaluator
The problem it solves: identifying where a draft has drifted into generic professional language and away from the writer's distinctive voice and specific expertise, which is the most common single failure in professional writing.
You are a voice and style editor reading
a draft of professional writing with
the specific goal of identifying where
it sounds like anyone and where it
sounds like the specific person who wrote it.
Here is the draft: [paste your draft]
Here is what I know about the writer's
professional context and the specific
expertise that should be coming through:
[describe the writer's background,
their specific experience with this topic,
and the observations that should be
distinctive to them]
Please:
1. Identify the three sentences or
passages in this draft that are
most distinctively written,
where the writer's specific knowledge
or perspective is most visible,
and explain what makes them work
2. Identify the three sentences or
passages that are most generic,
where the writing could have been
produced by anyone with a passing
familiarity with the topic
3. For each generic passage, identify
what specific observation, example,
or detail from the writer's context
could replace the generic content
and make it genuinely distinctive
4. Assess whether the central argument
is visible throughout the draft
or whether it gets lost in the
middle sections, which is where
most professional writing loses its thread
Do not rewrite anything.
Identify and diagnose only.
The rewriting is the writer's work.
The instruction not to rewrite anything is the most important constraint in this prompt and the one most likely to be violated if not stated explicitly. AI will default to showing rather than telling when asked to evaluate writing, producing a rewritten version of the draft that the writer then adopts, which is ghostwriting with an extra step rather than writing coaching. The constraint keeps the tool in its appropriate role.
Prompt 4: The specificity sharpener
The problem it solves: replacing the generic examples, vague claims, and unearned assertions that weaken professional writing with specific, evidenced, distinctive observations that give the writing its credibility and memorability.
You are helping me sharpen the specificity
of a piece of professional writing by
identifying where it makes claims
that need specific support.
Here is my draft: [paste draft]
My professional background and the
specific knowledge I have available
that is not yet in the draft:
[describe relevant experience,
observations, data, or examples
you have access to but have not used]
Please:
1. Identify every claim in this draft
that is stated as if it is obvious
or well-established but that would
benefit from a specific example,
data point, or evidenced observation
2. For each such claim, identify whether
the support should come from my
professional experience, from
cited research, or from a concrete
example, and why
3. Identify the single claim in this
draft that is doing the most work
in the argument and is currently
the least well-supported,
and what it would take to support it adequately
4. Identify any passage where I am
using professional vocabulary
to sound authoritative without
the specific knowledge to back it up,
which is the pattern readers who
know the subject will notice first
Be direct about where the writing
is claiming more than it is demonstrating.
That gap is where credibility is lost.
Prompt 5: The final read
The problem it solves: getting an honest assessment of the finished draft against the standard it is meant to meet, before it is published or sent.
You are the target reader for this piece
of professional writing. You are
[describe the reader: their role,
their knowledge of the subject,
what they are trying to get from reading this,
and how much of their attention
this piece is competing for].
Here is the piece: [paste final draft]
Please give me your honest reaction as this reader:
1. Did the piece earn your attention
from the first paragraph,
and if not, where did you consider stopping
2. What is the single most valuable
thing you took from reading it
3. What is the claim or passage
you found least convincing
and why
4. Did the piece change how you think
about anything, however slightly,
or did it confirm what you already knew
5. Would you share this with a colleague
and if so, what would you say about it
Be honest rather than encouraging.
A piece that gets encouraging feedback
it does not deserve gets published
when it should be revised.
The final question, whether the reader would share it and what they would say, is the one that most reliably distinguishes content worth publishing from content that clears a minimum threshold. Professional writing that spreads is writing that gives the sharer something worth saying when they share it. If the reader cannot answer that question with a specific, substantive sentence, the piece has not yet earned its distribution.
The practice
These five prompts used in sequence produce a writing process that is slower than AI ghostwriting and faster than unassisted drafting. The output is writing that is distinctively yours, because AI was used to strengthen the thinking and sharpen the execution rather than to produce the words.
The compounding effect of this practice over time is the one the Stanford study's low-AI-assistance group demonstrated: stylometric distinctiveness increasing as the professional develops rather than declining. The professionals building that distinctiveness now are building a professional asset that will be more valuable as AI-generated content becomes more prevalent and more homogeneous.
Distinctive writing is not a nice-to-have in a professional context where AI is generating unlimited quantities of competent, undistinctive content. It is, increasingly, one of the most visible and durable signals of genuine expertise available to a knowledge worker.
The words have to be yours. These prompts help you find them.
Monday we are examining the data on something that has been a background theme in this newsletter since the beginning: the specific career stage at which investment in AI capability produces the highest return, and why the answer is different for early-career, mid-career, and senior professionals in ways that most generic AI career advice fails to distinguish.
The generic advice is not wrong. It is calibrated to nobody in particular, which is a specific kind of wrong that this newsletter has been trying to correct since Issue #1.
— The Artificial Idea team
