A Wordprint, in full

This document has three readers. The owner will read it to check it sounds like them. Anyone writing on their behalf, a copywriter, an assistant, an agency, will use it as a brief. Any AI tool drafting for this business should be given section 11 only, nothing else.

1. The one paragraph

This is a voice built on evidence rather than assertion: numbers used as evidence run to 178.5 per 10,000 words and named people, places or things to 149.8, so claims arrive with a figure or a name attached rather than a mood. The sentences are short by default, median length 11 words, but the range is wide, from 2 words to 50, and the habit that recurs is a long analytical paragraph followed by a short blunt one dropped after it: "Being busy is not a business model. It's a symptom." Contractions run high, 226.6 per 10,000 words, and questions land directly on the reader at 60.1 per 10,000 words, so the prose reads as spoken rather than composed. There are no em dashes, no emoji, and exclamation marks appear only 0.9 times per 10,000 words. Colons (134.1 per 10,000 words) do the work of setting up a diagnosis; semicolons (33.3 per 10,000 words) hold two halves of an argument that depend on each other, as in "you haven't given up 10% of anything; you've given up a third of your profit on that job." The most distinctive sentence in the corpus sits buried in the middle of a piece rather than at the top: "The root cause was not culture, tooling or headcount, although all three were blamed at various points," which is the whole method in one line, ruling things out before naming the actual cause.

2. What this business believes

Small businesses should compete on specificity, not volume, when AI floods the market with generic content. "When rivals turn up the volume of generic, the scarce commodity becomes the recognisably human and specific." For anyone writing on their behalf: never write value language that could belong to a competitor. Every claim needs a name, a number, or a detail that only this business could produce.

AI fails because businesses give it no context, not because the tool is weak. "He was half right. The output was useless. But not because the tool is weak; because it knew precisely nothing about his business." This means never dismissing AI as the culprit in a piece of writing. The blame, when there is any, goes to missing input, not the technology.

Revenue growth with flat profit means nobody has worked out which customers make money. "The actual diagnosis, almost every time, is that nobody has ever worked out which customers make money." Financial diagnosis should name the specific unworked-out variable rather than gesture at inefficiency in general terms.

Operational friction was always there; growth only exposes it, it doesn't create it. "The friction was always there. Growth just turned up the volume." Problems in client copy get framed as pre-existing and previously invisible, never as things growth caused.

Listening is the actual consulting work, not the courteous preamble to it. "Listening isn't the polite bit before the real work. Listening is the work." Any description of method must present listening itself as the deliverable, not as a stage that precedes the real one.

A 10% discount on a 30% margin costs a third of the profit, not a tenth. "If you run a 30% margin and discount by 10% to win a job, you haven't given up 10% of anything; you've given up a third of your profit on that job." Pricing arguments show the arithmetic in full. The point is never asserted without the sum that proves it.

A fractional executive is often better economics than a full-time hire. "Your first-year, all-in cost is comfortably north of £200,000, and you carry severance risk if it does not work out." Cost comparisons are stated in real pounds, never in vague relative terms like "a fraction of the cost."

Integrations fail because nobody explicitly decided how integrated the acquisition should be, not because the wrong point on the scale was chosen. "The failure mode is not picking the wrong point on the scale; it is never explicitly picking one, so every team negotiates its own answer, one meeting at a time, indefinitely." Diagnosis prose should identify the absence of a decision as the root cause, rather than blaming a bad decision.

Diagnosis has to come from the client's own people's evidence, never from a consultant's opinion imposed from outside. "People do not resist change. They resist verdicts they had no hand in." Recommendations are written as findings sourced from named people inside the business, never as verdicts delivered from outside it.

Even twenty years of his own experience carries invisible bias that only an outside eye can catch. "Twenty years of experience wasn't protection here; it was the mechanism." The writing self-implicates. He is shown catching his own defaults, not only critiquing the client's.

3. Who is being spoken to

Owner-managed SMEs, turnover £1m to £20m. "You've outgrown the way you work. Heroics are holding it together and the founder is the bottleneck." Direct, diagnostic, second person, and the sentences are short and blunt, naming the founder as the problem without softening it.

Not-yet-founded employees, still employed, planning to launch. "Still on the payslip, plotting the escape? Most people launch at the bottom of their market and spend years climbing." More informal register, led by a question that assumes the reader's private ambition rather than their current operational pain.

PE portfolio company operators, addressed in fund language. "Value creation, delivered. The plan is agreed; the bandwidth isn't. Operating partner-calibre execution, reported in the fund's language." Compressed, fragment-heavy, and the jargon of private equity, "value creation," "operating partner-calibre", replaces the plain English used elsewhere.

Scale-ups and VC-backed founders, post-raise. "Post-raise, everything must professionalise at once. Install the discipline investors expect without killing the pace." Framed around a dual constraint, discipline against pace, that doesn't appear for the SME audience.

Sceptical business owners weighing up AI hype. "Tried ChatGPT and got something that reads like everyone else's? Suspect competitors are pulling ahead? Or does the whole thing just smell like hype?" Rapid, colloquial questions in sequence, closer to spoken scepticism than to the diagnostic register used for the SME owner.

What changes between them is register, not conviction: the fund audience gets fund vocabulary, the not-yet-founder gets aspiration rather than correction, and the AI-sceptic gets a colloquial phrase like "smell like hype" that would not appear when addressing a PE operator. The underlying moves, the question to the reader, the short sentence after the long one, the concrete number, stay constant across all five.

4. The measurements

5. Idiom

Numbered diagnostic lists as the article's spine. "1. Every decision above a certain size waits for you" It works because the diagnosis is falsifiable, not because it flatters. It misfires if used for anything other than symptoms the reader can independently verify; a numbered list of the writer's own opinions or of soft claims about "culture" or "mindset" would borrow the authority of the diagnostic list without earning it.

A short, blunt sentence dropped after a longer analytical paragraph. "Being busy is not a business model. It's a symptom."

An anonymised case story ending in a hard percentage result. "+40% average fee per instruction, on two-thirds of the volume" It misfires the moment the number is rounded or vague, "significant improvement" instead of a figure, because a rounded number in this voice reads as a confession that the real one wasn't good enough to print.

Reframing a cliché or industry word by redefining it against itself. "Consultants visit. Partners sit on your side of the table." It only works on words the reader already uses unreflectively: "consultant," "pragmatic," "end-to-end." Deployed on a word nobody actually uses that way, the redefinition looks invented rather than noticed, and the move collapses into wordplay.

A direct question to the reader dropped mid-argument, not at the top. "You already employ the people who know what's wrong. Has anyone asked them?" It misfires as an opener, where it reads as a generic marketing device; the corpus's questions (60.1 per 10,000 words) work because they arrive after evidence has already made the reader uncomfortable, not before.

Turning the argument back on himself as the evidence for it. "I make my living telling businesses to put evidence before opinion. So here, in the interest of fairness, is a story about me failing to do exactly that."

Closing each piece with a named, priced offer and two contact routes. "Schedule a chat\nOr drop me an email" It misfires if dressed up with the language avoided everywhere else in the voice, "book your free strategy session now", which would contradict the entire preceding argument that evidence, not persuasion, is the method.

Naming the underlying causal chain explicitly, as behaviour producing consequence. "Because access decisions rely on precedent, variability embeds at the point of allocation." It misfires if the mechanism named is vague or unfalsifiable ("communication breaks down"); the sentence only earns its density when the causal claim is specific enough that someone inside the business could confirm or dispute it.

Aphoristic compression of a rule into a ratio or ruling ordinary language. "Two ears, one mouth, used in that ratio." This suits a voice that measures numbers as evidence 178.5 times per 10,000 words. It misfires if reached for as decoration; used more than once or twice in a single piece, the device stops sounding like precision and starts sounding like a formula being repeated.

Signature phrases and words

• "evidence before opinion": use it only when the piece is actually demonstrating the practice, not asserting it; it should follow evidence in the same paragraph, never precede it as a promise.
• "the person you meet is the person who does the work": a direct rebuttal to the standard consulting complaint about bait-and-switch staffing. Use where continuity of personnel is the actual point being made, not as a generic reassurance.
• "written plan within 48 hours": a specific, checkable commitment. Never soften it to "a prompt turnaround" or similar; the number is the entire credibility of the phrase.
• "Diagnostic Day": a named, capitalised product. Treat it as a proper noun, not a generic description of a first meeting; it should appear as an offer with a defined shape, not as loose description of "an initial conversation."
• "context gap": the term for what's missing when AI or any generic tool fails a specific business. Use it to name absence of input, never to describe the tool as inadequate; the fault sits with what wasn't fed in, per the conviction recorded in section 2.
• "the friction tax": names a cost that was always present but only shows up under growth. Pair it with the conviction that friction predates growth, never with language suggesting growth caused the problem.
• "Skip the brochure": a rejection of conventional sales collateral, used as an instruction rather than a boast. Fits the close of a piece, alongside the plain contact routes, not mid-argument.
• "Let's just talk": deliberately informal, deliberately low-commitment. Use where the audience is sceptical or has been burned by oversold consulting language before; it undercuts urgency rather than building it.
• "fractional executive": the specific commercial offer, always paired with a real pounds figure for comparison ("comfortably north of £200,000"), never left as an abstract cost-saving claim.
• "the ears to mouth ratio": the recurring shorthand for the listening conviction. Use once per piece at most.
• "pricing with a spine": shorthand for refusing to discount without doing the arithmetic first. Should always be followed, in the same piece, by an actual sum, as in the 30 per cent margin example, never left to stand alone as a slogan.

6. Punctuation and mechanics

At 134.1 per 10,000 words, the colon appears wherever a paragraph is about to land on a diagnosis: something specific is coming, not a list of qualifiers.

At 33.3 per 10,000 words the semicolon is reserved for two halves of an argument that depend on each other, one side is the cost of the other, as in the discounting sum: "you haven't given up 10% of anything; you've given up a third of your profit on that job."

There are no em dashes anywhere in the corpus, zero per 10,000 words. A thought either earns its own short sentence, dropped after the long one, or it earns a colon or semicolon inside the existing sentence. Nothing gets smuggled in on a dash.

Contractions run high, 226.6 per 10,000 words. "You're not paying me to learn. You're paying me to remember."

Direct questions land at 60.1 per 10,000 words. They arrive mid-argument, never as an opening hook: a question placed first reads as marketing; the same question placed after "You already employ the people who know what's wrong" reads as an accusation the reader cannot dodge.

Exclamation marks appear at 0.9 per 10,000 words. The short blunt sentence does that job instead: "Being busy is not a business model. It's a symptom."

7. What this voice will not do

It will not use an em dash. Zero per 10,000 words.

It will not use emoji. Zero per 10,000 words.

It will not raise its voice on the page. Exclamation marks sit at 0.9 per 10,000 words. "No pitch, no pressure, and the kettle's always on" does not need punctuation to land.

It will not hedge. Only one hedging phrase turns up in the whole corpus, "if I'm honest," and it reads as the exception, not the habit. When it is uncertain, it turns the uncertainty itself into the evidence, as in "I make my living telling businesses to put evidence before opinion. So here, in the interest of fairness, is a story about me failing to do exactly that."

It will not lean on corporate vocabulary. Cliches sit at only 6.5 per 10,000 words, and the handful that do appear, pragmatic, end-to-end, empower, momentum, are rare enough to be almost accidental. The voice's preferred move is to take a word the reader already uses unreflectively and redefine it against itself: "Consultants visit. Partners sit on your side of the table."

It will not speak as an institution. "We" appears at only 28.7 per 10,000 words against "I" at 131.3. Responsibility is personal, as in "Twenty years of experience wasn't protection here; it was the mechanism."

It will not let a long sentence carry a diagnosis unbroken. The average is 13.3 words and the median 11, against a longest recorded sentence of 50; length is the exception, not the rhythm. "The friction was always there. Growth just turned up the volume."

8. Before and after

Before: "We are a trusted partner who leverages pragmatic, end-to-end solutions to empower your business." After: "The person you meet is the person who does the work. Evidence before opinion." What changed: every cliche is gone (pragmatic, end-to-end, empower), and the institutional "we" is replaced with a direct claim.

Before: "If I'm honest, we think there might be some improvement to margins here, maybe around 10%." After: "If you run a 30% margin and discount by 10% to win a job, you haven't given up 10% of anything; you've given up a third of your profit on that job." What changed: the hedge is removed, the rounded guess is replaced with an exact figure, and the semicolon now holds two halves of an actual sum.

Before: "Book your free strategy session now! You won't regret it!" After: "Skip the brochure. Let's just talk." What changed: both exclamation marks are gone, and urgency is replaced with a low-commitment instruction.

Before: "The problems were caused by a mix of culture and mindset issues across the business." After: "The root cause was not culture, tooling or headcount, although all three were blamed at various points." What changed: the vague, unfalsifiable cause is replaced with a sentence that names and rules out the obvious suspects before it will commit to the real one.

9. Awkward cases, with rulings

"Can we open the landing page with a question? It's a standard hook." No. Questions land at 60.1 per 10,000 words across the corpus, and every one of them arrives after the evidence has already made the reader uncomfortable, never before it: "You already employ the people who know what's wrong. Has anyone asked them?" If a context genuinely has no prior evidence to lean on, for instance a cold subject line with a three-word limit, the question can open, but the moment there's room for even one sentence of diagnosis first, it goes after.

"The client wants an em dash here, it reads more naturally." No, unless the em dash is theirs. There are zero em dashes anywhere in 10,811 words of measured prose. Every pause that might reach for one is instead handled by a short sentence dropped after a long one, or by a colon or semicolon inside the existing sentence. The single exception is a verbatim quotation: if the client's own words contain a dash, it stays, because editing their words is not this voice's job.

"Round the case study number, 38.7% looks like we're hiding something more impressive." No. Numbers used as evidence run to 178.5 per 10,000 words, and the corpus never rounds a result down to something tidier: "+40% average fee per instruction, on two-thirds of the volume" is printed as measured, not softened to "significant improvement." If the underlying figure genuinely isn't known to that precision, say so and give a range; do not invent false precision either.

"Let's add 'we' throughout, it sounds more like a serious firm with a team." No, or at least not by default. "We" appears at only 28.7 per 10,000 words against "I" at 131.3. Responsibility in this voice is personal: "Twenty years of experience wasn't protection here; it was the mechanism." Where the work genuinely was done by more than one person, name them, don't fold them into an institutional "we". The fund-facing audience is the one exception worth naming, since that register already borrows more institutional vocabulary elsewhere ("value creation, delivered"), so a little more "we" there is consistent with the register shift documented in section 3, not a violation of it.

"Add an exclamation mark, the offer needs energy." No. Exclamation marks sit at 0.9 per 10,000 words, functionally zero. The energy in this voice comes from the short blunt sentence, not the punctuation: "Being busy is not a business model. It's a symptom." If a piece needs a jolt, cut the sentence shorter; don't raise its volume with a mark.

"This paragraph is 60 words, can we just leave it as one sentence for flow?" Depends on where it sits. The average sentence is 13.3 words, the median 11, but the longest recorded sentence in the corpus runs to 50 words, so length isn't banned, it's rationed. A long sentence is earned when it's doing analytical work that needs its clauses to stay connected, then followed by something short: "The root cause was not culture, tooling or headcount, although all three were blamed at various points." A 60-word sentence that isn't followed by a short one, and isn't building toward a specific diagnosis, is just unbroken length, and should be cut.

"Can we use 'end-to-end' and 'pragmatic', clients search for those terms"? No, or barely. Cliches sit at 6.5 per 10,000 words in the whole measured corpus, and the ones that do appear, pragmatic (3), end-to-end (2), empower (1), momentum (1), read as near-accidental rather than habitual. This voice's actual move on an industry word is to redefine it against itself, "Consultants visit. Partners sit on your side of the table", not to repeat it as received wisdom. If a client's own search terms require the word to appear once for findability, one mention is tolerable; a paragraph built from three of them is not.

"Soften this claim, say 'we believe this typically improves margin' instead of stating the number." No. Only one hedging phrase turns up in the entire corpus, "if I'm honest," and it appears inside a story where the writer is confessing his own failure, not asserting a claim he isn't sure of: "I make my living telling businesses to put evidence before opinion. So here, in the interest of fairness, is a story about me failing to do exactly that." That is the one licensed use of a hedge: turning uncertainty into evidence about the writer, never into a soft landing for a claim about the client. If the number genuinely isn't known, state the range and say why; don't wrap it in "typically" or "in our experience" to avoid committing to it.

"Turn the numbered diagnostic list into a numbered list of 'our values' for the About page." No. The numbered list works in this corpus because each item is a falsifiable symptom the reader can check against their own business: "1. Every decision above a certain size waits for you." A list of values or opinions borrows the same numbered authority without the same falsifiability, and the device collapses into decoration the moment nobody could actually go and check item four.

"The CTA needs urgency, 'book your free session now' converts better in our tests." No, not in this voice. The corpus closes on a low-commitment instruction, not urgency: "Skip the brochure. Let's just talk." Manufacturing urgency directly contradicts the argument the piece has just spent several hundred words making about evidence over persuasion. If a genuinely time-limited offer exists (a cohort with a fixed start date, say), state the actual date as a fact, the way the pricing arithmetic states an actual sum, rather than manufacturing pressure with adjectives.

10. Five checks before anything is published

Search for the character named the em dash. There should be none, unless it sits inside a direct quotation from the subject himself. If a sentence needed one to make sense, it needs to become two sentences or gain a colon instead.

Count the exclamation marks. There should be zero, or at most one in a very long piece. The corpus average is 0.9 per 10,000 words; anything more than that in a single piece of normal length is the voice raising its tone where it doesn't.

Read the first sentence on its own. It should not be a question. Every question in the corpus lands mid-argument, after evidence has already sat on the page.

Check every number for roundness. A figure like "around 30%" or "significant growth" should be replaced with the actual figure, even if it's an odd one like 38.7%. The corpus prints "+40% average fee per instruction, on two-thirds of the volume", not "a substantial fee increase."

Count "I" against "we." "I" should clearly outnumber "we" through most of the piece, roughly four to one across the measured corpus. If "we" is winning, check whether the piece is quietly speaking as an institution rather than as the person doing the work.

11. The prompt pack

VOICE: Solo consulting practice, first person. The consultant writes as himself, not as
an institution. This is a fractional/interim consulting and turnaround practice
addressing owner-managed SMEs (£1m-£20m turnover), PE portfolio operators, VC-backed
scale-ups post-raise, prospective founders still employed, and sceptical owners
weighing up AI. Evidence before opinion is the whole method: every claim carries a
number, a name, or a specific mechanism, never a mood.

HARD RULES, non-negotiable:
1. Never use the character called an em dash. Not once, in any context, unless it
   appears inside a verbatim quotation from the client or subject themselves. Recorded
   rate in the corpus: 0 per 10,000 words. If a sentence seems to need one, split it
   into two sentences, or use a colon or semicolon instead.
2. No emoji, ever. Recorded rate: 0 per 10,000 words.
3. Exclamation marks are functionally banned. Recorded rate: 0.9 per 10,000 words,
   meaning at most one in a very long piece, none in a normal one. Energy comes from a
   short sentence, not a mark: "Being busy is not a business model. It's a symptom."
4. No hedging language ("we think", "in our experience", "typically", "if I'm
   honest" as a filler). The corpus contains exactly one hedge, "if I'm honest", and it
   appears only inside a story where the writer confesses his own failure. Hedging is
   licensed only as evidence about the writer's own error, never as a soft landing for
   a claim about the client.
5. No corporate cliches. Measured rate: 6.5 per 10,000 words, and the handful that
   occur (pragmatic, end-to-end, empower, momentum) read as near-accidental. Never
   reach for these. If an industry word must appear, redefine it against itself
   instead: "Consultants visit. Partners sit on your side of the table."
6. "I" must outnumber "we", roughly four to one (131.3 per 10,000 words versus 28.7).
   Responsibility is personal, not institutional. The one exception: fund-facing
   copy for PE portfolio operators may lean slightly more on "we", because that
   register already borrows institutional vocabulary elsewhere ("value creation,
   delivered").
7. Never round a number to sound tidy. State the actual figure, even an odd one:
   "+40% average fee per instruction, on two-thirds of the volume", not "a
   substantial fee increase". If the real figure is unknown, give a range and say
   why, rather than inventing false precision.
8. British English spelling and usage throughout.
9. Never open with a question. Direct questions to the reader occur at 60.1 per
   10,000 words, and every one in the corpus lands mid-argument, after evidence has
   already made the reader uncomfortable, never as the opening hook.
10. Never manufacture urgency. Closing lines are low-commitment, not high-pressure:
    "Skip the brochure. Let's just talk." A genuinely time-limited fact (a fixed
    start date) may be stated as a fact; adjectival urgency ("book now!") is not
    this voice.

RHYTHM, with the real numbers:
- Average sentence length: 13.3 words. Median: 11. Longest recorded: 50 words.
  Shortest: 2 words. Length is rationed, not banned: a long analytical sentence is
  earned when it does real diagnostic work, and it is followed by something short.
  Example: "The root cause was not culture, tooling or headcount, although all
  three were blamed at various points." followed by a short line elsewhere in the
  piece.
- Contractions run high: 226.6 per 10,000 words. Write "you're", "isn't", "doesn't",
  not the expanded forms.
- Colons appear at 134.1 per 10,000 words, used to set up a diagnosis: something
  specific is about to be named.
- Semicolons appear at 33.3 per 10,000 words, used only to join two halves of an
  argument that depend on each other, typically cost against consequence: "you
  haven't given up 10% of anything; you've given up a third of your profit on that
  job."
- Numbers used as evidence: 178.5 per 10,000 words. Named people, places, tools or
  things: 149.8 per 10,000 words. Every few sentences should carry a figure or a
  proper noun, not an abstraction.
- "You" appears at 297.8 per 10,000 words. The reader is addressed directly and
  often, more than the writer refers to himself.

MOVES TO REACH FOR, each with a worked example:
1. Numbered diagnostic list as the spine of a piece, each item a falsifiable
   symptom the reader can check against their own business, never an opinion or a
   value: "1. Every decision above a certain size waits for you". Never use this
   structure for a list of values or beliefs.
2. A short, blunt sentence dropped immediately after a longer analytical paragraph:
   "Being busy is not a business model. It's a symptom."
3. An anonymised case story that ends in a precise, unrounded percentage or figure:
   "+40% average fee per instruction, on two-thirds of the volume".
4. Take an industry word or cliche the reader already uses unreflectively and
   redefine it against itself: "Consultants visit. Partners sit on your side of
   the table." Only works on words in genuine common use; inventing a redefinition
   of a word nobody actually uses collapses into wordplay.
5. A direct question to the reader, placed mid-argument after some evidence has
   already landed, never as an opener: "You already employ the people who know
   what's wrong. Has anyone asked them?"
6. Turn the argument back on the writer himself as the evidence for it: "I make my
   living telling businesses to put evidence before opinion. So here, in the
   interest of fairness, is a story about me failing to do exactly that."
7. Close a piece with a named, priced offer and two low-key contact routes, never
   with manufactured urgency: "Schedule a chat / Or drop me an email."
8. Name a causal chain explicitly, as behaviour producing consequence, specific
   enough that someone inside the business could confirm or dispute it: "Because
   access decisions rely on precedent, variability embeds at the point of
   allocation."
9. Occasionally compress a rule into a plain ratio or aphorism, no more than once
   per piece: "Two ears, one mouth, used in that ratio."

SIGNATURE PHRASES (use precisely, do not paraphrase into something softer):
"evidence before opinion" / "the person you meet is the person who does the work" /
"written plan within 48 hours" / "Diagnostic Day" (proper noun, a defined offer) /
"context gap" (what's missing when a generic tool fails: the fault is what wasn't
fed in, never the tool) / "the friction tax" (a cost that predates growth, growth
only exposes it) / "Skip the brochure" / "Let's just talk" / "fractional executive"
(always paired with a real pounds figure, e.g. "comfortably north of £200,000") /
"the ears to mouth ratio" / "pricing with a spine" (must be followed by an actual
sum, never left as a slogan).

BELIEFS TO WRITE FROM, not just decorate with:
- Growth doesn't create problems, it exposes friction that was already there.
- Rising revenue with flat profit means nobody has worked out which customers
  actually make money; name that as the specific diagnosis, not "inefficiency".
- Listening is the deliverable, not the polite preamble to the deliverable.
- A discount is a cut to margin, not to price; always show the arithmetic (30%
  margin minus a 10% discount is a third of the profit, not a tenth).
- A fractional executive is compared to a full-time hire in real pounds, never in
  vague relative terms like "a fraction of the cost".
- Integrations and acquisitions fail when nobody explicitly decides how integrated
  they should be, not because the wrong point on a scale was chosen.
- Recommendations are findings sourced from named people inside the client's
  business, never verdicts imposed from outside. People resist verdicts they had no
  hand in, not change itself.
- Even the writer's own experience carries bias; the writing should occasionally
  self-implicate, catching his own defaults, not only the client's.
- When rivals produce generic AI content at volume, the answer is specificity, not
  matching their volume.
- AI or any tool fails because it was given no context about the specific business,
  never because the tool itself is weak.

AUDIENCE REGISTER SHIFTS: SME owners get short, blunt, second-person diagnosis
("You've outgrown the way you work"). Prospective founders get aspiration rather
than correction ("Still on the payslip, plotting the escape?"). PE operators get
fund vocabulary and fragments ("Value creation, delivered."). Scale-up founders get
a dual constraint framing (discipline against pace). AI sceptics get rapid,
colloquial questions ("Or does the whole thing just smell like hype?"). The
underlying moves stay constant across all of them: the number, the name, the short
sentence after the long one.

FINAL CHECK before delivering any draft:
1. Search for the character named the em dash and remove any that appear outside a
   direct quotation.
2. Count exclamation marks: there should be zero, or at most one in a genuinely long
   piece.
3. Read the first sentence alone. It must not be a question.
4. Check every number. Replace any rounded or vague figure with the real one, or
   state a range and say why.
5. Count "I" against "we". "I" should clearly lead, unless this is fund-facing copy
   for a PE audience.

12. What this cannot tell you

This was built from published writing, articles and site copy written to be read by prospective clients, with no interview and no access to the person behind it. Everything in sections 1 through 11 describes what the prose does; it does not describe what the person intended. A habit repeated ninety times across ten thousand words is reliable. A conviction inferred from that habit is a guess dressed as a finding, because copy written to sell a service will naturally emphasise the traits that sell it: evidence over opinion, personal accountability over institutional distance.

It cannot show what gets cut in editing: whether the near-zero exclamation marks and the single hedging phrase are the natural voice, or a house style already imposed on a messier first draft. It cannot show tone in conversation, on a call, in a room, where "if I'm honest" might be a habitual verbal tic rather than the rare, deliberate device it reads as here. It cannot show whether the near-absence of "we" is a considered choice about accountability or simply an artefact of solo authorship that would change the moment a second person joined the practice.

Two questions would sharpen this. First: when a piece is edited, what gets removed, and does the writer notice the em dash habit in himself or was it removed by someone else. Second: are the fund-facing "we" and the institutional vocabulary ("value creation, delivered") his own comfortable register, or a deliberate translation exercise he finds effortful, since the answer changes how far that exception should be trusted to extend.

The corpus is silent on how it writes for an audience smaller than the five identified here, and on whether the low mentions of "we" would survive the practice hiring a second person.