Table of Contents
TL;DR
- A chat surface fills every gap in a thin brief with the statistical middle of its training data. The middle reads as average prose, which is why short prompts produce generic copy.
- A working brief carries six load-bearing components — the named buyer, the position, the voice, the proof order, the structural pattern, the output format. Drafts written against the six pass the editorial bar 70-80 percent of the time.
- Three converging frameworks describe the same shape — Palmer six-component, Everworker CARE, Da Silva CART. The frameworks differ in name, but the load-bearing components do not.
- Voice prompting works through examples, not adjectives. Five to fifteen short passages of brand writing teach the chat surface more than any abstract description ever will.
- The brief saves the time it costs. Forty-five minutes upfront on the brief beats ninety minutes downstream on rework. The chat surface earns its speed at the drafting stage. The human earns the brand standard at the brief stage.
A small-business owner pastes a 90-word prompt into ChatGPT. Forty seconds later, a homepage hero comes back, polished and forgettable.
The owner edits for an hour. The result still reads like a category description. The buyer scrolls past the page on the way to a competitor.
A thin brief is an open mic at a chat surface. A fat brief is a sound check that calibrates the room before the singer walks on.
Most teams blame the chat surface when the draft reads generic. The fix sits one step earlier. The brief decides what comes back.
A brief that names six load-bearing components produces drafts that pass the editorial bar most of the time. A brief that names the topic and waves at the rest produces drafts that read like every other page in the category.
This is not prompt engineering. This is the same brief discipline a senior copywriter has carried for thirty years, written down so the chat surface can act on it.
Why does a generic brief always produce generic ChatGPT copy?
A chat surface has read the entire internet. It remembers no client specifically.
Whatever the brief leaves out, the chat surface fills with the statistical middle of its training data. The middle of the training data is, by construction, average prose. The output reads as the average page in the category because the brief described the category, not the company.
A brief that names the buyer, the position, the voice, and the proof produces copy that can only belong to one business. A brief that names a category produces copy that reads like every other page in the category. The shape of the output mirrors the shape of the input.
This is why the same flat result keeps coming back. The fix starts with seeing why prompts produce generic copy before you blame the AI tool. Name the buyer, and the draft has somewhere to stand.
The lesson is not that the chat surface writes badly. The lesson is that the chat surface writes whatever the brief told it to write. The brief is the prompt, and the prompt is the upstream decision the draft expresses.
What does a working ChatGPT brief look like?
A working brief carries six things the chat surface cannot infer.
The buyer named specifically. Not "small business owners," but the named ICP, the named industry, the named annual revenue band, the three sentences on the problem in the buyer’s own words. The chat surface needs the buyer’s vocabulary, not a description of the buyer.
The position. What context should the buyer place the offer in? Which competing alternatives does the page argue against?
Which unique attributes does the offer carry that the alternatives do not? The brief carries the answer — the page expresses it.
The voice. Five sentences on how the brand sounds. Three sentences on how it does not sound.
Three banned phrases. Three brand-axis commitments from the five-vector voice spec.
The proof order. The named testimonials, results, and authorities the page should lead with. The brief says which proof goes in the hero, which goes mid-page, which goes pre-CTA.
The structural pattern. The named framework — a problem-agitate-solve, a four-question landing page, a long-form sales letter — or a custom section order. The chat surface needs the page shape stated, not implied.
The output format. Length. Heading rules.
CTA shape. The brief that names all six produces drafts that pass the editorial bar 70 to 80 percent of the time without major edits.
For more on the editing pass that runs after the draft comes back, see how to edit AI-generated content.
Which six components close the AI-vs-human quality gap?
Three converging frameworks describe the same underlying discipline.
Joel Palmer’s six-component prompt covers reader specifics, awareness level, the feeling the copy must move the reader through, the differentiated mechanism, the proof order, and the structural pattern. Six load-bearing parts.
Everworker’s CARE covers Context, Ask, Rules, Examples. Four parts that compress the same six into a tighter shape.
Carlos Da Silva’s CART covers Context, Audience, Role, Task. Same load-bearing components, repackaged again.
The three frameworks describe the same shape from three angles. The lesson is not to memorize one framework. The lesson is that every working AI brief carries the same components, regardless of which acronym sits on top. A brief is working when the buyer, the position, the voice, the proof, the structure, and the output format are all named on the same page.
The acronym is a memory aid. The work is the same.
How do you put your voice into the brief without cloning yourself?
Voice prompting works through examples, not adjectives.
Paste five to fifteen short passages of writing the brand sounds like. The chat surface infers the register from the examples better than from any description of it. "Write in a confident, friendly, expert voice" tells the chat surface nothing it cannot already do. Five paragraphs of the brand’s actual writing tells it everything.
Add a banned-phrase list. The words and phrases the brand never uses, written in plain rows. The chat surface obeys the negative list more readily than the positive instruction.
Add three brand axes from the five-vector voice spec. Formality. Warmth. Directness.
Humor. Technicality. Pick three. Commit to a position on each.
"Plain over formal. Direct over hedged. Practical over theoretical." The chat surface now has scoreable targets to write against.
The voice spec fits on a single page. The page does more work than ten paragraphs of stylistic description.
Which phrases should the brief tell ChatGPT never to write?
A documented family of phrases dominates AI-drafted copy.
"Innovative." "Next-level." "Seamless." "High-quality." "Synergy." "Utilize." "We are excited to announce." "Cutting-edge." "Best-in-class." The cluster was generic before AI tools existed. The chat surface produces the cluster at higher rate because it dominates the training data.
The fix is mechanical. Maintain a banned-phrase list per brand. Include the list in every brief.
Update the list when a new phrase starts repeating. The list grows over time, and the brief gets stronger because of it.
A short banned-phrase list does more work than a long stylistic instruction. Ten well-chosen banned phrases prevent more generic prose than a hundred words of positive guidance. The reason is structural.
The chat surface defaults to the high-frequency phrases in its training data. Removing those defaults forces the chat surface to reach for less-frequent phrases, which read as more specific because they are.
The banned-phrase list is the single highest-impact paragraph in a brief.
Why does a longer brief save time on the back end?
A short brief produces a draft that needs heavy rework. A long brief produces a draft that needs light editing.
The total time per finished page is lower with the long brief, even though the upfront work feels heavier. A team that spends 45 minutes on a 600-word brief and 20 minutes editing the resulting draft beats a team that spends 10 minutes on a 100-word brief and 90 minutes editing the resulting draft. The math is unforgiving.
Most of the saved time is not at the keyboard. It is in the rework round-trip. A bad draft pulls the editor back into questions the brief should have closed. "What is this page actually for?" "Who is this written to?" "What proof do we have?" Those questions belong upstream of the chat surface, not downstream of it.
The brief is where customer-research, positioning, and voice work concentrate. The draft is where the chat surface earns its speed. The two stages do not compete. They cooperate, and only when the brief carries its weight.
A page that ships from a thin brief is a page that ships at the median quality of the training data. A page that ships from a fat brief is a page that ships at the brand’s quality bar. For the before-state — what generic AI copy looks like and why — see why your AI website copy reads generic.
Other questions worth answering
When should you scrap a draft entirely and re-prompt the chat surface rather than editing it?
Re-prompt when the brief was thin and the draft’s shape was wrong from the first sentence. Edit when the brief was solid and only word choice or rhythm drifted.
Reverse-outline the draft into bullet points first. If the argument arc reads wrong, re-prompt. If the arc is right but the prose is flat, edit. The two-minute diagnostic mirrors Joel Palmer’s brief-as-prompt frame.
How do you mine customer reviews for the words your buyers use about their problem?
Pull verbatim phrases from support tickets, recorded sales calls, one-star reviews, and survey free-text fields. Cluster the phrases by theme. The highest-frequency wording per cluster becomes the anchor language for headlines and the problem section of the spec.
Joanna Wiebe’s voice-of-customer methodology built the modern shape of this exercise. Forty reviews and three sales calls usually surface enough material for a small business.
How should two teammates align on the spec when each prefers a different style?
Score each axis from one to five and have the team commit. Hold roughly a 30-minute meeting where each editor picks a spot on five axes — formality, warmth, directness, humor, technicality. The lowest score wins. The exercise forces a compromise that scattered drafts will otherwise hide.
Update the doc when a new editor joins. Mailchimp’s voice-and-tone playbook treats this codification as the founding act of consistent prose.
Can the same spec carry across multiple chat surfaces such as Claude or Gemini?
About 80 percent of the spec carries across surfaces. ChatGPT, Claude, and Gemini all infer register from example passages and obey banned-word lists. Roughly 20 percent of the work is per-surface tuning. Claude tends to soften directives.
Gemini tends to add transitional padding. In 2026, Search Engine Land documented well-prompted drafts hitting 70 to 80 percent on-voice across surfaces, regardless of which engine drafted them.
How do you keep the spec current when positioning shifts mid-quarter?
Revisit the spec on two triggers, not on a calendar. First trigger: when the offer or buyer materially shifts, often from a new segment or price change. Second trigger: when three drafts come back off-voice for the same reason.
April Dunford’s positioning framework gives the easiest test — if the competitive alternatives changed, the spec needs rewriting.
What does your next ChatGPT brief actually look like?
One page. Six headed sections.
Buyer specifics. The named ICP. Three sentences on the buyer’s problem in the buyer’s own vocabulary, pulled from support tickets, reviews, or recorded sales calls.
Position. The five Dunford components on a single page. Competitive alternatives. Unique attributes.
Value with proof. Best-fit buyer. Market category.
Voice. Five passages of brand writing. Three banned-phrase rows. Three brand-axis commitments.
Proof. The named testimonials and results in the order they should appear on the page. Hero-adjacent first, mid-page next, pre-CTA last.
Structure. The named section pattern, in the order the page should follow. Hero, problem-agitate, solution, proof, objections, CTA, FAQ. Or whatever the offer demands.
Output format. Length in words. Heading rules. CTA shape and language.
Forty-five minutes to write the first time. Fifteen minutes to update for the next page. The draft that comes back fits the brief because the brief left no room for guesswork. The chat surface produces what it was told to produce, and the editor’s pass becomes a final taste round rather than a rescue operation.
If you have written a few briefs and the drafts still come back generic, you can contact me here. Send the brief and one paragraph of the draft it produced. I will read both and name the load-bearing component the brief is missing.
No charge and no follow-up call. The brief that ships next week is the one that earned its weight.