Buyer read
Use this skill when a finished page, deck, or piece of client copy is about to go to the client or be published, and a real ICP or buyer research document exists. Produces a ranked list of what would make the target buyer stop trusting it, read in character, with every finding citing a line in a real source, plus what to cut and a score. Catches what craft reviews miss: scenarios that can't happen in the buyer's week, an offer never plainly stated, missing price, unused proof, seller vocabulary. Also use for "/buyer-read", "review this as the buyer", "have the ICP look at this", "what would a CFO think of this page", "be hypercritical about this copy". Not for grammar or voice checks, and never for rewriting.
Chris DuBoisFounder and Coach, Dynamic Agency OSApplies to finished copy, run last, before it goes to the client or goes live. Produces a ranked, fully sourced buyer read with a score, or a refusal when no real buyer document exists.
Craft reviews grade what is on the page. They can't see what is missing, and they can't tell whether the writer understands how the buyer's business runs. This puts on the buyer's body, reads the piece cold, and reports where they would have stopped believing it.
The play
- Gather the piece (file, URL, or pasted text; if a URL won't open, ask for the text), the ICP or buyer document, any real buyer quotes (call transcripts, sales notes, reviews, survey answers), and client context (profile, proof, competitors, brand voice). Read references/worked-example.md before the first pass.
- No ICP or buyer research? Stop. Say what's missing and ask for it. Don't proceed on assumptions, even if the user asks you to guess.
- Build the persona before reading a word of the copy, following references/persona-and-checks.md. Where real buyer quotes exist they outrank the ICP: build from what buyers said, use the ICP for gaps, and report any disagreement between the two.
- Cold walk. Read the piece top to bottom in first person, in character, without looking at any brief or rubric. Note where you'd stop, disbelieve, get confused, or feel talked down to. Fix nothing yet.
- Run the nine checks in references/persona-and-checks.md, one at a time. Most produce nothing on most pieces, and that's correct. A check missing its source is marked "not run" with the source named.
- Evidence pass, separately from step 5. Attach a source to every candidate finding: a quoted line from the buyer document, a moment in a transcript, a documented competitor fact. A failed cold question needs no buyer line; cite the piece and what it never answers. Anything else with no citation gets cut, not softened. Then drop anything that is a writing preference wearing a persona.
- Rank by cost: the finding that loses the most buyers goes first. Own the writer's errors in a sentence. Name which existing section pays for the additions, or say the piece is too short to have one. Findings that cost conversions go above findings that are survivable.
- Score in a separate pass with references/scoring-rubric.md, then revise the lowest dimensions and re-score the whole read. Maximum three cycles. The rubric says what a stuck dimension means.
- Deliver, in order: the persona in two or three sentences with citations; ranked findings, each with what breaks, the evidence line, and the fix or decision it forces; a second tier; what to cut; questions for the client; what was cut in the evidence pass, one line each; the score sheet and
Buyer read: __/100, weakest <dimension>, cycles run: N.
What good looks like
- First thing to check that most people skip: the three cold questions. What am I being sold, what does it cost, show me one you've built. A beautifully written page that can't answer all three has an absence problem.
- The second thing: domain reality. A scene that can't happen in the buyer's week (a dead phone on their busiest day) triggers the deadliest objection there is: these people don't understand my business.
- The mediocre version: a persona that reads "a busy owner," ten findings that are style notes in costume, ranked in the order they were noticed, with no citations and nothing named to cut. It looks rigorous and is unfalsifiable.
- The read is right when every finding survives the question "where does it say that?", the persona is named, sized and situated from the documents, and it scores 85/100 or better with evidence traceability at 8 or higher.
- A three-finding read that passes beats a padded ten. If the piece is mostly fine, say so.
Rules
- MUST refuse to run without a real ICP or buyer research document. An invented persona produces confident, plausible, unfalsifiable findings, the worst failure of this skill, and it looks exactly like success.
- MUST cite a real source for every finding, or cut it.
- MUST write "not in the documents" for any persona element the sources don't cover. Invent only the name.
- MUST rank by cost, not by discovery order.
- NEVER rewrite the copy. This critiques; it names the fix or decision and stops.
- NEVER pad to reach a number of findings.
- NEVER present a failing read as finished. Deliver it marked not done, with the score sheet and the one line blocking the pass.
- NEVER quote the worked example as evidence.
references/persona-and-checks.md
Persona and the nine checks
Open this for the persona (step 3) and the checks (step 5).
The persona block
Write a short character block before reading the copy. Pull every element from the documents and cite where each came from.
- Name, age and role. Invent only the name. The rest comes from the documents.
- The business, sized. Locations, staff, tenure, revenue band.
- What they name first when asked what's hard, in their own words if the documents have them.
- Their failure history with vendors like this one. The burn scars make them skeptical.
- How they arrived at this page. A referral, a search, a conference badge scan. Different arrivals need different things above the fold.
- The physical moment. Where they are, on what device, with how long.
If the documents don't cover an element, write "not in the documents" and leave it blank. Never fill it with a guess. Blanks lower BR1, which is the honest result of a thin document. A range stays a range.
Specificity is what makes the findings concrete. "A busy owner" finds nothing. "Fifty-one, runs a nine-truck HVAC company his father started, checking his phone in the van between calls" finds real problems.
The nine checks
Run each one deliberately. If a check needs a source you don't have, mark it "not run" and name the missing source.
- The three cold questions. What am I being sold, what does it cost, and show me one you've built. A page that can't answer all three has an absence problem however well it's written.
- Domain reality. Does every scene survive this buyer's operating reality? Check hours, staffing, seasonality, and how they get customers. This catches errors no writer notices and every buyer does.
- Absence audit. What would this buyer go looking for that isn't here? Price, work samples, geography, who they'd work with, what happens if it fails.
- Proof inventory. List the proof the client already has and check it against the page. Proof sitting unused in a transcript from three weeks ago is a research failure.
- Peer vocabulary. Would this buyer say this word to a peer, unprompted? Apply it hardest to the main CTA and anything above the fold.
- Falsifiability. Does the page make a claim about the reader's business that they can disprove in thirty seconds? If it's wrong about them, the page breaks.
- Self-assessment honesty. If the page asks the buyer questions (a quiz, a scorecard), can they answer them? A diagnostic that asks what only a vendor can measure returns a guess.
- Incumbent adjacency. Who is the buyer comparing this to, and what does that competitor lead with? If the incumbent leads with a credential this page can't answer, that's a finding even on a strong page.
- The named-first problem. Does the page acknowledge what this buyer names first? Run this only if the documents say what comes first; list order is not evidence. Surfacing the gap is different from recommending it go on the page. A problem the client can't solve usually belongs in the sales conversation. Report the gap and say which you think it is.
references/scoring-rubric.md
Buyer read scoring rubric
Open this to score (step 8). Score in a separate pass, pessimistically, taking the lower number when torn. Rate each dimension 0-10, multiply by its weight, divide by 10, and add them up for a total out of 100.
| # | Dimension | Weight | 10 looks like | 6 looks like | 2 looks like |
|---|---|---|---|---|---|
| BR1 | Persona fidelity | 12 | Named, sized and situated, every attribute traceable | Real but generic in places | "A busy owner," no sourcing |
| BR2 | Evidence traceability | 20 | Every finding cites a source | One or two rest on reasoning | Confident assertions with nothing behind them |
| BR3 | Domain reality | 18 | Every scenario tested, violation found or none stated | Checked loosely | Never asked if it could happen |
| BR4 | Absence detection | 16 | Names what the buyer wouldn't find, including what craft review can't see | Mostly obvious absences | Only critiques what's on the page |
| BR5 | Severity honesty | 12 | Ranked by cost, honest tiers | A mid finding is inflated | Discovery order, or everything urgent |
| BR6 | Actionability | 12 | Each finding names the fix or decision, plus what to cut | Fixes implied, or no cut | Complaints with no path |
| BR7 | Restraint | 10 | Nothing padded, says so if the piece is mostly fine | One or two preferences in costume | Padded to hit a number |
PASS: 85/100, no dimension below 6, and BR2 at 8 or higher.
Draft, score, revise the lowest dimensions, then re-score the whole read. Maximum three cycles.
When a dimension is stuck
- BR1, BR2 or BR3 stuck means the source material is too thin. Stop revising, deliver the read marked not done, and ask for the transcript or the missing ICP section. Reasoning around the hole is the fabrication this skill exists to prevent.
- BR4 stuck usually means the piece has no significant absences. Say that and deliver a short read.
- BR7 stuck means findings are writing preferences. Drop them and shorten.
references/worked-example.md
Worked example
Everything here is invented for illustration, including the agency, the buyer document and the scores. Use it to see the shape and the standard of a read, and never quote it as evidence.
The piece: the homepage for Northline, an agency selling websites and local ads to HVAC contractors.
The hero: "It's 2pm on a Tuesday in July and your phones are dead. Northline fills your schedule with AI-powered demand generation."
The rest of the page: a three-step "Our Process" section, a testimonial, and a "Book a strategy call" button. It shows no work samples and no price.
Buyer document lines used:
- Section 1: "Owns a nine-truck HVAC company, second generation, age 51. Names hiring techs as his hardest problem."
- Section 5: "Most find vendors through Google, on a phone, between calls."
- Section 2: "Busiest weeks are July and January. Slow months are April and October."
- Section 3: "Won't call a marketing company without seeing a contractor's site they built."
- Section 4: "Got burned on a $2,500/mo contract with no idea what he was paying for."
- Sales call, March: "Don't tell me about demand gen. Tell me how many calls."
The read
Persona. Rick Sobol, 51, runs a nine-truck HVAC company his father started (ICP, section 1). He's on his phone in the van between calls, and he found Northline through a Google search for HVAC marketing (ICP, section 5). His last agency cost him $2,500 a month and he never learned what it did (ICP, section 4).
Findings that cost conversions
1. The opening scene tells Rick these people have never run an HVAC company. A July Tuesday is the week his phones don't stop. He'd be turning work away, and the page opens on the one afternoon that can't happen to him. Evidence: "Busiest weeks are July and January. Slow months are April and October." (ICP, section 2) Fix: move the scene to April. This one is the writer's mistake.
2. He can't find a site Northline built, so he leaves. The page has no work samples, and that's the first thing he looks for. Evidence: "Won't call a marketing company without seeing a contractor's site they built." (ICP, section 3) Fix: show two contractor sites directly under the hero. If none exist, the client has to decide what to show.
3. There's no price, and that repeats his worst vendor experience. Evidence: "Got burned on a $2,500/mo contract with no idea what he was paying for." (ICP, section 4) Fix: state a starting price and what it includes.
Worth fixing
4. "Demand generation" is seller vocabulary in the hero. Evidence: "Don't tell me about demand gen. Tell me how many calls." (Sales call, March) Fix: name calls booked.
5. The page ignores what he names first. Evidence: "Names hiring techs as his hardest problem." (ICP, section 1) Northline can't solve hiring, so this belongs in the sales conversation and stays off the page.
What to cut
The "Our Process" section pays for the work samples and the price. Rick didn't ask how the agency works, and nothing in the source documents says he cares.
Questions for the client
Can Northline publish a starting price, and which two sites can it show?
Cut in the evidence pass
The page's green color felt off, but no source supports that, so it's gone.
Score
BR1 9, BR2 9, BR3 9, BR4 8, BR5 9, BR6 9, BR7 9
Buyer read: 88/100, weakest BR4, cycles run: 1.