Prowrit / prompt engineering studio
pipeline · 3 steps

The Anti-Slop Article — the workflow, engineered

You want: a post on your topic that no competitor could have published.

why this shape choosing what only you can say and saying it well are two different jobs with different failure modes — and 'could a rival run this paragraph unchanged' must be judged by fresh eyes, or the writer grades their own homework

✓ Reviewed before you saw it passed all 5 checks — an automated, fresh-context review

how this was engineeredanalyzed 2 assumptions surfaceddesigned 3 steps · 1 gatereviewed 5/5 checks ✓

3 chat runs · 3 things you paste in · quality-checked (retries ≤2×)

you start withyour topic · what your product knows firsthand · three competing articles (optional)↓ the angle sheet: one defended angle, a tagg…↓ the full draft, 900-1,400 words, every OURS…↓ ✓ pass — continuesyou getthe finished 900-1,400-word post plus its differentiationreport — quoted evidence that every section spends a cla……1agentAngle & Claimspick one contrarian-but-defensible angle and budgetevery claim as OURS-ONLY or COMMON, section by…2agentDrafterwrite the post section by section under a voicecontract, spending exactly the claims each section…3quality checkDifferentiation Gatetest every paragraph against one question: could acompetitor have published this unchanged?↩ fix-list — back to 2. Drafter · ≤2×

The Anti-Slop Article — the workflow in words

Shape: pipeline. 3 steps, in this order.

  1. Step 1 of 3: Angle & Claims — Agent. pick one contrarian-but-defensible angle and budget every claim as OURS-ONLY or COMMON, section by section. Takes the material you supply. Produces the angle sheet: one defended angle, a tagged claims inventory, and a section outline with claims budgets.
  2. Step 2 of 3: Drafter — Agent. write the post section by section under a voice contract, spending exactly the claims each section budgets. Takes Angle & Claims. Produces the full draft, 900-1,400 words, every OURS-ONLY claim marked with its inventory number.
  3. Step 3 of 3: Differentiation Gate — Quality check. test every paragraph against one question: could a competitor have published this unchanged?. Takes Drafter. Produces the differentiation report: pass/fail per rubric item with quoted evidence, and a numbered fix-list on failure. On failure, sends its fix-list back to Drafter, up to 2 times.

Final output: the finished 900-1,400-word post plus its differentiation report — quoted evidence that every section spends a claim only you can make

in plain words Your topic and firsthand knowledge go to Angle & Claims, which picks one defensible angle and tags every claim OURS-ONLY or COMMON with a spend budget per section; the Drafter writes under a voice contract, spending only what's budgeted; the Differentiation Gate reads with fresh eyes and sends any paragraph a competitor could run straight back — you get a post only you could publish.

What you'll walk away with — an excerpt from a real run of this workflow

THE OPENING — from the real draft:

"A trades business with a 19% no-show rate on unpaid quote visits has a 4%
no-show rate on paid diagnostic visits. Same customers, same markets, same
technicians driving the same trucks. The only variable that changed is
whether money moved before the appointment. That gap — fifteen percentage
points — is the whole argument."

THE DIFFERENTIATION REPORT — PASS, one paragraph flagged:
"Every sentence in it could appear in a Calendly blog post, a Square
merchant tips article, or a HubSpot explainer without alteration… the
draft allowed connective tissue to expand into a full paragraph that earns
no proprietary weight."
The gate reads your article the way the market will.

What I assumed — if a guess is wrong, fix the brief and rerun

  • assumed 'our voice' can be pinned by concrete bans plus an optional sample paragraph — not a full style guide
  • assumed you can paste real firsthand material (measured numbers, customer stories, opinions you'll sign) — the workflow spends that material, it cannot invent it

Would sharpen the result

  • where the post will live and who reads it (practitioners vs. executives) — register shifts with the reader

Run it — 3 steps, in this order

Run each step in a fresh chat, in order. Step 1 wants your topic, your firsthand material (pull real numbers from analytics, real quotes from support tickets and sales calls — ten minutes of gathering buys the whole differentiation), and the top three articles ranking for your topic (search it, copy the top three organic results, or write none). Paste each step's output into the next. If the gate fails, paste its fix-list under the angle sheet in the Drafter and run it again.

  1. 1

    Angle & Claims

    Agent◆ your strongest model

    pick one contrarian-but-defensible angle and budget every claim as OURS-ONLY or COMMON, section by section

    give it your material — whatever this step's {{PLACEHOLDER}} asks for it returns the angle sheet: one defended angle, a tagged claims inventory, and a section outline with claims budgets
    Show the prompt for this step
    ## Role
    You are an editorial strategist choosing the one angle a piece can actually defend. Your enemy is the consensus post — the article that ranks because it repeats what already ranks.
    
    ## The material
    The topic:
    ```
    {{PASTE_YOUR_TOPIC_HERE}}
    ```
    What we know firsthand — data we measured, customer stories we witnessed, opinions we are willing to sign:
    ```
    {{WHAT_YOUR_PRODUCT_KNOWS_FIRSTHAND}}
    ```
    Three competing articles — search the topic, paste the top three organic results (title plus body or a solid excerpt), or write none:
    ```
    {{THREE_COMPETING_ARTICLES_OR_"none"}}
    ```
    
    ## Instructions
    1. Read all the material before proposing anything — the angle must come from what we know, not from what the topic usually attracts.
    2. List what the competing articles agree on. That consensus is the no-fly zone: anything all three say, we don't lead with. If none were pasted, write the three points a generic page-one post would make and treat those as the consensus.
    3. Pick ONE angle that is contrarian-but-defensible: it pushes against the consensus, and every load-bearing claim behind it is backed by the firsthand material. Reject angles that are merely contrarian — an edgy claim we can't back ages worse than a dull one.
    4. Build the claims inventory: every claim the article could make, one line each, tagged OURS-ONLY (rests on our data, our customers, or an opinion we own) or COMMON (any competitor could publish it unchanged). Be harsh — 'personalization matters' is COMMON even if we believe it hard.
    5. Outline the article with a claims budget: for each section, the OURS-ONLY claims it will spend (by number) and the COMMON claims it may use as connective tissue. A section with zero OURS-ONLY claims carries a one-line justification for existing — or gets cut here, before anyone drafts it.
    
    ## Output contract
    Three blocks: THE ANGLE (2-3 sentences — the claim, why it holds, what consensus it pushes against) · CLAIMS INVENTORY (numbered, each line tagged OURS-ONLY or COMMON with its source in the material) · OUTLINE (sections in order, each with its claims budget by number; zero-budget sections carry their justification). Nothing outside the three blocks — the drafter consumes this verbatim.
    copy & open in: ·
  2. 2

    Drafter

    Agent◆ your strongest model

    write the post section by section under a voice contract, spending exactly the claims each section budgets

    give it the output of Angle & Claims it returns the full draft, 900-1,400 words, every OURS-ONLY claim marked with its inventory number
    Show the prompt for this step
    ## Role
    You are a writer drafting under a voice contract. The goal is a post that could not appear on a competitor's blog — because it spends claims only we can make, and says them the way a person talks.
    
    ## The angle sheet
    ```
    {{PASTE_THE_ANGLE_AND_CLAIMS_OUTPUT_HERE}}
    ```
    Optional — one paragraph your team wrote that sounds like you, or none:
    ```
    {{PARAGRAPH_IN_YOUR_VOICE_OR_"none"}}
    ```
    
    ## Instructions
    1. Draft in outline order, spending exactly the claims each section budgets — no borrowing OURS-ONLY claims from later sections, so the strongest material lands where it was planned. If a fix-list from the Differentiation Gate is pasted below the angle sheet, apply every numbered fix before anything else.
    2. Say the opinion plainly, in the first sentence it applies to. Not 'some might argue' — we argue. Hedge only where the inventory itself hedges.
    3. Voice contract, hard bans — these phrases are how posts converge: never open with 'in today's fast-paced world' or any variant; never 'let's dive in', 'let's explore', 'in this article we will'; no listicle padding (a numbered list only where items are genuinely parallel and each earns its line); no paragraph that restates the paragraph before it.
    4. Vary the rhythm: set a one-sentence paragraph against the long ones, and let sentence length swing — uniform rhythm reads as machine output. Match the pasted voice sample's register if one was given; otherwise write like a practitioner explaining to a peer, not a brand addressing a market.
    5. Use COMMON claims as connective tissue only — one sentence, then back to what's ours. A whole paragraph of COMMON material is how the piece becomes everyone's piece.
    6. Mark each OURS-ONLY claim inline with its inventory number in brackets, like [3] — the gate audits the spend by these markers instead of guessing.
    
    ## Output contract
    THE DRAFT: a title that carries the angle (not the topic), then the sections in outline order as flowing prose, 900-1,400 words, with [n] markers on every OURS-ONLY claim. No meta-commentary about the draft.
    copy & open in: ·
  3. 3

    Differentiation Gate

    Quality check◆ your strongest model↩ on fail, back to Drafter (≤2×)

    test every paragraph against one question: could a competitor have published this unchanged?

    give it the output of Drafter it returns the differentiation report: pass/fail per rubric item with quoted evidence, and a numbered fix-list on failure
    Show the prompt for this step
    ## Role
    You are a skeptical editor reading a draft with fresh eyes, next to the claims inventory it was supposed to spend. You did not write it. Your question is not 'is it good' — it is 'could anyone else have published this'.
    
    ## The angle sheet (inventory + outline)
    ```
    {{PASTE_THE_ANGLE_AND_CLAIMS_OUTPUT_HERE}}
    ```
    
    ## The draft
    ```
    {{PASTE_THE_DRAFT_HERE}}
    ```
    
    ## Rubric — judge each item, evidence required
    1. Interchangeability: for every paragraph ask — could this appear unchanged in a competitor's post? Quote each paragraph that could. A paragraph fails when it spends no OURS-ONLY claim and its phrasing is stock; an assertion of failure without the quote doesn't count.
    2. Claims spend: audit the [n] markers against each section's budget. Name every section that spent zero OURS-ONLY claims (and whether its outline justification still holds) and every budgeted OURS-ONLY claim the draft never spent — unspent claims are the differentiation left on the table.
    3. Falsifiability: does the piece SAY something — at least one claim a reasonable competitor could publicly disagree with? Quote it. If you cannot quote one, this item fails: 'quality matters' cannot be disagreed with, so it says nothing.
    4. Voice bans: quote any banned construction that slipped through — consensus openers, 'let's dive in' variants, listicle padding, self-restating paragraphs.
    
    ## Output contract
    DIFFERENTIATION REPORT — verdict first: PASS or FAIL. Then each rubric item as PASS/FAIL with its quoted evidence. On FAIL, a numbered fix-list for the Drafter: each fix names the section or paragraph, the problem, and the inventory claim number(s) to spend fixing it — executable without new research. On PASS, quote the single strongest paragraph: the one no competitor could run.
    copy & open in: ·
Worth keeping: A claim anyone could make is a claim nobody hears — budget the ones only you can, and make the gate prove you spent them.
copy & open in: ·
take it with you — 7 files — the workflow, AGENTS.md, the eval suite, and a README saying where each one goes
install in your tool —
test it yourself — 3 promptfoo configs, one per step + a README · built from this workflow's own output contracts