BUILD auto+gate answer-ready-page

Every page answers the question it ranks for

The file

take it and run it.

One markdown file: the goal prompt, the verifier prompt, and the stopping condition. Read it before you run it. MIT licensed, no account, nothing to sign up for.

Download answer-ready-page.md View the raw file

curl -fsSL https://vibemarketers.ai/loops/answer-ready-page.md -o .claude/loops/answer-ready-page.md

The loop

copy it.

A page ranks for a question and then makes the reader hunt for the answer. AI assistants do the same thing readers do: if the answer is not near the top, they skip the page. This loop scores each page on how directly it answers its own question, then rewrites the opening.

Use it when you have pages that get traffic but do not get quoted, or a set of definitional or comparison pages where the answer is buried under throat-clearing.

You need the pages, in any format your agent can read: markdown, a CMS export, or HTML. No third-party tool or subscription.

You get a score per page before and after, and a rewritten opening that answers the question in the first sentence. Every claim, number and source stays exactly as it was. The loop is not allowed to change what a page says, only the order it says it in.

The gate is you reading the diff. It should be small: if the loop rewrote more than the opening, it exceeded its brief and you should reject it.

Goal prompt

For every page in <collection>, score the opening against the answer-first
rubric, rewrite the opening to answer the page's question in the first
sentence, keep every claim and source unchanged, and record before and
after scores in the receipt.

Verifier prompt

Diff every rewritten page. FAIL if any number, claim, or source changed, if
any opening still starts with a topic instead of an answer, or if the
receipt lacks before and after scores with the page count. Otherwise PASS.

Stopping condition

Every page in the collection has a before and after score in the receipt and the verifier marks PASS.

MCPs it needs

none required

Want everything at once

The installer writes the three starter skills and every published loop into .claude/. Read the script first.

curl -fsSL https://vibemarketers.ai/install.sh | bash

Or have us run it on your site →

The receipt

Receipt answer-ready-page 2026-09-21

Baseline: 13 term pages scored against the answer-first rubric. Mean 9.69 of 10, with 11 of 13 already perfect. Two pages lost points for a first sentence over 40 words: what-is-a-marketing-agent at 48 and what-is-an-agent-loop at 42.
Hypothesis: Splitting an overlong opening sentence after its definition raises the answer-first score without changing a single claim, because the definition is already in the first clause.
Intervention: Scored all 13 pages with a deterministic rubric, then rewrote 2 openings by replacing one colon with a period and capitalising the next word. No other page was touched.
Result: Mean answer-first score 9.69 before and 10.0 after across the same 13 pages, 13 of 13 now perfect. Diff is 2 characters across 2 files, with 0 words added and 0 words removed. No number, claim or source URL changed.
Commercial signal: none
Attribution limits: The score is the loop's own rubric, not a reader outcome. Nothing here shows that a reader or an answer engine prefers the rewritten openings. Citation movement cannot be read at all: no engine is configured, so there is no baseline to compare against.
Learning: The first rubric counted only copula verbs as assertions and wrongly scored 3 pages as having no assertion, when they open with runs, splits and executes. Rewriting to satisfy that scorer would have made 3 good pages worse. The rubric was fixed before any page was edited, which moved the true baseline from 8.77 to 9.69.
Next intervention: Add a length floor to the rubric. A stub page can score 10 while saying nothing, so answer-first should not pass a page that has no sourced example in it.

0 human touches

9 minutes

0 errors

raw run logopen →

Sources: run log, lab/runs/2026-09-21-answer-ready-page.json, collected 2026-09-21 , window: same 13 pages, scored before and after one run .

One entry per run, append-only, newest first. Corrections append a dated line.

Inside a company

where the lane changes.

VibeMarketers tests the loop. SuperMarketers runs it on a real team.

Problem: Pages rank for a question and then make the reader hunt for the answer.
Source of truth: The CMS export plus the Brain's buyer-questions file.
Permitted actions: Score pages, draft rewrites of the opening paragraph, open a PR. Never change a claim or a number.
Required approvals: Content owner approves each PR before publish.
Owner: Content lead
Metric: Answer-first score per page, and citations on the fixed query set per page.

On a real team the rewrite lands in the CMS as a draft with a diff, and the gate is the content owner, not a repo PR.

Running this on your own site is one person reading a diff. Running it inside a company changes four things, and the guide covers them. When it stops being a file you copy and becomes a build, SuperMarketers does the same loop model inside your stack, on your keys.