Saved a reusable Keyword Planner classification playbook and applied it to the prompt-builder 98-keyword export (0 build; absorb into prompt-template-builder). | Dev Logs

Date: 2026-09-01 20:38 Summary: Saved a reusable Keyword Planner classification playbook and applied it to the prompt-builder 98-keyword export (0 build; absorb into prompt-template-builder). Visibility: people

[question] 综合之前的对 keywords/prompt-builder 的keyword planner进行归类分析，并给出保存分析规则，后续直接沿用或迭代

[try to solve]

Context

docs/seo/keywords/prompt-builder/ held only a Google Keyword Planner CSV (98 rows, 12 ad groups). Prior theme batches (CIDR, magnets, text-diff/Compare-Text, social-share) already used a shared but undocumented workflow: volume buckets → drop noise → §3.3 H job types → harvest/absorb → optional SERP. The ask was to classify this CSV with that method and persist the rules so later themes reuse or iterate one playbook instead of reinventing steps.

Catalog already has /tools/prompt-template-builder: local Role/Task/Constraints/Output structuring, not an LLM prompt generator.

Process

1. Re-read Skill keyword-to-tool-funnel, long-tail strategy §3.3, and authoritative analyses for CIDR and text-diff as structural templates.
2. Parsed the prompt-builder CSV (BOM-aware): volume bands 3×(1k–10k), 16×(100–1k), 67×(10–100), 12×(0–10); ad groups are synonym clusters (Free/Online/Maker/Creator), not scenes.
3. Built job-type table (A structure template → existing slug; B LLM invent; C image/Midjourney; D writing/character; E random; F text→structure; G Copilot scene; H head synonyms).
4. Wrote durable playbook docs/seo/keyword-planner-analysis-rules.md and theme authority keywords/prompt-builder/2026-09-01-prompt-builder-keyword-planner.md.
5. Wired playbook into keywords README, Skill, ops §2.5/§2.6, funnel docs, strategy §3.3 F note, tracker + ~10 pool rows. No work-tasks/, no SERP claim of long_gap.

Root cause / analysis

Ad-group labels look like product lines but are doorway fuel if used to split URLs. The real split is controls: structuring fields (already shipped) vs calling a model vs image-parameter builders vs writing randomizers vs agencies. Head terms ai prompt generator / prompt maker (1k–10k) must not become H1; existing H1 Prompt template builder already differentiates. Without Bing/Google SERP, competition stays provisional head and verdicts stay absorb/defer/drop.

Solution

Verdict for the batch: 0 build. Absorb structured/builder/templates/create/maker·creator/text-to-structure/Copilot into prompt-template-builder. Drop services, brand betterprompt, “best/pro”. Defer Midjourney, writing dialogue/character, random generator until named + SERP.

How to reuse next theme: drop CSV under keywords/{theme}/ → tell the agent to follow keyword-planner-analysis-rules.md → expect authority md + ~10 pool rows + tracker line; iterate the playbook’s revision log when hard steps change.

Notes / boundaries

• No SERP this round → do not mark long_gap or week attack.
• Do not auto-open work-tasks/ for absorb-only findings.
• Later copy changes on prompt-template-builder should fill the Ads/Planner long-tail table in 02 (text + capability), still without chasing generator H1.

[actions]

• Added docs/seo/keyword-planner-analysis-rules.md
• Added docs/seo/keywords/prompt-builder/2026-09-01-prompt-builder-keyword-planner.md + README
• Updated keywords README, Skill, ops, funnel, strategy link, tracker, keyword-daily-pool.tsv, keyword-to-tool-funnel.mdc