Content & SEO · CoreCMO

The framework — strategy first

Content & SEO — the strategic foundation.

STRATEGY & PROCESS

Content is [COMPANY NAME]'s most durable marketing asset. Our goal is to own the conversation around [PRIMARY TOPIC CLUSTER] across search, social, and syndication.

THE CONTINGENCY — WHEN THIS IS TRUE AND WHEN IT ISN'T

The blanket version of this advice is "content compounds, paid doesn't." That's right at $50K+ Annual Contract Value (ACV) with a 6+ month cycle — content authority stacks and paid wraps around it.

At sub-$15K ACV with a 60–90 day sales cycle, the math inverts. Search Engine Optimization (SEO) content takes 6–18 months to outrank an established competitor. Your buyer is in-market for 30 days. The Google Search query is the brand-discovery moment, and paid Search captures it before content has time to compound. Run the content engine — but fund it from paid-driven revenue, not the other way around. Treat content as the retention + expansion asset; treat Google Search ads as the acquisition channel until you cross ~$10M Annual Recurring Revenue (ARR).

Product-Led Growth (PLG) inverts again. Product-led companies should treat content as category education + activation help, not lead capture. Stripe's docs, Linear's blog, Notion's templates — distribution is the product, content makes the product easier to find and easier to use. The 4–6 SEO posts per month cadence below is the sales-led horizontal Software-as-a-Service (SaaS) pattern. Developer-tools and PLG companies should aim for fewer, deeper, technical pieces tied to product surface area.

Shift to the AEO mindset — but don't abandon SEO.

AEO VS SEO — THE 2026 REFRAME

Answer Engine Optimization (AEO) is the discipline of optimizing for AI answers — ChatGPT, Perplexity, Google AI Overviews, Claude. It's not a replacement for SEO. It's the new surface that sits above SEO, and it rewards different work. The senior-operator framing from the AEO discipline:

SEO MINDSET	AEO MINDSET
Win the click	Win the answer
Target short keywords	Target full buyer prompts / questions
Build authority via backlinks	Build consensus via citations across platforms
Penalized for duplicate content	Rewarded for consistent cross-platform content
Organic traffic as primary signal	LLM referral traffic as primary signal
Takes 6–12 months to see results	Can show measurable signal in weeks

The two-discipline rule: run AEO and SEO in parallel. AEO content (full-prompt long-form answers, cross-platform consistency, structured data, citation-friendly authority statements) feeds the AI inference layer. SEO (clusters, internal links, comparison pages, programmatic SEO) feeds the lingering human-search traffic. They draw on the same content but the optimization criteria differ. Plan accordingly.

Content Strategy

Your Content Pillars

Four pillars. Each one anchors a quarter of your content. Saves to your Brief — every content prompt on the site will know what topics you own.

Pillar 1

Pillar 2

Pillar 3

Pillar 4

Primary topic cluster (the conversation you're trying to own)

Saved as [CONTENT PILLARS] and [PRIMARY TOPIC CLUSTER].

Content Calendar Cadence

CONTENT TYPE	FREQUENCY	OWNER	PRIMARY DISTRIBUTION
Original research / benchmark report	2× per year	PMM + Demand Gen	Press, LinkedIn, email, gated landing page
SEO-optimized blog posts	4–6× per month	Content Marketing	Organic search, LinkedIn, email newsletter
Customer case studies	2× per month	Content + CS	Website, sales enablement, review sites
Video content	Bi-weekly	Content + Creative	LinkedIn, YouTube, website
Webinars / virtual events	2× per month	Demand Gen + PMM	Email, LinkedIn, content syndication
Product release content	Per release	PMM + Content	Email, in-product, social, website

SEO Strategy

Keyword Clusters (Priority)

Branded: '[COMPANY NAME]', '[COMPANY NAME] reviews', '[COMPANY NAME] pricing' — Protect and own.

Category: '[CATEGORY NAME] software', '[CATEGORY] platform', '[CATEGORY] tools' — Define the category.

Problem-based: '[PROBLEM YOUR PRODUCT SOLVES]', '[ALTERNATIVE PROBLEM DESCRIPTION]' — Capture intent.

Comparison: '[COMPANY NAME] vs [Competitor]', 'best [CATEGORY] for [VERTICAL]' — Intercept evaluation searches.

Emerging: '[NEW TECHNOLOGY] for [YOUR AUDIENCE]' — Own the emerging category early.

SEO Execution Standards

Every piece of owned content must have a target keyword, meta description, and internal linking plan before publishing.

Volume by motion (replaces the blanket "2 posts per week"): Sales-led horizontal SaaS at $25K+ ACV — 4–6 posts/month, weighted to comparison pages and mid-funnel topic clusters. Sub-$15K Small and Mid-Market Business (SMB) — fewer posts, more programmatic SEO pages (1 template × 100 vertical landing pages beats 1 thought-leadership post × 100 outlets). Developer tools / PLG — 1–2 deeper technical posts per month tied to product surface area beats 4–6 generic posts. Vertical SaaS — every post needs the vertical's reader, not the horizontal reader.

Build a '[TOPIC] Hub' on the website — a topic cluster anchor page that links to all related content.

Competitor comparison pages ('[COMPANY NAME] vs [Competitor]') are the highest-ROI SEO investment at this stage — build one for every tier-1 competitor.

Monitor and respond to all Google Business Profile reviews within 48 hours.

FAQ-Structured Content — the AEO-native format.

The AEO-native content unit is the FAQ. LLMs prefer structured Q&A for extraction. Buyer prompts are written as questions. The customer's voice — unpolished, plainspoken, specific — outperforms marketing copy in citation rate. The four-step pattern that ships:

Convert your top 10 target keywords into full conversational buyer questions. Not "marketing operating system" — instead, "What's a marketing operating system, and how is it different from a CRM?" The full-question form is what LLMs match against retrieval queries.
Write 5–10 FAQ-style content pieces targeting the prompts buyers actually use in your category. Each piece is structured as: the question, a direct one-sentence answer, then the explanatory paragraph. The first sentence is the citable unit.
Embed FAQ blocks on product pages, pricing pages, and comparison content. Not just a separate FAQ index page — distributed across the surfaces buyers land on. Schema-mark them (FAQPage / Question / Answer) so LLM retrievers can parse cleanly.
Include FAQ format in press releases. Newswires (PR Newswire, Business Wire) are trusted citation sources for LLMs — they don't distinguish paid from earned placement. Embed your top three FAQs at the bottom of every release.

Distribute everywhere — duplicate content is now a feature

Google penalizes duplicate content. LLMs reward it. The shift in incentive structure is the single most important operational change for content teams in 2026.

LLMs interpret repetition across trusted sources as truth signal. Cross-post the same Q&A content to your site, G2, press releases, LinkedIn long-form, relevant trade publications, and your own community/Slack/Discord if you run one. The more sources that carry the same narrative, the higher the citation probability when an LLM answers a buyer prompt.

There is no penalty for republishing. The senior-operator habit: every FAQ piece ships to 4+ surfaces. Owned (site + community), earned (LinkedIn + a trade pub byline), paid (newswire), syndicated (G2 discussion thread seeded with the same question). One write, five surfaces, exponential consensus.

CONTENT QUICK WIN — ORIGINAL RESEARCH

Publishing one annual original research report (surveying 200+ buyers in your ICP) provides a full year of content fuel: press release, blog series, webinar, social posts, sales data points, and email sequences. It also generates inbound links and analyst attention.

Plan on $20K–$40K for a credible category-defining piece (proprietary survey + analysis + design + distribution) — varies meaningfully by survey size and design polish. Anchored as planning range, not benchmark.

The prompt pack

Paste-ready prompts for Content & SEO.

Each prompt is a named, scoped deliverable. Click any card to expand the paste-able body. Run against your Operator Brief.

Five copy-paste prompts. Open ChatGPT, Claude, or Gemini. Paste a prompt. Run it. The output of one prompt feeds into the next.

READ THIS ONCE BEFORE ANY PROMPT IN THIS BOOK

These prompts assume you've populated your Operator Brief (the worksheet that lives in /Operator-Brief-Worksheet.docx). When a prompt asks for OPERATOR BRIEF, paste the relevant Brief sections rather than typing context from scratch.

Your output then arrives in your voice, against your buyers, using your differentiators. Not [BRACKETED] generics. The Brief is the difference between an LLM helper and a tool that sounds like you.

Prompt 1

Keyword cluster map

A 5-cluster keyword map: pillar topic → 8–12 supporting terms per cluster, with intent and difficulty.

Build a keyword cluster map for our SEO program. Five clusters anchored on five pillar topics. PRODUCT: [one-liner] TARGET BUYER: [persona] PILLAR TOPIC CANDIDATES: [list 8-10] Pick the five pillars with the highest commercial intent for our buyer. For each: 8-12 supporting keywords with (a) search intent (informational/ navigational/commercial/transactional), (b) difficulty bucket (low/med/high), (c) one-sentence why-it-matters for our buyer. Output as a markdown table. End with the single keyword we should target FIRST and rationale.

Prompt 2

Editorial calendar (90 days)

A 90-day editorial calendar with weekly cadence: 2 blog posts + 1 newsletter + 3 social posts.

Build a 90-day editorial calendar. PILLARS: [paste 5 pillars from Prompt 1] CADENCE: 2 blog/week + 1 newsletter/week + 3 social/week BUSINESS GOAL: [pipeline / brand / retention / expansion] For each week, output: - Blog 1 + Blog 2 (title, target keyword, format) - Newsletter (subject + 1-line angle) - 3 social posts (hook + format) Sequence pillars so each week reinforces the previous. Build to a campaign moment (gated asset, webinar, or product update) on weeks 4, 8, 12.

Prompt 3

Long-form blog brief

A SEO content brief: H2/H3 outline, target word count, internal links, citations needed, FAQ schema, CTA.

Produce a content brief for the keyword below. The brief goes to a writer — be specific. TARGET KEYWORD: [keyword] INTENT: [informational / commercial / transactional] PILLAR CLUSTER: [which pillar] WORD COUNT TARGET: [800/1500/2500] Output: 1. SERP analysis: top 3 ranking pages + what's missing 2. H1 + meta description options (3 each) 3. H2/H3 outline with 1-line description each 4. Internal links to add (3–5) 5. External citations needed (data, studies, named sources) 6. Featured-snippet target (which question) 7. FAQ schema questions (5) 8. CTA placement (mid-article + end)

Prompt 4

Internal linking plan

A linking matrix: which existing pages should link to the new piece, and which existing pages the new piece should link out to.

Plan the internal linking for a new article. NEW ARTICLE: [title + URL slug] PILLAR: [which pillar] EXISTING TOP-PERFORMING PAGES: [list 10 with URL + topic] Output two tables: 1. INBOUND — which existing pages should link IN to the new article, what anchor text, and where in the page the link should sit. 2. OUTBOUND — which existing pages the new article should link to, anchor text, and the section that justifies the link. End with the one orphan page that gets the biggest lift from this new piece.

Prompt 5

Content refresh audit

A prioritized list of 10 old posts ranked by refresh upside, with a 5-line refresh plan per post.

Audit our old content for SEO decay. Prioritize for refresh. POSTS TO REVIEW: [list 10–20 with URL, publish date, current ranking, monthly traffic] PILLAR ALIGNMENT: [our 5 pillars] For each post, score 1-5 on: - Drop in ranking last 6 months - Traffic potential at #1 ranking - Strategic pillar alignment - Lift effort (low/med/high) - Conversion track record Rank the top 10 by refresh ROI. For each, write a 5-line refresh plan: (1) what's outdated, (2) what to add, (3) new SERP target, (4) internal links to upgrade, (5) CTA refresh.

Prompt 6

FAQ-content production batch (AEO)

Convert top 10 buyer prompts into 10 FAQ-structured content pieces with cross-platform distribution checklist. The AEO-native production prompt — outputs ready for site, G2 discussion, LinkedIn long-form, newsletter, and newswire embed.

You are the Content Operations Agent for [COMPANY NAME]. Produce a batch of 10 FAQ-structured content pieces optimized for LLM citation (AEO), with a cross-platform distribution plan for each. Context: [PASTE OPERATOR BRIEF] Our top 10 target keywords: [LIST] Our named competitors: [LIST] Buyer prompts we currently lose (from AEO Baseline): [LIST or "unknown — run /aeo-baseline first"] For each of the 10 buyer prompts: 1. The full conversational buyer question (not a keyword — a sentence) 2. The one-sentence direct answer (this is what LLMs cite verbatim — get it right) 3. A 200-300 word explanatory paragraph in customer voice, not polished marketing copy. Use the exact terms our ICP uses in interviews. 4. 3 supporting bullet points with named evidence (specific data, named customer outcomes, citation-worthy stats) 5. FAQPage schema markup ready to paste into the page 6. A distribution plan with 4 surfaces: - Where on our site this lives (product page / pricing / comparison) - The G2 discussion question we seed that mirrors this prompt - The LinkedIn long-form version (200-300 words, native to the platform) - The newsletter / newswire FAQ block (3 questions, embedded at the bottom) Distribute deliberately. LLMs reward repetition across trusted sources (G2 + LinkedIn + your site + newswire = consensus signal). The same answer, worded the same way, across 4 surfaces, beats 4 different versions across 4 surfaces. Output as a single document with 10 sections, one per buyer prompt. At the end, list the suggested publication cadence (e.g., 2 per week for 5 weeks, distributed across each surface on the same day).

The agent spec

The agent for Content & SEO.

How to install this agent

Five steps from spec to running agent.

System prompt — copy the system prompt block below into your AI tool's system prompt field (Claude Project instructions, Cowork Skill instructions, custom GPT config, or your agent platform's equivalent).
Inputs — wire the inputs as the agent's reference files. The Operator Brief is always input #1; the other inputs vary by agent.
Outputs — the output schema tells you what the agent produces. Use it as a structured-output instruction in the system prompt, or as the format you expect to see back.
Evals — before publishing any output, score it against the eval criteria. Don't ship anything that doesn't pass.
Cadence — set the run cadence on your calendar (or your agent platform's scheduler). Log every run in your wins log.

Content Operations Agent

Runs the editorial calendar end-to-end across BOTH disciplines — SEO content for the human-search layer AND AEO/FAQ content for the LLM inference layer. Queues briefs, monitors SERP positions daily, tracks LLM citation share-of-voice quarterly, surfaces refresh candidates when rankings drop, maintains cluster-coverage AND consensus-source coverage. Keeps content compounding instead of decaying.

Who is this agent

Identity card

NameContent Operations Agent

RoleEditorial calendar + SEO operations — the content compounding layer

OwnerHead of Content

Reports toVP Marketing

Versionv0.5 (supervised)

SurfaceReplit + Postgres (SERP history, content corpus index)

Output target/editorial/queue.md, /editorial/refresh-queue.md, /content-status.md

Review cadenceDaily SERP sweep, weekly editorial review, quarterly cluster audit

Mission

Run the editorial calendar like an operating program, not a wish list, across BOTH the SEO discipline (human search) and the AEO discipline (LLM inference). Queue briefs the writer can pick up — both blog/cluster posts and FAQ-structured AEO content. Monitor SERP positions on the top 50 keywords daily. Track LLM citation share-of-voice (% of target buyer prompts where the company is cited across ChatGPT / Perplexity / Claude / Gemini) quarterly. Surface refresh candidates when rankings drop OR when an AEO Baseline run reveals a prompt the company is losing. Maintain the five-pillar cluster-coverage map for SEO AND the cross-platform consensus-source map for AEO (site + G2 + LinkedIn + newswire + Reddit when authentic). Keep content compounding instead of slowly decaying into a graveyard.

Goals & KPIs the agent moves

Leading indicators — the agent controls these

Brief-queue depth (briefs ready to write, not over-stuffed)3–4 weeks of writer capacity

Cluster coverage balance (max pillar vs. weakest pillar ratio)< 2× maintained quarterly

Lagging indicators — downstream outcomes with review triggers

Refresh precision (flagged refresh holds its ranking improvement for 14 days). Trigger: 2 consecutive quarters below 75% pages the Content Lead for refresh-rubric review.≥ 90%

Average organic traffic per published post at +90 days. Trigger: trend flat or declining for 2 consecutive quarters pages the Head of SEO and the VP Marketing for cluster-and-topic review.Trending up quarter-over-quarter

What it does

Task list

Daily Pull SERP positions for top 50 tracked keywords. Compare to prior day. Tag persistence (1-day vs. 14-day moves).
Daily Surface any post that dropped ≥ 3 positions and held the drop for ≥ 14 days. Add to refresh-queue with cited evidence.
Weekly Generate 2–3 new briefs for the writer. Prioritize the pillar with weakest cluster coverage. Each brief: target keyword, SERP analysis, recommended H2/H3 structure, internal-link plan, word-count target.
Weekly Generate 2 FAQ-structured pieces for AEO. Each one: full conversational buyer question, one-sentence direct answer (the LLM-citation unit), 200-300 word customer-voice explanation, FAQPage schema, distribution checklist (site + G2 discussion thread + LinkedIn long-form + newsletter/newswire FAQ block). Customer voice over polished marketing copy — LLMs cite plain language.
Weekly Cross-platform distribution audit. Every FAQ piece shipped this week should appear on minimum 3 surfaces by Friday: site + G2 (discussion seed or review reference) + LinkedIn long-form + optionally newsletter or newswire. Consensus comes from repetition.
Weekly Compile content-status.md — what shipped, what’s queued, what dropped in rankings, what climbed, what needs attention.
Weekly Audit recently published posts (T+30, T+60, T+90). Track organic traffic + conversion vs. forecast.
Monthly Pillar coverage audit: are all five clusters within 2× coverage of the weakest? Recommend rebalance briefs.
Monthly Internal-link audit. Every new post should link to ≥ 3 existing posts; every existing post should be linked to from ≥ 1 new piece per quarter.
Quarterly Cluster-strategy refresh. Are the five pillars still the right five? Surface evidence for adds / consolidations.
Quarterly AEO Baseline review. Run the /aeo-baseline tool against the top 20-30 buyer prompts. Compare to last quarter. Surface: (1) which prompts the company is winning with positive sentiment, (2) which it's losing and to whom, (3) which 2-3 prompts to target next quarter with new FAQ-content production.
Event When Win/Loss surfaces a new objection theme, brief a post that disarms it within 14 days — one SEO version, one FAQ version.
Event When the Brand & Positioning Agent updates voice rules, audit recent posts for voice drift; queue refreshes as needed.
Event When a competitor publishes new comparison content or runs a category-defining campaign, queue an "Alternatives to [competitor]" FAQ piece within 30 days — honest pro/con tables are highly cited by LLMs.

Schedule grid

Task	Frequency	Duration	Output goes to
Daily SERP sweep	Daily 04:00	~10 min	Head of Content + refresh-queue
Daily refresh-candidate surfacing	Daily 04:15	~5 min	Refresh-queue + Head of Content if material
Weekly brief generation	Weekly Mon 09:00	60–90 min	Writer + Head of Content (approval)
Weekly content-status compile	Weekly Fri 14:00	~30 min	Head of Content + VP Marketing
Weekly published-post audit	Weekly Fri 14:30	~30 min	Head of Content
Monthly pillar coverage audit	Monthly 1st	~60 min	Head of Content + VP Marketing
Monthly internal-link audit	Monthly 15th	~45 min	Head of Content + Web Operations Agent
Quarterly cluster refresh	Quarterly Q-1 days	~3 hours	Head of Content + Head of PMM + VP Marketing

Triggers

Scheduled (cron-style):

Schedule	What it runs
`0 4 * * *`	Daily SERP sweep + refresh surfacing
`0 9 * * 1`	Weekly brief generation
`0 14 * * 5`	Weekly status + audit
`0 9 1 * *`	Monthly pillar coverage audit
`0 9 15 * *`	Monthly internal-link audit

Event-driven:

Event	What it runs
Tracked keyword drops ≥ 5 positions in a single day	Flag for refresh-queue immediately; alert Head of Content
Win/Loss surfaces a new objection theme tagged Content	Generate a brief that disarms it within 14 days
Brand & Positioning updates voice rules	Audit last 30 days of published posts for voice drift within 7 days
Web Operations Agent ships a hero/pricing page copy variant	Audit related blog content for message-match within 7 days
Cluster pillar drops below 60% of the strongest pillar	Prioritize that pillar in next 4 briefs

Who it works with

Inputs

Source	Type	Cadence	Required?
Operator Brief (Sections 2, 3, 6, 8)	Markdown	Read every run	Required
Editorial calendar + 90-day plan	Markdown	Read every run	Required
Published posts corpus with frontmatter (title, slug, KW, publish date)	Markdown	Continuous	Required
SERP positions data (Ahrefs / SEMrush / Search Console API)	CSV / API	Daily	Required
Five-pillar cluster map	Markdown	Quarterly	Required — core config
Win/Loss Agent themes	Markdown	Per-interview	Required
Brand Voice Agent score history	Postgres	Weekly	Required — for voice-drift catch
Revenue Attribution Engine per-post conversion data	JSON	Weekly	Required for post-90-day evaluation

Outputs

Output	Format	Target path	Audience
/editorial/queue.md	Markdown	/editorial/queue.md (4 briefs max)	Writer + Head of Content
/editorial/refresh-queue.md	Markdown	Top 5 refresh candidates with 5-line plan each	Writer + Head of Content
/content-status.md	Markdown + Slack message	Weekly digest: published, queued, dropped, climbed	Head of Content + VP Marketing
Per-brief specification	Markdown	/editorial/briefs/<slug>.md	Writer (approval gate before publish)
Monthly pillar coverage report	Markdown + chart	/editorial/pillar-coverage/YYYY-MM.md	Head of Content + VP Marketing
Internal-link audit	Markdown table	/editorial/internal-links/YYYY-MM.md	Web Operations Agent + Head of Content

↑ Upstream — agents/sources that feed this one

Operator Brief (human-maintained). Voice rules + ICP + personas ground every brief.
Win/Loss Agent. Surfaces objection themes the content function should disarm.
Persona Researcher Agent. Persona profiles inform which dimensions need content coverage.
Market Intelligence Agent. Competitor content moves that should provoke response.
Brand & Positioning Agent. Voice + positioning rules that briefs must respect.
Revenue Attribution Engine. Per-post pipeline-trace confirms which content actually drove revenue.

↓ Downstream — agents/humans that consume its output

Writer (human). Picks up briefs from the queue. Drafts get reviewed via Brand Voice Agent before publish.
Head of Content (human). Approves briefs + refreshes; reviews weekly status.
Web Operations Agent. Receives internal-link recommendations + new-content alerts for landing-page integration.
Performance Marketing Agent. Receives newly published high-performing content for paid amplification consideration.
Brand Voice Agent. Scores every brief + every draft before publish.
Account Intel Hub. Receives content-engagement signals for the per-account record.

Human escalation paths

Trigger condition	Escalate to	Within
Refresh queue grows > 15 items	Head of Content + VP Marketing	Same week (capacity gap)
Brief approved without passing Brand Voice Agent scoring	Head of Brand + Head of Content	Before writer starts
Tracked KW drops > 10 positions sustained 21 days	Head of Content + VP Marketing	< 48 hours (likely algorithm hit)
Pillar coverage gap > 3× weakest-to-strongest	Head of Content + Head of PMM	Same week
Published post at T+90 has < 10% of forecast traffic	Head of Content	Add to refresh queue

How to build it

System prompt

You are the Content Operations Agent for [COMPANY]. YOUR JOB Run the editorial calendar like a program. Queue briefs the writer can pick up. Monitor SERP daily. Surface refreshes. Maintain cluster balance. Make content compound, not decay. INPUTS (always read in this order) 1. /operator-brief.md (Sections 2, 3, 6, 8) 2. /editorial/calendar.md (90-day plan) 3. /editorial/published/*.md (corpus) 4. /seo/positions.csv (daily SERP) 5. /seo/clusters.md (5-pillar map) OUTPUTS - /editorial/queue.md (4 briefs max - writer capacity) - /editorial/refresh-queue.md (top 5 candidates) - /content-status.md (weekly digest) - /editorial/briefs/<slug>.md (per-brief spec) RULES 1. Never queue more than 4 briefs per week (writer capacity). 2. A post enters refresh queue only if it drops 3+ positions for 14+ days (filter noise). 3. Brief queue prioritizes the pillar with weakest cluster coverage. 4. Refresh recommendations cite SERP data + traffic delta. 5. Every brief cites: target keyword, SERP top-5 analysis, recommended H2/H3 structure, internal-link plan, word-count target, conversion intent. 6. Voice rules from Brief Section 8 apply to every brief and every refresh plan. 7. Never invent a SERP position - cite the actual /seo/positions.csv entry. ESCALATION - Refresh queue >15: Head of Content + VPM same week. - KW drop >10 positions sustained 21 days: page Head of Content + VPM <48h.

Tools & integrations

Platform / tool	Used for	Required?
Replit + Postgres (SERP history)	Daily sweep + corpus state	Required
Ahrefs / SEMrush / Search Console API	SERP position tracking	Required
Google Analytics 4 + Search Console API	Per-post traffic + query data	Required
CMS (Webflow / WordPress / Contentful)	Reading published corpus	Required
Brand Voice Agent API	Pre-publish brief + draft scoring	Required
Slack API	Weekly status delivery + alerts	Required

Guardrails — what it must not do

Never queue more than 4 briefs per week. Writer capacity is the bottleneck; over-queuing creates backlog rot.
Never publish content without Brand Voice Agent scoring + Head of Content approval.
Never refresh a post based on a single day’s SERP volatility — require 14-day persistence.
Never invent traffic numbers, SERP positions, or keyword volumes — cite actual data sources.
Honor cluster discipline — don’t let pillars drift more than 2× from each other in coverage.
Never refresh a post without ROI projection — high-traffic + low-conversion may not be worth the work.
Never share keyword targeting outside the marketing function without VP Marketing approval — competitive intel.

Evals + hallucination defense

Evals — output quality checks:

Brief quality. Weekly: writer rates 3 briefs 1–5 on usefulness. Target average ≥ 4.0.
Refresh precision. Monthly: when refresh flagged, did the post actually persist below baseline? Target ≥ 90%.
Cluster balance. Monthly: pillar-coverage ratio strongest-to-weakest. Target < 2×.
Post-ship performance. Quarterly: % of published posts hitting ≥ 75% of forecast traffic at T+90. Target ≥ 60%.

Hallucination defense — specific checkpoints:

SERP positions must trace to the actual /seo/positions.csv entry with date stamp.
Traffic + conversion claims must come from GA4 / GSC export, never extrapolated.
Internal-link recommendations must verify the destination page exists + is currently published.
Keyword volumes must cite the tool source (Ahrefs / SEMrush) + date.
When data is missing, surface the gap rather than guess.

Maturity curve + first-run checklist

v0.1 — Manual-assistAgent generates briefs on-request. Head of Content runs SERP watch by hand. Useful from day 1.

v0.5 — SupervisedDaily SERP + weekly brief generation + weekly status + monthly audits autonomous. Head of Content approves every brief + every refresh. Default ship state.

v1.0 — Semi-autonomousAfter 90 days of clean evals + brief quality ≥ 4.0, agent auto-publishes refresh-queue without HoC gate (Head of Content still gates new briefs). Voice + cluster rules stay supervised.

First-run checklist — 5 steps from spec to running agent:

Author the five-pillar cluster map with Head of Content + Head of PMM.
Wire SERP + GA4 + GSC integrations. Verify daily data lands cleanly.
Run the agent in shadow mode for 2 weeks. Writer rates first 3 briefs daily.
Turn on the queue + refresh-queue. Subscribe Head of Content + VP Marketing to weekly status.
Schedule the quarterly cluster refresh + monthly audits. Log every run in /editorial/agent-log.md.