SEVOsmith v2.2.0: n8n Content Creation Automation for Multi-Site Publishing

I ran SEVOsmith on “DataForSEO review” on a Wednesday morning. Thirty-six minutes later: a 4,800-word DataForSEO review was live in WordPress — six images attached, Rank Math metadata injected, pricing tables formatted, and a full audit trail in my inbox showing every source the research agent pulled. I didn’t touch the keyboard once.

The day after, I queued “best AI content writer tools” — a competitive listicle covering 16 tools. SEVOsmith switched to listicle format automatically, pulled live pricing data for each tool, ran the audit loop, and returned a 5,600-word draft with feature comparison tables already formatted. Forty-one minutes. Same workflow, no changes.

That’s the baseline. Most AI writing tools have one setting: generate. Give them a keyword, get back 2,000 words with no audit trail, no quality gate, no verification that the stat in paragraph 3 came from a real source. SEVOsmith is the opposite of that — a 6-workflow production system you own outright, running on your own n8n instance, with a full paper trail on every decision the agents made. Here’s exactly how it works.

Who Is SEVOsmith v2.2.0 For?

SEVOsmith is designed for operators — people running content production as a repeatable system, not a one-off task. If you self-host n8n and manage more than one site, this was built for you.

SEVOsmith is built for:

• Content agencies running 20–200 articles/month across multiple client sites — DataTable credential management means one workflow instance serves every client, every site
• SEO-focused brands that need consistent, audited quality at volume without a 10-person writing team
• n8n developers who want to understand, extend, and build on a production-grade content automation architecture — every sub-workflow is independently editable
• Publishers targeting AI citation — the GEO layer, stat+consequence pairings, and brand proximity rules are designed specifically for ChatGPT, Perplexity, and Google AI Overview inclusion
• Multi-language content operations — set target_language in Project_Setting and every agent outputs in that language. Active in production for English and Vietnamese.

Not the right fit for: users who want a one-click SaaS tool (this requires API credential setup and n8n configuration), or single-article use cases where setup overhead outweighs the value.

What You Get After Every Run

A completed run produces four deliverables. Here’s exactly what each one contains.

Research & Intelligence Report

Stored to Cloudflare R2 and your Google Drive folder. URL included in the completion email. Contains:

• Keyword metrics: volume, difficulty, CPC, search intent classification
• SERP analysis: Page 1 results, featured snippets, full SERP feature inventory
• YouTube SERP: video competition signals — informs whether video embeds strengthen the article
• AI Overview presence signals via Google AI Mode and AI Keyword Data endpoints
• Competitor landscape: gap analysis, negative space, battle plan sourced from Perplexity sonar-pro
• Audience profile: knowledge level, pain points, decision criteria
• E-E-A-T source list: curated citations, tier classification (primary / secondary / supporting), CRAAP test scores
• Content cluster map: topic cluster and subtopic structure
• Internal link map: existing site pages matched to anchor opportunities (via Firecrawl sitemap scan)
• Content blueprint: UVP angle, SEvO strategy, brand proximity plan

Article & Audit Trail Report

Also stored to Cloudflare R2 and Google Drive. This is the compliance record — not a summary, a forensic log:

• Quality score breakdown across 5 categories: Content Architecture, SEO Compliance, E-E-A-T & Citations, GEO & AI Citability, Anti-AI Detection
• Forensic Accountant raw metrics: exact keyword counts, URL validation results, paragraph length distribution, stat density per 100 words
• Issue log with codes: each compliance issue listed with severity, location, and fix applied
• Human review suggestions: edge cases the system flagged but left for editorial judgment — distinct from auto-applied fixes
• Content lifecycle assessment: freshness classification, update triggers, and evergreen viability rating

Published CMS Draft

WordPress: post created as draft, featured and section images attached to the media library, RankMath SEO title / meta description / slug injected via REST API post-creation.

Shopify: article published in draft status, Gemini post-processing step applied to convert the article HTML to Shopify-compatible format before publishing.

A standalone HTML article file is also saved to Cloudflare R2 on every run — independent of the CMS. Useful if you need the article outside the CMS or want to feed it into another workflow.

Completion Email

Sent to the configured notification address in Project_Setting. Contains: article title, meta description, CMS draft link, research report link, audit trail link, quality score, word count, execution time, project name, and keyword. Everything you need without opening the CMS.

Every run also logs automatically to the Article_Report tab in your Content Planner — one row per article across every project. Your cumulative content production ledger.

What Is in the Full System — 6 Workflows, 467 Nodes, 12 Agents?

SEVOsmith v2.2.0 ships as 6 workflow files. The main orchestrator calls 5 independent sub-workflows — each with one job, each independently maintainable, each callable from any workflow you build on top.

YOUR GOOGLE SHEET  (Manager tab — one row per article)
         │  Run Now = TRUE  or  Scheduled DateTime
         ▼
AI-Citable Content Engine v2.2.0  (194 nodes · 12 agents)
    │
    ├──▶  KW Research Engine v2.2.0   (126 nodes · 10 <a href="https://nextgrowth.ai/dataforseo-review/">DataForSEO</a> endpoints)
    │         └── all results cached in Cloudflare R2 per keyword
    │
    ├──▶  Tools v2.2.0               (33 nodes · ~33 calls per run)
    │         └── cache check / store / run logging for every agent stage
    │
    ├──▶  Image Generator v2.2.0     (29 nodes · Gemini Imagen API)
    │         └── fails safely — article still publishes if this fails
    │
    ├──▶  WordPress Publisher v2.2.0  (34 nodes)
    │         └── DataTable auth · images · RankMath meta injection
    │
    └──▶  Shopify Publisher v2.2.0   (28 nodes)
              └── DataTable auth · Shopify-compatible HTML post-processing
    │
    ▼
Published Draft + Research Report + Audit Trail + Completion Email

YOUR GOOGLE SHEET  (Manager tab — one row per article)
         │  Run Now = TRUE  or  Scheduled DateTime
         ▼
AI-Citable Content Engine v2.2.0  (194 nodes · 12 agents)
    │
    ├──▶  KW Research Engine v2.2.0   (126 nodes · 10 DataForSEO endpoints)
    │         └── all results cached in Cloudflare R2 per keyword
    │
    ├──▶  Tools v2.2.0               (33 nodes · ~33 calls per run)
    │         └── cache check / store / run logging for every agent stage
    │
    ├──▶  Image Generator v2.2.0     (29 nodes · Gemini Imagen API)
    │         └── fails safely — article still publishes if this fails
    │
    ├──▶  WordPress Publisher v2.2.0  (34 nodes)
    │         └── DataTable auth · images · RankMath meta injection
    │
    └──▶  Shopify Publisher v2.2.0   (28 nodes)
              └── DataTable auth · Shopify-compatible HTML post-processing
    │
    ▼
Published Draft + Research Report + Audit Trail + Completion Email

Technology Stack

Seven external services, coordinated by n8n. Each is swappable independently — switch LLM providers, data sources, or storage without touching workflow logic.

What Does Each SEVOsmith AI Agent for Content Creation Do?

12 agents. Each specialized. Each with a distinct system prompt, a defined input format, and a verifiable output contract. The Forensic Accountant is not an LLM — it’s deterministic code that counts, measures, and validates. Quality scores you can reproduce.

Why a code node for the audit? LLMs cannot count reliably. Ask one how many times a keyword appears in a 4,000-word article and it estimates. The Forensic Accountant counts — exactly. Keyword density, stat density, paragraph length distribution, anchor text length — measured deterministically. That’s why the quality score is reproducible, not approximate.

The Article Reviser has full writer authority. It receives the original production brief and rewrites entire flagged sections from scratch when the Compliance Auditor flags structural issues. This is not a copy editor making surface-level tweaks. If a section fails stat density or uses a prohibited opener pattern, the Reviser generates new content.

What Content Quality & SEO Capabilities Are Built Into Every Agent?

Most AI content tools are a black box. SEVOsmith’s agents have documented, verifiable SEO logic in every system prompt — enforcement rules with audit codes, not suggestions.

9 Content Formats with Format-Specific Intelligence

Set the content format in the Manager tab per keyword row. All agents — outline, writing, audit, revision — adapt automatically. Each format has its own structural rules, word count ranges, required elements, and prohibited patterns.

E-E-A-T Universal Mandate

Post-December 2025 Core Update: E-E-A-T applies to all competitive content, not just YMYL. SEVOsmith enforces this unconditionally across every content format:

• Minimum 3 external citations for all competitive content
• Per-format Experience Signal Minimums — each format has documented, audited requirements
• Elite E-E-A-T Curation Agent: three-tier citation system, CRAAP test criteria, domain extension prioritization

Anti-AI Detection — Structural Variation Directive

Both the Content Writer and Compliance Auditor enforce these rules independently — a failed check at writing gets caught again at audit:

• Varied list lengths across sections — no uniform bullet counts throughout
• Mixed paragraph lengths — no consistent 3-sentence paragraph pattern
• Varied opener styles — sections cannot start the same way consecutively
• Rhetorical variety — questions, statements, imperatives mixed across sections

Technical SEO Built Into the Audit

The Strategic Compliance Auditor checks these on every article, every run:

• Anchor text length: 2–8 words (single-word brand names excepted)
• Stat density: minimum 1 sourced statistic per 100–150 words
• Mobile heading length: H2 under 50 characters, H3 under 40 characters
• PAA (People Also Ask) distribution across body sections, not concentrated only in FAQ
• Search intent cross-check: article structure must match detected intent classification
• Format-conditional freshness policy: “we update this annually” language is a violation in evergreen formats
• Helpful Content Self-Assessment: 4-item checklist on every run

How Does the AI Citation & GEO Layer Work?

GEO — Generative Engine Optimization — is a distinct layer, not a footnote to traditional SEO. The Master Content & SEvO Strategist’s entire job is structuring content to be cited by AI systems. This is designed into the article before the Content Writer generates a single word.

What’s built into agent prompts specifically for AI citation readiness:

• Stat + consequence pairings: every claim structured as “[stat] which means [consequence]” — the format AI systems prefer when pulling citations into generated answers
• Brand proximity rule: brand name mentioned within 2–3 H2s and in the Key Takeaway block — ensures the brand is cited alongside the insight, not just buried in the intro
• Citable body passage per H2: one quotable statement per section, under 30 words, stat + insight format — extraction-ready for AI answer generation
• Competitor SERP Feature Audit: maps which competitors own which featured snippets for the target keyword
• AI Overview Defensive Strategy: identifies AI Overview triggers and structures content to compete for inclusion or defend against displacement

The Intelligence Layer — KW Research Engine v2.2.0

Every article starts here. The KW Research Engine is called once per run. It hits 10 DataForSEO endpoints, normalizes everything into a single keyword_research_orchestrator JSON object, and caches the result in Cloudflare R2. Every downstream agent reads from this object — no agent calls DataForSEO directly.

1. Keyword Overview — volume, difficulty, CPC baseline
2. Keyword Ideas — semantic expansion of the primary term
3. Related Keywords — SERP co-occurrence signals
4. Keyword Suggestions — intent-adjacent terms
5. Search Intent — informational / commercial / transactional classification
6. AI Keyword Data — AI Overview presence signals for this term
7. HTTP Subtopics — topic cluster map for the keyword
8. SERP Google — live Page 1 results, featured snippets, full SERP feature inventory
9. SERP YouTube — video competition signals (informs whether to add video embeds)
10. AI Mode Data — Google AI Mode presence signals

The Cloudflare R2 Cache Layer

Every DataForSEO result is cached per keyword. Same keyword on second run: all 10 API calls skipped — cached results returned in seconds. Running 50 articles on the same topic cluster doesn’t mean paying for 50 identical keyword research calls. The first run pays. Every subsequent run on cached keywords costs near zero for that stage. For setup instructions, see the Cloudflare R2 + n8n integration guide.

• Cache key: keyword + config hash
• TTL: configurable per project in Cache_Settings tab (default 30 days)
• KW_Force_Refresh flag: override per Manager tab row when you need live SERP data

The Infrastructure Layer — Tools Sub-Workflow v2.2.0

The sub-workflow that runs more than anything else. ~33 calls per article run — more than all other sub-workflows combined. Three operations: cache check (S3 read), parse and store (S3 write), run logging. Set up n8n workflow monitoring to catch failures early.

It exists as a dedicated sub-workflow rather than inline code for one reason: one fix propagates to all 33+ callers. One log format. One cache TTL logic. One place to change behavior for the entire pipeline. A cache hit on any agent stage means that agent is skipped entirely — its cached output returned and passed forward. On a warm cache, a first-run 35-minute pipeline completes in under 10 minutes.

How Does Multi-Site Publishing Work for WordPress and Shopify?

SEVOsmith ends with a published CMS draft — images attached, metadata injected, correct slug set. Two dedicated publisher sub-workflows, each targeting one CMS. Both use n8n DataTables so credential management scales without workflow copies.

In practice: we run SEVOsmith across three client sites from one n8n instance. Adding a fourth took 90 seconds — one new row in WordPress_Credentials, one new project row in Project_Setting. No workflow edits, no credential re-wiring, no duplicated workflows.

Multi-Site Management via DataTables

n8n DataTables are built-in database tables inside your n8n instance. Each site is one row. No credential per workflow copy, no duplicate workflows per client.

WordPress_Credentials table: project_name, wp_api_url, wp_username, wp_app_password, wp_author_id. Add a new WordPress site: add one row. Zero workflow edits. The publisher looks up credentials at runtime from the project_name in the Manager tab row that triggered the run.

Same pattern for Shopify: Shopify_Shop_Credentials — store_name, my_shopify_unique_domain_name, my_shopify_admin_token, my_shopify_unique_api, author_name. Ten Shopify stores = ten rows, one workflow.

WordPress Publisher v2.2.0

• Creates post as draft via WordPress REST API
• Uploads featured and section images to WordPress media library (or embeds R2 URLs directly)
• Injects SEO title and meta description via RankMath REST API post-creation
• Sets correct slug from Metadata Architect output

Shopify Publisher v2.2.0

• Publishes to Shopify blog via Admin API (draft status)
• Gemini post-processing converts the article HTML to Shopify-compatible format before publishing
• Same DataTable auth pattern — credentials looked up at runtime from Shopify_Shop_Credentials

The Image Generator v2.2.0

Controlled entirely from Project_Setting — two Yes/No flags: Auto Generate Featured Images and Auto Generate Images H2 (one image per H2 section). Zero workflow edits to enable or disable.

Calls the Gemini Imagen API after the Content Writer completes. Two storage modes: upload to WordPress media library, or upload to Cloudflare R2 and embed the URL in the article HTML. Because Image Generator runs as a separate sub-workflow, an API failure there does not stop the article from publishing — the publisher continues with images omitted and the failure is logged.

The Google Sheet as Command Center

The Google Sheet isn’t just a trigger — it’s the operating system. Multi-project management, content refresh, scheduling, and per-stage cache control all live here. Nothing requires workflow edits for normal operations.

Manager Tab — The Article Queue

One row = one article run. Key columns: Project, Keyword, Word Count, Content Format, Target Audience Knowledge Level, Additional Instruction, Run Now, use_cache, Old Post URL, Scheduled DateTime.

• Run Now = TRUE triggers immediately; Scheduled DateTime triggers at an exact time
• use_cache = FALSE on any row forces a full fresh run for that article, ignoring all cached agent outputs
• Old Post URL — paste a URL to an existing article and the Content Writer switches to refresh mode: it reads the existing article and rewrites or extends it rather than generating from scratch. Same audit and publishing pipeline runs on the result.

Project_Setting Tab — Per-Project Configuration

Each unique Project value = one project. Add a row: project name, target domain, S3 bucket, CMS Platform (WordPress or Shopify), language, image generation flags, notification email. That project is live on the next trigger. No workflow edits.

Cache_Settings Tab — Per-Stage TTL Overrides

Set any agent stage TTL to 0 to force re-run on next trigger. Right way to iterate on agent prompts: set the target stage to TTL 0, trigger a run, review the output, restore the TTL. All other stages run from cache — you pay only for the stage you’re testing.

The n8n Content Creation Automation Workflow — From Keyword to Draft

The complete content creation workflow runs 14 steps across 8 execution zones — from a Google Sheet trigger to a published CMS draft with images and metadata. Every stage is logged. Every agent output is cached. Every failure retries only the stage that failed.

See It in Action — Real Runs, Real Output

Below are real runs. A keyword goes in; a published article comes out. The screenshots show the KW research report, the compliance audit log, and the completion email — with quality score, word count, and execution time. Each linked article was written entirely by SEVOsmith v2.2.0, from brief to published draft.

Already Using v1.5.0? Here’s What Changed

SEVOsmith v2.2.0 is a full architecture rebuild — not an update to import over your existing setup. Here’s what drove the rebuild and what the new version ships.

Three problems surfaced consistently in v1.5.0 user feedback. First: when any zone failed inside the monolithic workflow, the entire run was lost — restart meant paying for every API call from the beginning. Second: the single-workflow design made safe customization nearly impossible — editing one section risked breaking something three zones downstream. Third: every run called all 10 DataForSEO endpoints regardless of whether the keyword had been researched days ago.

v2.2.0 was designed around all three. Every major change maps to a user report:

By the Numbers

Full specification comparison between the two versions.

See the complete v1.5.0 system breakdown for a detailed look at what the original architecture included.

Frequently Asked Questions

What exactly is SEVOsmith — is it a SaaS tool or something I own?

SEVOsmith is a deployable n8n automation — a set of workflow files you import into your own n8n instance (self-hosted or n8n Cloud). You own it. You configure it. It runs on your infrastructure with your API credentials. There are no usage fees, no per-article charges, and no vendor dependency beyond the underlying APIs (DataForSEO, Gemini, OpenRouter, Cloudflare R2). One-time purchase. Perpetual license.

Do I need coding experience to use SEVOsmith?

You need basic n8n familiarity — enough to import workflow files, connect credentials, and configure nodes. The Setup Wizard walks through all 9 API credentials in sequence. If you’ve used n8n for other automations, setup is straightforward. If you’d rather skip the configuration entirely, the Done For You option (below) handles every step and hands you a running system.

How many API credentials do I need to set up?

Nine: n8n (self-hosted or Cloud), Google Sheets (OAuth), Google Drive (OAuth), DataForSEO, Firecrawl, Gemini (Google AI Studio), OpenRouter (for Claude Sonnet), Cloudflare R2, and your CMS (WordPress Application Password or Shopify access token). The Setup Wizard walks through all nine in sequence. See the complete setup guide before purchasing to verify you have access to all required services.

Can I run multiple client sites from one SEVOsmith instance?

Yes. Each row in Project_Setting is one project. Each row in WordPress_Credentials or Shopify_Shop_Credentials is one site. Add a project name, domain, CMS platform, and language — that project is live on the next trigger. The Manager tab routes each article run to the correct credentials and Project_Setting row automatically. No workflow edits per site.

Do I pay for keyword research on every article run?

Not on repeat runs. Every DataForSEO call and every agent output is cached in Cloudflare R2. Second run on the same keyword: all 10 research API calls are skipped — cached results returned in seconds. A full cache hit turns a 35-minute run into under 10 minutes at near-zero API cost. The KW_Force_Refresh flag overrides this per row when you need live SERP data.

What if I already have a published article I want to update?

Paste the URL into the Old Post URL column in the Manager tab. The Content Writer switches to refresh mode — it reads the existing article and rewrites or extends it rather than generating from scratch. The same Forensic Accountant audit, Compliance Auditor review, Reviser pass, and publishing pipeline runs on the result.

Does SEVOsmith support languages other than English?

Yes. Set target_language in Project_Setting. Every agent — headlines, body copy, metadata, citations — outputs in that language. Active in production for English and Vietnamese. Any language supported by Gemini and Claude Sonnet via OpenRouter is available. Different projects can run in different languages from the same n8n instance.

Can I use this without WordPress or Shopify?

Yes. A standalone HTML article file is saved to Cloudflare R2 on every run, independently of the CMS publisher. Pull it from R2 and publish manually to any CMS or feed it into any other workflow. The article, research report, and audit trail are all accessible via R2 links in the completion email regardless of which publisher ran.

What does the quality score actually measure?

Five categories: Content Architecture (structure, heading hierarchy, section balance), SEO Compliance (keyword placement, anchor text, meta readiness), E-E-A-T & Citations (source tier, experience signals, citation density), GEO & AI Citability (stat+consequence pairings, quotable passages, brand proximity), and Anti-AI Detection (paragraph variation, list length diversity, opener style variety). The Forensic Accountant produces the raw metrics deterministically; the Compliance Auditor interprets policy compliance. Production runs score 95–96 / 100. The audit trail report documents every check, its result, and any fix applied — you can see exactly why the score is what it is.

How is v2.2.0 different from v1.5.0?

Three core architectural differences. v1.5.0 was one monolithic workflow — one failure meant restarting everything from the beginning. v2.2.0 has 5 independent sub-workflows; when one fails, only that sub-workflow retries. v1.5.0 had no caching — every run paid for all 10 DataForSEO endpoints and re-ran every agent. v2.2.0 caches at every stage in Cloudflare R2. v1.5.0 published only to WordPress and required a separate workflow copy per client site. v2.2.0 supports WordPress and Shopify with unlimited sites from one workflow via DataTable rows. Full comparison in the “Already Using v1.5.0?” section above.

Build Your Content Engine — Two Ways to Start

The full SEVOsmith v2.2.0 bundle includes: AI-Citable Content Engine v2.2.0, all 5 sub-workflows (KW Research Engine, Tools, Image Generator, WordPress Publisher, Shopify Publisher), Setup Wizard, and the SEVOsmith Content Planner Google Sheet template. One-time purchase. Perpetual license.

Don’t want to configure 467 nodes yourself? We set up the entire system on your n8n instance — all 9 API credentials, DataTables, Google Sheet, Apps Script trigger, and first test run — and hand it over ready to produce articles on day one. Starting at $200. No subscription, no ongoing cost. You focus on the keywords.

Your competitors are running monolithic prompts and calling it a content strategy. This is what the next tier looks like.