The AI Content Intelligence Platform.
Data-driven topic selection. Voice cloning. Universal publishing.
An end-to-end AI content intelligence platform that knows what to write, when to write it, who it should sound like, and where to publish it.
TBK-IQ is not an article generator. It is a 6-layer platform that combines real-time performance data (GSC, GA4, Semrush, Ahrefs), competitive intelligence, voice cloning, multi-agent article generation, and universal CMS publishing into a single pipeline. The user provides a category. TBK-IQ returns data-backed article recommendations, generates content in the voice of the client's existing writers, and publishes it directly to any CMS.
SRMG, Asharg, and media groups producing 2,000+ articles/week. Manual SEO workflows can't scale — TBK-IQ automates the intelligence layer.
Content teams managing 10+ client blogs. TBK-IQ replaces the spreadsheet-and-gut-instinct approach with data-driven topic selection per client.
Product-aware content generation. TBK-IQ writes blog posts that reference the store's catalog, driving organic traffic to product pages.
WordPress, Ghost, Webflow, Contentful, Strapi, custom CMS, or plain HTML. If it publishes content, TBK-IQ works with it. Zero lock-in.
Each layer feeds the next. Data flows top to bottom. Every article is backed by real performance data, matched to a human voice, and published directly to the client's CMS.
Every decision TBK-IQ makes is backed by real performance data. This layer collects, normalizes, and stores it.
| Source | Data Pulled | Auth Method | Refresh Cycle |
|---|---|---|---|
| Google Search Console | Clicks, impressions, CTR, avg position per page per query; index coverage; crawl stats | OAuth 2.0 | Daily |
| Google Analytics 4 | Sessions, engagement rate, scroll depth, time on page, conversions per URL | OAuth 2.0 | Daily |
| Semrush API | Keyword research, keyword gap, domain analytics, topic research, SERP features | API Key | Weekly |
| Ahrefs API | Backlink profile, domain rating, referring domains, top pages by traffic, keyword explorer | API Token | Weekly |
| Google Trends | Seasonal interest curves, trending queries, regional demand signals | Public / Pytrends | Weekly |
| CMS Crawler | Every article URL, title, word count, publish date, author, categories, tags | CMS API / HTTP | On-demand |
All sources are normalized into a unified Content Performance Record per URL:
The brain. Analyzes performance data to recommend exactly what to write, when, and why.
User provides a category. The agent returns a ranked list of specific articles to write, each scored and justified with data.
User provides a specific keyword directly — for agent override to accommodate priorities or preferences beyond the data.
Mode B runs 6 analyses in sequence to produce scored recommendations:
| # | Analysis | What It Does | Data Source |
|---|---|---|---|
| 1 | Content Inventory | Crawls the site, builds a map of every article with metadata | CMS API + HTTP |
| 2 | Decay Detection | Flags articles with 3+ months of declining impressions or 20%+ single-month drop | GSC + GA4 |
| 3 | Gap Analysis | Finds high-volume keywords the client doesn't rank for but competitors do | Semrush + Ahrefs |
| 4 | Seasonality Check | Adjusts opportunity scores based on seasonal demand curves (write before the peak, not during) | Google Trends |
| 5 | Saturation Index | Scores SERP difficulty: are top results thin/outdated (high opportunity) or deep/authoritative (low)? | Semrush SERP |
| 6 | Cannibalization Guard | Checks if client already ranks for a keyword — recommends refresh instead of new article | GSC + Content Map |
Detect existing writers, clone their style, produce content that's indistinguishable from their own work.
Instead of "write like a human," TBK-IQ says "write like THIS specific human who already writes for this publication." The Voice Analyzer Agent runs a 4-step process on the client's existing content.
Crawl the client's site and collect 50–100 articles minimum (statistical significance threshold). Extract raw text, strip HTML, preserve paragraph structure. Associate each with author name where available.
Run stylometric analysis on each article to classify as HUMAN HYBRID or AI-GENERATED. Signals measured:
| Signal | Human Pattern | AI Pattern |
|---|---|---|
| Sentence length variance | High (mix of short punchy + long complex) | Low (uniform 15–20 word sentences) |
| Vocabulary richness (TTR) | Higher — uses unexpected words | Lower — defaults to common synonyms |
| Hedging frequency | Occasional, deliberate | Excessive ("it's worth noting", "it's important to") |
| Cliche density | Low — personal expressions | High — "delve", "landscape", "leverage" |
| Em-dash frequency | Varies by author | Unusually high (AI signature) |
| Paragraph rhythm | Irregular, driven by thought flow | Uniform 3–4 sentence blocks |
Using only the HUMAN articles, cluster by writing style fingerprint using NLP features: sentence structure, tone, formality, vocabulary, use of analogies, humor, technical depth. HDBSCAN clustering finds natural writer groups — no need to predefine the number of writers.
This persona is injected into the draft-writer agent as a style constraint. The output reads like the client's own writer wrote it.
7 AI agents orchestrated in sequence. Each one is specialized, autonomous, and produces structured output for the next.
| # | Agent | Input | Output | Tools Used |
|---|---|---|---|---|
| 1 | Topic Recommender | Category + client site URL | Ranked list of article recommendations with scores and reasoning | Layers 1–2 data |
| 2 | Voice Analyzer | Client site URL | Writer personas (DNA profiles) for style matching | HTTP crawler, NLP |
| 3 | Project Analyzer | Target project directory or URL | Shell detection, design tokens, component inventory, tech stack | File system, HTTP |
| 4 | Research Engine | Topic + domain lock | 6-round research report + 4–6 image prompts | Gemini MCP, WebSearch, WebFetch |
| 5 | Article Architect | Research report + component inventory | 5 concepts → selected architecture → TOC with sidebar labels → image plan → section metadata | Internal analysis |
| 6 | Draft Writer | Architecture + research + writer persona + design tokens | Framework-native article (.tsx, .vue, .svelte, .html) with inline edit UI and section-level editing | Framework adapters |
| 7 | Quality Gate | Generated article + target persona | 7-signal quality score. Below 7/10 = auto-revision. Max 2 passes before human review flag. | SEO scoring rubric |
The pipeline adapts to whatever the target project provides:
Project has its own component library. TBK-IQ detects and uses them — cards, callouts, tables, heroes — native to the site.
Project has no components. TBK-IQ uses its 193 structural component blueprints, styled with the project's design tokens.
No project detected at all (standalone mode). TBK-IQ generates self-contained HTML with inline CSS, professional defaults, and the full edit UI. Works as a static file, email attachment, or raw upload.
Every generated article includes an inline edit UI. Users click "Edit" on any section, type a revision instruction ("make this more technical", "add a comparison table"), and TBK-IQ rewrites just that section via a bridge server that spawns a Claude subprocess. No full regeneration needed.
Generate once, publish anywhere. CMS adapters, WordPress plugin, Shopify app.
Each adapter translates TBK-IQ's standardized article output (HTML + metadata + images + SEO fields) into the CMS's native format.
| Platform | Integration | Capabilities | Builder Compat |
|---|---|---|---|
| WordPress | REST API + Plugin | Create posts, categories, featured image, Yoast/RankMath meta | All (Gutenberg, Elementor, WPBakery, Classic) |
| Shopify | Admin API + App | Create blog posts, SEO meta, product references | All themes (Liquid-native) |
| Contentful | Management API | Create entries, publish, media upload | N/A (headless) |
| Strapi | REST / GraphQL | Create content, upload media | N/A (headless) |
| Ghost | Admin API | Create posts, set cards, metadata | Native editor |
| Webflow | CMS API | Create collection items, SEO fields | Webflow Designer |
| Custom | Webhook / REST | POST article payload to any endpoint | Any |
wp_posts database level via wp_insert_post(), not at the builder level. WordPress stores content in the same table regardless of whether Gutenberg, Elementor, WPBakery, or Classic Editor rendered it. This is why it works with every builder — it doesn't compete with them.
The WordPress plugin adds:
Embedded Shopify admin app using the Blog API. Product-aware — can reference the store's catalog in generated articles. Simpler than WordPress because Shopify has one content path, not multiple builders. Distributed via Shopify App Store.
What happens when the client's setup doesn't match the happy path.
| Scenario | What Happens | Fallback |
|---|---|---|
| Site is not WordPress or Shopify | CMS adapter system handles this. If the CMS has a REST/GraphQL API (Contentful, Strapi, Ghost, Webflow, Sanity), use the matching adapter. | If no adapter exists: generic webhook adapter POSTs the article payload to any endpoint. If no API at all: output standalone HTML file (current v1 behavior) for manual upload. |
| Site is a custom-built CMS with no standard API | Generic webhook adapter sends structured JSON (HTML body + metadata + images) to a client-provided endpoint. | Client implements a small webhook receiver on their end. We provide a reference implementation. Estimated effort: 2–4 hours for any backend developer. |
| Client doesn't have GSC or GA4 connected | Layer 2 (Intelligence) can still function using Semrush + Ahrefs data alone for gap analysis, keyword opportunities, and saturation scoring. | Decay detection requires GSC data — without it, the system recommends only NEW articles (no refresh recommendations). Prompt client to connect GSC for full functionality. |
| Client doesn't have a Semrush or Ahrefs subscription | TBK-IQ uses its own API keys (cost absorbed into subscription). The client doesn't need their own account. | If TBK-IQ API costs need reduction: fall back to free/lower-cost data sources (Google Keyword Planner API, Ubersuggest API, or web scraping SERP results). |
| Site has fewer than 50 articles (insufficient for voice analysis) | Voice Analyzer needs 50+ articles for statistically meaningful clustering. | Fall back to single "brand voice" persona defined manually with the client (tone guide, vocabulary preferences, example articles they admire). Or use the best 20–30 articles with reduced confidence. |
| All existing content is AI-generated (no human baseline) | Voice Analyzer detects this and flags it. | Two options: (1) Client provides 5–10 reference articles from other publications they want to emulate, or (2) TBK-IQ uses a curated "professional editorial" persona as baseline with client-specific vocabulary overlaid. |
| Site is in a language TBK-IQ hasn't seen before | Research engine and draft writer work in any language (LLM-native). Voice analysis uses language-agnostic stylometric signals (sentence length, variance, structure). | Arabic/RTL is first-class (SRMG). Other languages: research quality depends on Gemini/web coverage in that language. Low-resource languages may produce shallower research. |
| Google OAuth verification takes longer than 3 weeks | Use "Testing" mode (supports up to 100 test users — sufficient for pilot clients). | If verification is blocked entirely: accept manual GSC/GA4 CSV data exports. Less automated but functional. Push verification as a background task. |
| Semrush/Ahrefs API changes pricing or access terms | All SEO data sources are behind an adapter layer. | Swap to alternative providers (Moz API, SpyFu, SimilarWeb) with a new adapter. The intelligence engine consumes normalized data — it doesn't care which provider produced it. |
| Client wants to use a different SEO tool (e.g., Moz, Sistrix, SE Ranking) | Build a new connector adapter for that tool. | Each new SEO tool adapter is ~1–2 days of work. The adapter normalizes into the same ContentPerformanceRecord format. No changes needed in the intelligence layer. |
Every article passes through a scoring rubric BEFORE delivery. Below 7/10? Auto-revise.
Adapted from the Master Kit's Content SEO Scoring framework — a multi-signal quality assessment that catches weak content before it reaches the client.
| Signal | Weight | What It Measures | Threshold |
|---|---|---|---|
| E-E-A-T Signals | 20% | Experience markers, expertise depth, authority signals, trust elements | ≥ 7/10 |
| Topical Completeness | 20% | Coverage vs top 5 competitors' subtopics (completeness matrix) | ≥ 80% coverage |
| Voice Match | 15% | Stylometric distance from target writer persona (sentence length, TTR, cadence) | ≤ 0.3 distance |
| AI Detection Score | 15% | Probability of passing AI detection (sentence variance, vocabulary, cliche density) | ≥ 85% human |
| Freshness Signals | 10% | Current statistics, recent examples, working links, up-to-date screenshots | All data ≤ 6mo old |
| Technical SEO | 10% | Heading hierarchy, schema markup, meta description, image alt text, internal links | ≥ 9/10 |
| Readability | 10% | Flesch-Kincaid grade, paragraph length, section structure, scanability | Grade 8–12 |
The moat nobody else has. TBK-IQ learns from its own output.
After publishing, TBK-IQ tracks the article's GSC/GA4 performance at 30, 60, and 90 days. This creates a closed loop:
If an article underperforms the prediction, the intelligence engine recalibrates: "Articles about [topic X] in this niche don't convert as well as the saturation index suggested." Over time, recommendations get sharper. This is how TBK-IQ becomes more valuable the longer a client uses it — their data makes the engine smarter.
Full scope. 4 parallel workstreams. 3 weeks. All external dependency applications submitted Day 1.
Four teams work in parallel. No cross-dependencies until end of week.
| Stream | Team | Week 1 Deliverables |
|---|---|---|
| A: Data Pipes | 2 engineers | GSC OAuth2 connector, GA4 connector, Semrush connector, Ahrefs connector, Google Trends integration, CMS HTTP crawler, ContentPerformanceRecord schema, data normalization pipeline, caching layer |
| B: Intelligence | 2 engineers | Topic Recommendation Agent spec + scaffold, decay detection engine (port Master Kit logic), gap analysis module, keyword-content mapping (port cannibalization rules), saturation index scorer |
| C: Voice | 1 engineer + 1 NLP | Voice Analyzer Agent, corpus crawler, AI vs human classifier (stylometric signals), writer clustering (HDBSCAN), persona schema design, PoC on 3 real client sites |
| D: Publishing | 2 engineers | CMS adapter base architecture, universal article payload format, WordPress plugin (PHP skeleton, wp_insert_post, Yoast/RankMath meta), Shopify app scaffold (embedded admin, Blog API), media upload pipeline design |
Streams converge. Intelligence consumes data pipes. Voice feeds into draft-writer. Publishing connects to pipeline output.
| Stream | Team | Week 2 Deliverables |
|---|---|---|
| A: Data Pipes | 2 engineers | End-to-end data pull for 1 real client, Supabase schema extension (performance_snapshots, keyword_opportunities, writer_personas, client_connections), data refresh scheduling, API cost monitoring dashboard |
| B: Intelligence | 2 engineers | Wire Mode B to live data pipes, seasonality model (Trends curves), competitor content velocity scoring, scored recommendation output format, SEO quality scoring gate (7-signal rubric), auto-revision loop (below 7/10 = re-draft) |
| C: Voice | 1 engineer + 1 NLP | Persona generator (DNA profiles), inject persona as style constraint into draft-writer agent, voice match scoring (stylometric distance), validate output on 3 client sites — blind test: can client tell AI from their own writer? |
| D: Publishing | 2 engineers | WordPress plugin full build (all builder testing: Gutenberg, Elementor, WPBakery, Classic), Shopify app full build, Contentful + Ghost + Webflow adapters, generic webhook adapter, media upload pipeline (images → CMS CDN) |
Full system integration. Feedback loop. Multi-client dashboard. Demo-ready.
| Stream | Team | Week 3 Deliverables |
|---|---|---|
| ALL: Integrate | Full team | End-to-end demo: category in → Mode B recommendations → user picks → voice-matched article → published to WordPress/Shopify as draft. 30/60/90 day performance tracker, prediction vs actual recalibration engine, multi-client dashboard, scheduled content calendars, client-facing performance reports, A/B headline testing integration |
Everything that needs to happen before each phase can start. Prioritized by dependency order.
The 3 P0 security issues (service_role on disk, tokens on disk, prompt injection) from the v1 audit must be fixed before any new data connectors are added. Data connectors handle OAuth tokens — the same class of vulnerability.
Semrush API access requires a Guru+ plan ($249/mo) or API units purchase. Apply now — approval can take 1–2 weeks. Without this, Gap Analysis and Saturation Index cannot function.
Ahrefs API requires an Enterprise plan or separate API subscription. Provides backlink data, domain rating, and keyword explorer that Semrush doesn't duplicate well. Apply in parallel with Semrush.
GSC and GA4 both require OAuth 2.0. Need a Google Cloud project with the Search Console API and Analytics Data API enabled, plus an OAuth consent screen (starts in "Testing" mode, needs verification for production). Set up now — Google verification takes 2–6 weeks.
Current Supabase schema handles articles and auth. Needs extension for: content_inventory, performance_snapshots, keyword_opportunities, writer_personas, client_connections (OAuth tokens for GSC/GA4/Semrush/Ahrefs per client). Design schema before building connectors.
This is the most complex new agent. Needs a formal spec before implementation: input format, scoring algorithm (how to weight volume vs KD vs decay vs seasonality vs saturation), output format, confidence thresholds. The scoring weights will need tuning against real client data.
Before building the full Voice Analyzer, run a proof-of-concept on 3 real client sites: can the stylometric signals reliably distinguish writers? Can the clustering produce meaningful groups? If the signal isn't strong enough with 50 articles, we may need 100+. This determines feasibility before full build.
The 36-seo/content-seo/content-decay-refresh.md contains the full decay detection methodology (4 detection methods, priority scoring formula, refresh triggers). Extract the decision logic and implement as a scoring module that runs against ContentPerformanceRecords.
The keyword-content-mapping.template.md contains cannibalization detection rules, one-keyword-per-page enforcement, intent alignment checks. This becomes the content map that prevents the agent from recommending articles that would cannibalize existing rankings.
WordPress.org plugin submissions require review (1–4 weeks). Submit the plugin shell early with basic functionality. Full features are added via updates after approval. Also need a WordPress test environment with Gutenberg, Elementor, WPBakery, and Classic Editor for compatibility testing.
Shopify apps require a Shopify Partner account (free) and app listing submission. The app review process takes 1–3 weeks. Submit early. Also requires a Shopify development store for testing.
All CMS adapters consume the same payload. Need to define the canonical format: HTML body, SEO fields (title, meta description, OG image), taxonomy (categories, tags), author, publish date, featured image (URL + alt), schema markup. This is the contract between the pipeline and every adapter.
Currently images are base64-embedded in HTML. For CMS publishing, images need to be uploaded to the CMS's media library (WordPress Media, Shopify Files, etc.) and referenced by URL. Need a media pipeline that: generates image → uploads to target CMS → returns URL → inserts in article HTML.
Per-client operational costs and key risks to monitor.
| Service | Est. Monthly Cost | Notes |
|---|---|---|
| Semrush API | $80–150 | Depends on API unit consumption. Caching reduces by ~60%. |
| Ahrefs API | $50–120 | Row-based pricing. Backlink data is the largest cost driver. |
| Google APIs (GSC + GA4) | $0 | Free within standard quotas (25K requests/day GSC, 10K GA4). |
| Google Trends | $0 | Unofficial API (pytrends) — no cost but rate-limited. |
| LLM Tokens (article generation) | $80–200 | Depends on articles/month. Voice analysis adds ~$20/mo. |
| Supabase (database + auth) | $25–50 | Pro plan. Scales with data volume. |
| Total per client | $235–520/mo | At Tier 2 ($5K/mo) = 90–95% gross margin |
| Risk | Impact | Likelihood | Mitigation |
|---|---|---|---|
| Semrush/Ahrefs API changes or pricing increase | HIGH | MEDIUM | Abstract behind adapter layer. Can swap providers without pipeline changes. |
| Google OAuth verification rejected | CRITICAL | LOW | Submit early. Follow Google's sensitive scope guidelines exactly. Have fallback: manual GSC data export. |
| Voice analysis doesn't produce distinct personas | HIGH | MEDIUM | Run PoC in Phase 1. If clustering fails, fall back to single "brand voice" persona defined manually. |
| WordPress plugin rejected from directory | MEDIUM | LOW | Self-hosted distribution as fallback. Most enterprise clients don't use wp.org directory anyway. |
| AI detection evolves, bypassing voice cloning | HIGH | MEDIUM | The goal isn't to "fool" detectors — it's to write AUTHENTICALLY. Voice cloning produces naturally varied output that reads human because it IS human-patterned. |
Three service tiers designed to match publisher scale — from self-serve intelligence to fully managed editorial operations.
Full AI platform access. Data-driven topic recommendations, voice analysis, automated article generation, and SEO quality scoring — delivered to the client's editorial team via dashboard and content calendar.
$3,000/mo · $36K ARR
Best for: publishers with in-house SEO teams
Everything in Tier 1, plus Chain Reaction managed service. Editorial validation by human specialists, content strategy consulting, voice calibration, and ongoing performance optimization.
$5,000/mo · $60K ARR
Best for: enterprise publishers without dedicated SEO resource
Complete managed intelligence. Automated content generation at scale, GEO optimization for AI search visibility, trending topic discovery, dedicated CR strategy team, and weekly prioritized action briefs.
$8,000–$12,000/mo · $96–144K ARR
Best for: large media groups requiring full content intelligence operations
TBK-IQ transforms content generation
from an execution engine into an intelligence platform.
Semrush tells you what keywords exist. TBK-IQ tells you which ones are worth writing about for YOUR specific site, in YOUR specific voice, and then proves it worked.