Content Loop Studio is six AI agents working in parallel, each scoped to one job. The Supervisor routes messages. The Analyst reads your analytics. The Writer drafts in your voice. The Publisher opens PRs. The Approver gates merges. The AEO worker pool samples AI assistants. The complexity is hidden. You only ever talk to the Supervisor in chat.

Below is what each part does, why it exists, and the explicit tradeoffs we made. If you want the full architectural detail, see the Platform PRD.

Chat-native interface

You @mention the bot in Microsoft Teams or Google Chat. No new dashboard, no separate login. The Supervisor agent classifies the intent (topic ideation, draft, audit, AEO check, approval) and routes to the right specialist. Replies post back to the same channel within seconds for fast commands, around 30 minutes for full drafts.

Why: B2B teams already live in chat. Adding another tool to the stack is friction. An @mention is the lowest possible cost of adoption.

Tradeoff: Chat is asynchronous. Long-running operations like a draft return a status reply first and the result later, not in one synchronous turn. The flip side is that the channel doubles as the audit log.

Grounded analytics, not generic SEO

Every topic idea and every draft is grounded in your actual data. The Analyst agent has tools for Google Search Console queries, GA4 analytics, semantic search over your existing site content, and competitor SERP checks. When it recommends a topic, it cites the impressions, position, and CTR behind the recommendation.

Why: Generic SEO advice (“write about industry trends”) doesn’t move the needle. Specific advice (“you rank #14 for DMARC with 4,210 monthly impressions; competitor Hooli ranks #3 with a similar post”) does.

Tradeoff: Setup requires read-only Google service accounts for GSC + GA4. That’s a 10-minute task for your ops team, but it’s not zero. The payoff is that nothing the bot says is invented.

Brand voice tuning + humanizer pass

The Writer agent loads your brand voice document and three to five example posts on every draft. In the first month, I iterate with you on the voice doc. Adjustments come from real drafts the agent produces against your brief. By week four, drafts read like your team wrote them.

Then every draft passes through a humanizer post-process step before opening a PR. Adapted from the open-source humanizer skill (MIT-licensed), it detects 29 known AI-writing patterns: significance inflation, AI vocabulary (“crucial”, “landscape”, “delve into”), em-dash overuse, signposting, copula avoidance (“serves as”, “stands as”), rule-of-three filler. It rewrites them in your voice.

Why two layers: Brand voice tuning gets the substance right. The humanizer pass closes the cadence gap. Customers can still spot AI-cadence if either layer is missing. Together they get you past the threshold where the “is this AI?” question stops mattering.

Tradeoff: Week one drafts are still mediocre even with the humanizer. The voice doc itself needs tuning before the humanizer has a target to calibrate against. It takes 3-5 rounds of “no, more direct”, “less qualifying”, “drop the lists of three” before the voice settles. The humanizer pass is opt-out via tenant config for the rare tenant who wants raw Writer output, like formal academic documentation where AI-cadence isn’t a tell.

AEO (Answer Engine Optimization)

The AEO worker pool samples Perplexity and Claude with web search in parallel, with up to 30 prompts in a single baseline. ChatGPT-with-search and Gemini-with-search are in active development for the next release. Each prompt gets entity-extracted: how often your brand is mentioned, which competitors are mentioned alongside you, which sources the assistants cite. Monthly baselines get diffed automatically.

Why: AI assistants are eating search. If Perplexity cites your competitor and not you when someone asks about your category, you’re losing market share. Tracking this is becoming as important as tracking Google rankings.

Tradeoff: AEO data is noisier than search data. It’s sampling, not census. We mitigate by running large prompt sets and tracking trends rather than single-snapshot positions. The trend over months is the signal. Week-to-week noise is real. v1 covers two assistants well rather than four superficially.

Human-in-the-loop approval

Every draft is a GitHub pull request with a Cloudflare Pages preview URL. You see the rendered post on your domain before it goes live. Approve in chat (approve), revise with feedback (revise: tighten the intro), or reject. The PR gating is enforced platform-side. The merge tool refuses to fire without a recorded approval on the Supervisor’s storage.

Why: Full AI publishing is a bad idea. You stay in control over what ships to your domain. The chat-based approval is fast enough that this doesn’t slow you down.

Tradeoff: You have to review. We don’t auto-publish even after N approvals or for “low-risk” categories. Your name’s on the byline.

Founder-led implementation

In the founding cohort phase, I onboard every tenant personally. That includes writing the initial brand voice doc with you, designing the AEO prompt set, configuring the chat backend, setting up the GitHub + Cloudflare Pages flow, and tuning prompts for the first 30 days. After ramp, you have a direct Slack channel with me for anything that breaks or needs adjustment.

Why: AI product at this stage is more service than software. The buyers who’d self-serve aren’t the ones we’re serving in the founding cohort.

Tradeoff: We cap the founding cohort at ten because I’m one person. After that, post-cohort customers get a dedicated channel but not necessarily founder-direct.

What you don’t get

Worth being explicit:

  • No image generation. Drafts are text. If you want hero images, your designer or a separate tool handles them.
  • No social media scheduling. Content Loop Studio drafts long-form blog posts. Repurposing to social is a v2 feature being considered.
  • No multi-language. English only for v1. If your audience is non-English, this isn’t the right tool yet.
  • No A/B testing of drafts. We ship the draft you approve, not multiple variants.
  • No email marketing. Just blog content. Repurposing to email or newsletter is a v2 candidate.
  • No analytics dashboards beyond what’s in chat. Status, AEO baselines, and request history come back in chat. There’s no separate web UI for now.

If any of those are dealbreakers, say so on the discovery call. I’d rather find that out before signing.