Workflow 2: Brief Generation

<aside> 📑

This stage turns an approved keyword database into an editorial brief. Research and structure generation are automated; the brief itself remains an editable approval artifact.

</aside>

Workflow map

flowchart LR
A["Approved keyword sheet"] --> B["Download + extract"]
B --> C["Google AI Overview"]
C --> D["Parse text, references,<br>related queries"]
D --> E["AI: Select relevant references"]
E --> F["Scrape up to 5 articles"]
F --> G["AI: Deconstruct article structures"]
G --> H["AI: Create article structure"]
H --> I["Create Google Doc<br>PRE-APPROVE brief"]
I --> J["Human edits + approval"]
J --> K["Move to Approved Briefs"]

What is automated

New Keyword Database Trigger watches the approved-keyword folder and reads the topic, intent and selected terms from the approved sheet.
The workflow queries Google's AI Overview and separates its text blocks, references and related searches into usable fields.
Extract Relevant References selects up to five article pages from the supplied reference set and also converts the AI Overview into a compact map of themes, user intent, pains and possible solutions.
Selected articles are fetched and cleaned. Extract Structures from Reference Articles deconstructs them into headings, topics, insights, examples, statistics and quotes while preserving source links when they are actually present.
Create the Article Structure1 receives the article evidence, topic, intent and keyword set. It builds Title, Description, H1 and a writer-facing article_structure, while keeping link integrity, evidence grounding, non-repetition and the requested SEO intent.
The result is normalized by code, written into a Google Doc, renamed [PRE-APPROVE] ..., and announced in Telegram.

Human checkpoint

The brief is the central control surface. A person reviews the generated structure, can change headings and scope, add product knowledge, anchors, internal links or custom instructions, and can reject weak evidence. Only moving the brief into Approved Briefs starts article writing.

<aside> 🧩

The automation can generate the brief, but it does not own the editorial decision. A fully custom human-written brief can enter the same downstream workflow.

</aside>

Selected node details

Node	Why it matters
`Filter Out AI Overview Text Blocks` / `Filter Out References`	Deterministic code separates search context from source candidates instead of passing the raw search response directly to the model.
`If`	Checks that AI Overview text and usable `https` references exist. The error branch stops the research path and sends a Telegram warning.
`Extract Relevant References`	AI gate that may choose only from supplied references; banned candidates and non-article pages are explicitly excluded.
`Extract Structures from Reference Articles`	Turns fetched article text into source-faithful research notes, with special checks around statistics, quotes and citations.
`Create the Article Structure1`	Main planning agent. It decides article route and section roles from evidence and intent, then returns a strict JSON outline.
`Finalize the Brief`	Builds the actual editable brief document with size, meta, keywords, related queries, custom-instruction space and the structured outline.

<aside> 🤖

The exact prompts for reference selection, source deconstruction and article-structure generation are in Prompts & Code Guide.

</aside>