> For the complete documentation index, see [llms.txt](https://oodlz-ai-studio.gitbook.io/docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://oodlz-ai-studio.gitbook.io/docs/outbound-agents/awareness/2-articles/writer-agent.md).

# Writer Agent

2,500–3,500 words per article in one structured pass — 1–1.5% focus keyword density, anti-AI rules at prompt level, images woven at instructed positions, citations sanitised deterministically.

> **2,500–3,500 words, written like an expert — not a summary, not training-data filler.**

The Writer Agent receives the full research brief plus image metadata and produces the article in one structured pass. Focus keyword density, anti-AI writing rules, image placement, branded H2 labels, pull-quote placement, table limits, and FAQ structure are all enforced at prompt level — not added by post-processing.

## How It Works

Before the writer runs, a pre-step extracts a clean 2–4 word focus keyword from the semantic keyword list. Pre-extraction removes LLM self-extraction variance — the writer always uses the same keyword, regardless of how the topic was originally phrased.

The writer produces 2,500–3,500 words in one pass. Focus keyword density holds at 1–1.5% (25–40 exact-match uses) with placement in the title, first paragraph, and multiple H2/H3 headings. Anti-AI writing rules are embedded in the prompt: banned phrases, no perfect paragraph symmetry, no formulaic opening structures. Images are woven into the content at the positions the Image Generation Agent specified, with captions and alt text already prepared. Every H2 carries a branded 2-word label prefix. Pull-quotes are placed at natural pause points. Tables are limited to 1 per article. FAQ sections are structured for rich-snippet eligibility.

After the writer runs, a SANITIZE Citations step strips any non-approved source attributions deterministically — using the research brief's approved source list. No LLM is involved in this step. No hallucinated citations survive into the link pipeline.

## What You Get

* 2,500–3,500 word article in one structured pass — no piecewise generation, no post-hoc stitching.
* 1–1.5% focus keyword density with placement in title, first paragraph, and multiple headings — Google's helpful content system has a harder time penalising this output.
* Anti-AI writing rules at prompt level — banned phrases blocked, formulaic structures suppressed.
* Images woven into the prose at the right positions — captions and alt text written upstream and reused, not regenerated.
* Pre-extracted focus keyword — same keyword used by writer and link pipeline, no self-extraction variance.
* SANITIZE Citations strips non-approved attributions deterministically — hallucinated citations never reach publish.