> ## Documentation Index
> Fetch the complete documentation index at: https://docs.koalr.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Drafts
> Setting up a draft, what the content agent does, the Evidence tab that shows whether a piece is actually optimised, and the content pack.
## Setting up a run
**Create content** opens a five-step wizard. Nothing runs, and nothing is spent, until the last step.
| Step | What you decide |
| ---------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Steer** | A **prompt to win** (optional, but the only way the piece becomes measurable; prompts are listed worst first by your visibility on them), a **page to work from** (optional: a rival page to out-write, or one of your own that is working), and **your own steer**, the brief in your words. You need at least one of the prompt or the steer |
| **Shape** | The content type and the length: Brief, Short, Standard or In-depth, with Koalr's recommendation and the reason for it |
| **Voice** | Your saved tone of voice document, or a one-word tone while you have none; the language (UK or US English); emojis on or off |
| **Ground rules** | Which competitors the piece may name (anything unticked is excluded) and anything specific it has to include |
| **Review** | Everything the agent will be told: target prompt, reference page, steer, shape, voice and ground rules, plus **Why this piece**, the plan with the evidence for doing it now. Edit anything, then **Run the agent** |
Why this piece is stored on the finished piece, so a draft opened months later still explains itself.
## What the agent does
A pipeline, not one prompt: gather your cached pages and the rival's, study the page to beat, search for fresh facts, plan the outline, write, polish with an expert panel, finalise. Runs happen in the background; you can navigate away. The job page shows the **Workflow** as a timeline with each step, its duration and the sources it used, so the draft is traceable to real research.
## The right-hand rail
Beside a finished draft:
* **Workflow**: the timeline.
* **Outline**: the document structure, for jumping around a long piece.
* **Evidence**: see below. It appears once the piece has been checked; drafts written before the checks existed do not show it.
## Evidence: is this actually optimised?
Above the draft is a one-line verdict: how many of the retrieval conditions the piece meets, out of the ones that could be measured. The Evidence tab breaks that into rows: answers the question in the opening, brand is named, specific enough to be quoted, sub-questions asked the way people ask them, passages sized to be lifted whole, deep enough to displace the page being cited. Each row opens to show why it matters and the exact instruction the writer was given.
* **It is live.** Every check re-runs as you edit. If your own polishing buries the answer, you see it immediately.
* **Show evidence** highlights the exact sentences each check is about, directly in the draft. It is a view only; nothing is saved into the article.
* **The page currently cited is scored with the same rules**, so the comparison is like for like.
* **Find the quote** shows the passage an AI would most likely lift, verified word for word against the draft before it is shown.
Evidence means the piece does the things cited pages do. Whether it is cited is measured after you publish, on every run. Nobody can promise a citation, and the Published tab is the only honest way to answer it.
## Editing
The editor has **Title** and **Meta description** fields and the usual formatting: headings, bold, italic, strikethrough, links, bullet and numbered lists, quotes. **Copy** takes the piece to your clipboard and **Save changes** keeps your edits. If you have drifted while editing, ask Kevin on the draft to **Put it back in your voice** and he re-runs the voice check against your tone of voice document.
## Content pack
**Repurpose this piece** turns the finished piece into a LinkedIn post, a thread, an FAQ block, a 60-second video script, and Article plus FAQPage JSON-LD ready to paste. Everything in it traces to something the article actually says.
## Publish and track
Open the draft, add the address you published it at, and measurement begins. From then on every run reports whether AI platforms have started citing it. See [Published](/act/content-studio-published).