How articles are written

A new article check runs hourly (0 * * * *, UTC). It picks up briefs and keyword clusters that are due, then writes each one in a fixed sequence: outline, intro, sections, images, then metadata. One article credit is spent per generation job, charged on the first attempt only — if a step fails and the job retries, it resumes from where it left off instead of starting over and re-billing you.

The writing pipeline

1. 1
   Outline

   The model is asked for 5–7 H2 headings covering the target keyword; anything under 4 headings is rejected and retried.
2. 2
   Intro

   A single ~150-word opening paragraph is generated in parallel with the outline, written to hook the reader and work the primary keyword in naturally.
3. 3
   Sections

   Each heading is expanded into HTML — paragraphs, subheadings, lists — one at a time. The target length per article comes from your minimum word count setting (in the scheduling step of a website's settings), defaulting to 1500 words and capped at 1800 regardless of what you set higher.
4. 4
   Images

   Once the text is assembled, image prompts are generated from the draft and illustrations are placed at the first four section breaks — see Article images for how many you get.
5. 5
   SEO metadata

   A focus keyword, meta title and slug are set from the brief; the finished draft also gets a rough SEO score based on keyword usage, heading structure and length.
6. 6
   Schedule

   If the brief had a target publish date, the article status becomes Scheduled and the publisher picks it up at that time. Otherwise it lands as a Draft in Content for you to review and publish by hand.

Every completed step — outline, intro, each section, the finished HTML with images — is checkpointed on the article as it happens, so a crashed or rate-limited retry resumes instead of regenerating (and re-billing) work that already exists.

Writing in your voice

Content settings control how each article reads: a writing style toggle (professional tone, shorter sentences) and a point of view (Expert or first person), set during onboarding or later in site settings, plus the website's configured language. Text generation itself runs on a DeepSeek Flash model — the cheap part of the pipeline; images are the real cost, which is why they're capped separately (see Article images).

You can also add brand voice article URLs — links to existing articles that represent how you already write — during the brand voice step of setup. These are stored on the website as a reference for tone the writing prompt draws on, not a strict style transfer.

How often articles are generated

Generation runs on the publishing schedule you picked when you connected your site, not on demand only. The presets are Light (2 posts/week), Standard (4 posts/week) and Aggressive (daily); Custom lets you choose the exact days of the week, the publish time and the time zone. You can change the frequency any time in content settings, and whatever you choose is still capped by your monthly article allowance.

Publishing after generation

A dispatcher checks for due, scheduled articles every minute and sends each to the right publishing queue for your connected platform (WordPress, Shopify, Webflow, Wix or a webhook). If your monthly article allowance runs out mid-cycle, pending jobs are held as "Paused - Limit Reached" rather than dropped, and resume automatically once your allowance frees up — no need to re-trigger them.