Skip to content
RankX AI
RankX AI Docs
Research & Content

Generating articles

Generating a draft from a researched brief in RankX AI. What it produces, why it never publishes, and the pre-spend checks that stop you paying for nothing.

Article generation takes an approved, researched brief and produces a full draft in Content Studio for a human to review. It is dispatched and polled: the call returns a content item id immediately and the writing happens in the background over a few minutes.

It never publishes. The draft lands for review, and putting it on a live site is a separate, deliberate step with its own confirmation.

Before it spends anything

Two pre-spend gates, and both exist so you cannot pay for an outcome that was never possible:

A brief with no completed research is refused before any credits are reserved. So calling too early costs nothing.

A failed dispatch charges nothing. Credits are reserved before the work and released in full if it does not happen, on the same reserve-then-consume model as everything else in RankX AI. See credits and metering.

Length tiers

Generation is priced in tiers by target word count: short, standard and long. The tier changes what you pay and what you get, and the current figures are on the credit cost reference.

Choose on what the piece needs to answer rather than on length for its own sake. A piece that answers its brief's reader questions completely in fewer words is a better piece than one padded to a tier.

What comes back

A draft with:

  • The body, structured against the brief's outline.
  • A meta title and description.
  • Target keywords, carried from the brief.
  • Quality and GEO scores, once computed. See content quality scores.
  • Counts of fact-checked claims and internal links.

A null body means it has not been drafted yet, which is what you will see while the generation is still running. It does not mean an empty article came back.

A null score means not yet computed, not zero.

Reviewing a draft

Treat it as a first draft from a competent writer who has read your brief and your brand profile and has never met your customers. Three things to check every time:

Every claim you would not defend. The draft carries a count of fact-checked claims, which is not the same as every claim being checked. Anything with a number in it is worth reading twice.

Whether it answers the brief's reader questions. Those questions are the sub-queries an answer engine fans a search out into, and a piece that answers them explicitly under their own headings is a piece that can be extracted from. A draft that covers the subject beautifully and answers none of them will not be cited.

Whether it sounds like you. Generation reads your brand profile, so it is closer than a generic tool would be, and it is not you.

Publishing

Publishing is separate, deliberately, and it goes through the WordPress integration if the Website has one connected. Every write there dry-runs first, is verified by reading the object back, and is refused rather than half-applied on a page it cannot safely edit.

See publishing to WordPress.

Driving this from an assistant

generate_article dispatches, get_content_item polls and reads the draft back.

"Generate an article from brief X. Check my credit balance and confirm the cost with me first, then poll until it is ready and show me the opening three paragraphs and the GEO score."

Two things a well-behaved assistant will do here, because RankX AI tells every connected client to: check the balance and confirm before spending, and poll rather than reporting the dispatch response as the result. An assistant that says "here is your article" a second after you asked has read the wrong response. See the tool reference.

What generation is not

It is not a publishing pipeline. Nothing goes live without a human.

It is not a substitute for knowing your subject. The draft is as good as the brief, and the brief is as good as the evidence you chose the keyword on. See content briefs.

It is not a way to produce volume. Scale without value is the thing search engines and answer engines both act against, and the credit model prices generation accordingly.

Where to go next