Prepare
Sourced draft + artwork
Kitek internal project
AI-assisted publishing with human approval
We built an AI-assisted editorial and social publishing workflow for Kitek’s own website. It prepares sourced article drafts and artwork, leaves publication to a human editor, and passes approved articles to a separate social publishing worker.
The project shows how AI can fit into an existing business process: useful preparation, a clear approval point and explicit handling when an external service fails.
The approval boundary
Preparation. Human approval. Distribution.
Human review
Check, revise, approve
Publish in WordPress
Editor authorises release
Distribute + record
Facebook / LinkedIn status
The business problem
Publishing a company article involves more than producing text. Sources and claims need checking, artwork must suit the article, and each social channel has different requirements. An approval in one system also needs to mean something precise in the next.
For Kitek, the workflow needed to connect research, drafting, WordPress and social distribution while keeping the decision to publish with a person.
The implemented workflow
Prepare a sourced draft
The editorial worker researches approved topic areas and prepares an article with supporting sources. Software checks required fields and known content restrictions before importing it.
Attach article-specific artwork
The workflow creates a branded social card and attaches it to the WordPress draft. The editor can inspect the image and its alt text or replace it with an approved alternative.
Review in WordPress
The editor checks sources, factual claims, article copy, links and artwork. The editorial importer creates drafts; it has no operation for publishing them.
Publish after human approval
Publishing the article in WordPress queues a separate social worker. That worker prepares channel-specific copy from the approved article and checks the resolved image before contacting a platform.
Record each channel’s result
The worker records publication status separately for each channel. Kitek has verified published posts on Facebook and LinkedIn, including their article links and artwork.
Where people remain in control
AI prepares work; it does not approve its own article. WordPress remains the editor’s review workspace, and publication is the handover to distribution. The draft can be corrected, held or left unpublished.
Validation supports that review by rejecting missing required content, missing artwork or an unapproved image source. The checks catch defined failure cases; the editor still needs to assess whether the article and image are accurate, useful and appropriate for Kitek.
What happens when something goes wrong
These controls make incomplete work visible and reduce duplicate-publication risk. External platform access and service availability still need ongoing attention.
- Missing artwork stops distribution until it can be reviewed and repaired.
- An unavailable channel does not stop other eligible channels or hide newer articles behind a failed attempt.
- Confirmed publication is recorded so a later run can skip that channel. If a platform’s response leaves the outcome uncertain, automatic retries stop for investigation rather than blindly posting again.
- When an article cannot be published to any channel and has no earlier successful publication, the workflow sends an email alert with the reasons.
What the project demonstrates
This is Kitek’s own working implementation, demonstrated through its connected workflow and published Facebook and LinkedIn output.
- Article preparation and distribution can be connected to an existing WordPress workflow.
- Draft creation, human publication and social distribution have distinct responsibilities.
- Article-specific artwork travels with the approved content into verified Facebook and LinkedIn posts.
- Channel status and exceptions remain visible instead of treating a partial result as complete.
Apply the same thinking to your workflow
The same design questions arise in enquiry triage, document preparation and internal reporting: what can AI prepare, what can software verify, who authorises the next step, and where do exceptions go?
A useful first project starts with one bounded process and representative examples. We can help map the handovers, assess the data and integrations, and define what a pilot must demonstrate before it handles live work.
A clear first conversation
Discuss one workflow worth improving
Tell us where information needs interpreting, work is being repeated or handovers are difficult to track. We can discuss a practical first step.