Designing AI-assisted editorial workflows.

Client context

QLM Media is an independent media company developing explanatory content, editorial products and audience propositions in Tunisia and Africa. PENCILS and QLM remain distinct organisations; this work drew on QLM’s own editorial operations.

Context

Editorial teams spend significant time discovering topics, consolidating sources, preparing research and adapting material between audiences and formats.

These processes create opportunities for AI assistance, but they also create quality, sourcing and reputational risks if automation is poorly bounded.

What we did

PENCILS and QLM used QLM’s editorial operations as a field environment to design and prototype AI-assisted workflows.

The objective was to identify tasks where AI could provide useful support while keeping source verification, editorial judgement and final approval under human control.

Workflow areas

Discovery

Support for identifying potentially relevant scientific, economic and media topics from defined source sets.

Editorial selection

Structured generation of possible subjects and angles for different audiences, followed by human selection.

Evidence management

Source provenance and verification requirements defined before drafting.

Draft assistance

Generation of structured first drafts only from approved evidence, with human review before publication.

Multilingual adaptation

Assistance adapting content between markets and languages while preserving editorial meaning and terminology.

Cost control

Matching task complexity with appropriate tools and models rather than applying the most expensive model to every step.

Operating principles

Human approval at critical stages

Traceable sources

No autonomous publication

Explicit quality criteria

Clear treatment of sensitive information

Cost and token discipline

Outputs

Workflow map

AI use-case prioritisation

Prompt and agent architecture

Verification protocol

Human-review gates

Operating guidelines

Cost framework