Challenge
Editorial research was a recurring efficiency bottleneck. Editors needed to build detailed briefs for freelance writers, but gathering the underlying research could be time-consuming.
The goal was to make that research process substantially more efficient without lowering quality or removing human judgment from the workflow.
My Role
I identified the use case, tested multiple AI tools, designed the prompt framework, created the operating process, documented appropriate use, trained editors, and iterated on the workflow as model capabilities changed.
I also established the human-review expectations and practical governance controls around what information could be shared with AI tools.
Workflow Design
I began by testing the process on several real assignments for which I already knew what strong research should contain. Those tests helped me identify the inputs, context, and instructions that consistently produced useful output. I converted those lessons into a repeatable prompt framework editors could use across topics.
Editors used the framework to generate research for article briefs, then reviewed and refined the output before assigning work to freelancers.
In the second phase, we incorporated the prompt framework into the editorial-planning process so AI could support batch research across multiple planned assignments. That reduced the amount of manual research editors had to conduct from scratch.
Human-in-the-Loop Governance
Human review was mandatory. Editors were expected to read the AI-generated research, identify anything that appeared inaccurate, incomplete, or questionable, and correct or remove it before using the material.
Freelance writers were also expected to verify the research they received and flag anything that did not appear correct. The workflow therefore treated AI output as a research starting point rather than authoritative source material.
We also maintained boundaries around sensitive information. Proprietary or potentially problematic company information, including internal sales information, was not provided to external AI systems.
Pilot & Rollout
I tested the workflow personally before broader use, using real content assignments to determine whether the output was useful enough to support editorial work.
Once the prompt framework was performing consistently, I introduced it to my editors and moved from testing to implementation within approximately one week.
Training & Adoption
- Documented SOPs
- Clear guidance on when and how AI should be used
- Reusable prompt templates
- Demonstrations of the workflow
- Ongoing discussion and coaching during weekly one-on-one meetings
The process was updated iteratively as AI models and capabilities changed.
Risk Management
The primary risk was over-reliance on AI, particularly the possibility that users would accept generated research without verification.
Models occasionally returned incorrect information, so the operating model deliberately kept humans responsible for validating the work. Rather than trying to eliminate that inherent model limitation, the workflow was designed around it: AI accelerated the first pass, while editors and writers remained accountable for accuracy and quality.
Impact
The program materially improved research efficiency for editors. Instead of spending most of their time gathering baseline information, editors could use their time to evaluate, refine, and improve AI-assisted research before turning it into stronger assignment briefs.
The batch-research phase increased that efficiency further by applying the same repeatable framework across multiple planned assignments.
No formal productivity KPI was established for the initiative, so the impact is best characterized as a substantial operational efficiency improvement rather than a quantified percentage gain.
Lasting Impact
The initiative established a repeatable AI-enabled research workflow rather than a one-time experiment.
The prompt framework, usage documentation, human-review expectations, governance boundaries, and training approach remained in place and were updated as models and capabilities evolved.
The lasting capability was a practical operating model for adopting new AI tools quickly without treating automation as a substitute for human judgment.