How Shira Lazar Built a Custom GPT for Her Editorial Team

Shira Lazar introduces a practical demonstration of the custom GPT built for What’s Trending. She and guest Jim Marsh plan to explain the underlying framework and how the team uses it in editorial work.

Marsh describes his background building digital marketing and social media capabilities at HBO. He says the arrival of generative AI coincided with his transition into a new phase of work.

He frames AI as a collaborator that can amplify human ideas rather than replace a person’s thinking. Shira asks how his marketing background informs his view of emerging creative tools.

Marsh predicts that lower barriers to producing polished visual work will enable new kinds of storytelling. Shira distinguishes that creative question from the use of AI to tailor marketing at scale.

The pair introduce a worksheet and their work together. Shira says she wanted a shared editorial framework for a small team, after finding that relying on individual chat memory did not give her the team workflow she needed.

Her goal was to give writers access to an editorial reference when she was unavailable: something that could help maintain brand voice and improve content for different platforms and search.

Marsh outlines three steps: encode intent, structure a repeatable system, and collaborate and refine. The first asks what a person or organization is trying to do; the second turns that purpose into reusable workflows; the third improves them through use.

Shira says writing these ideas down also forces a team to clarify its brand. If the organization is uncertain about its identity, AI can still produce a direction, but that direction may not be the one it wants.

The starting material was a document developed with her team from brand data, priorities and platform-specific principles. Marsh stresses that the original creative and strategic signal came from people before an AI system processed it.

Marsh also says people should review public-facing work. Shira describes the source material as something like a brand kit or internal guide that helps the team understand what it wants to publish.

In a demonstration recorded in 2025, Marsh opens the GPT area of ChatGPT and describes custom GPTs as specialized configurations for a particular use. Interface details shown in the recording may change.

They open the What’s Trending configuration, with its name, description, instructions and reference files. The written role is to help writers and editors generate, review and refine material in line with the brand.

The reference material includes mission and values, an editorial style guide and strategic goals. The instructions ask for a particular tone, clear answers and clarifying questions when a request is vague.

Shira asks how to define “problematic” content without leaving the term ambiguous. They discuss flagging material that appears off-brand and tying critiques back to the source documents instead of merely validating the user.

The configuration also offers conversation starters, such as asking for a headline or checking whether a post fits the brand. The prompts are intended to make it easier for team members to begin.

Marsh says he used AI to help draft the instructions from Shira’s material, then checked and revised the result. He emphasizes that this involved several rounds of quality assurance rather than a one-time setup.

The documents form a knowledge base intended to reduce repeated setup for common tasks. Marsh says this first version was useful but that updating it and sharing it across different tools presented limitations for their team.

He wanted the team’s working principles to travel between tools rather than depend entirely on one product’s memory. That led to what he calls a strategic augmented intelligence layer, or SAIL.

The approach has two documents: a human-facing manual explaining the system and how to use it, and a separate file intended to give an AI system context and instructions. The manual lists possible uses such as evaluating a story or drafting headlines.

The team guide also includes cautions against overreliance, entering private or confidential material, speculating about individuals and making unsourced claims. Marsh says the system is meant to assist people.

He calls the documents a living system: team members should compare notes on what works, update the material and share useful prompts. A prompt library gives the group a place to keep reusable examples.

Shira observes that many people need help turning individual habits into repeatable processes. Marsh connects that work to his experience guiding teams through earlier technological change.

They recap the progression from source documents to a custom GPT and then to portable documentation for the team. Marsh opens the SAIL file for the AI and prepares to test it in a fresh chat.

The AI-facing document begins with instructions and organizes the team’s context into sections. Marsh says it was written so people can read and update it as well as provide it to an AI system.

Shira asks how a beginner could build something similar. Marsh recommends starting with identity, purpose, how the person wants to work with AI and a few focused use cases; the system can develop from there.

They inspect sections on identity, decision principles, risks and examples. The document compares stronger and weaker headlines, using specificity and usefulness as editorial considerations.

Marsh recommends replacing examples and adding real team experience as the document evolves. He saves it as a PDF and uploads it into a temporary chat to see how it responds without his usual personal context.

The system proposes uses including evaluating story ideas, generating headlines, checking brand safety and improving engagement. Shira says her team should write its own draft first and then ask for feedback.

They paste an existing article into the chat and ask for feedback on search, readership, engagement and answer-engine optimization. Shira describes this as an editorial aid when she is not available to review every draft herself.

The response suggests alternative headlines, questions the article could answer and a stronger opening. Shira says a writer can review these suggestions and decide which to implement.

Marsh then asks AI to help create a repeatable prompt for the same evaluation. The pair discuss adding a useful version to the team manual or shared prompt library.

For the main takeaway, Marsh returns to intention and structure. A clear purpose and a little context can be a starting point; people can then develop focused use cases and update the system through practice.

Asked about the future, he says the speed of change makes predictions difficult. He encourages people to stay grounded in their own goals while learning how the tools can help.

Marsh says he deliberately takes walks and time away from generative technology so he can return with attention and creative energy. Shira thanks him, invites feedback on the demonstration and points viewers to their worksheet.