Integrating AI Tools Into Your Workflow: Custom LLMs

How and why to make a company GPT

Perhaps you’ve heard about custom GPTs. A generative pretrained transformer (GPT) is a specialized large language model (LLM) that generates and structures text. In other words, it writes for you. For businesses, a custom GPT can significantly streamline written content creation, particularly the type of hard-working, fact-based communications that fill so many corporate to-do lists.

A custom GPT will generate more accurate written content that is more specific to your unique business needs, all with more appropriate industry terminology than generic models. They also provide greater tailoring to your specific use cases.

Improving brand consistency with custom GPTs

This is precisely why we built a custom GPT for a major client recently. They had a number of professionals dispersed across the organization who needed assistance with professional writing. Since none were trained in developing copy, they wanted help getting started on composing and proofing highly structured content like press releases and user manuals.

That is an ideal application for a custom GPT.

Our 3 steps for building a custom GPT

• 1. Create your own walled space within ChatGPT. This ensures that the proprietary content we upload will be kept discrete and not used to train ChatGPTs LLMs.
• 2. Upload the client’s marcom style guide. This ensures the outputs later will adhere to the correct brand language and format.
• 3. Upload other brand content. Providing additional historical content allowed the machine to learn everything from style and tone to actual product specs.

The result? In short order, we created a gated tool that allowed the company’s professionals to generate polished brand-consistent drafts of written content that adhered to the company’s proofing and styling guide.

The drafts it returns literally cut hours from their development timelines: all consistently formatted and all adhering to their legal parameters.

You still need a human in the loop

There’s always a “but” and this one is crucial. Artificial intelligence is not perfect. It is not infallible, and it doesn’t truly think so much as aggregate and statistically generate.

So, you can’t simply enter a prompt, download the output and hit Send. At least not yet. As of now, it’s incredibly important to keep a human in the loop. Because, yes, AI can make mistakes, sometimes embarrassing little ones, sometimes fact-based bigger ones.

The advantage of a custom GPT is speed to draft, not speed to execution; that remains a human responsibility. You must always consider GPT output as a draft to be reviewed and edited for accuracy. So, keep a human in the loop.

It’s a small price to pay for the benefit of generating first drafts in minutes.