LLMs for the Rest of Us
“I made a custom GPT that started as a checklist we used on every new article for proper SEO structure. It addressed headers, metadata, article formatting, and other parameters,” said Essex, “to ensure that it met Google’s Search Essentials protocol so search engines would find it.” He next created a Spam Checker GPT to help improve open rates and prevent emails from being sent to the spam folder. “I put in 2,000 buzzwords that were likely to divert emails into the spam or promotions folder and gave the GPT instructions to rewrite the emails accordingly.” These applications were made available to customers once they were tested and proven to be effective.
One of the other applications is a prompt optimizer. “It’s for people who want to benefit from using AI but aren’t sure what to say to it,” commented Essex. “The output you get from AI is only as good as the instructions you give it. Using this application, you can type, in your own words, what you want it to do, and then it’ll give back a fully structured AI prompt.” Getting useful results is dependent not on just picking the right LLM model, but on how the prompts are engineered.
He also created an AI assistant named “Julie” that responds to his emails for him. “I wanted a gatekeeper of sorts to qualify incoming emails to help maximize my time. I also wanted to get a female perspective and an agent that would not hesitate to question me, so I asked for responses from my best friend since high school,” Essex explained. Julie now answers his emails using AI analysis provided by Google Gemini. He still reviews each message, noting, “We are nowhere near removing humans from the loop. Those who are trying to do this are failing spectacularly,” he maintained. “However, what I might generate is only what's in my own head, whereas AI has what’s in the heads of 400 million people and can augment what’s in mine.”
As a small business, POWR is acutely aware of the challenges faced in scaling operations throughout the company. “Time is a nonrenewable resource,” he pointed out. “I still spend 10 hours a week editing content, but I am able to complete three times as much work in that time.”
This process has been integrated into the workflow so that as soon as a new document is put in the folder, the AI agent edits it, plugs in images and backlinks, then schedules, and, after review, publishes it.
Top LLM Applications for SMBs
LLM use cases for SMBs and individual consultants should focus on functions that consume large amounts of time and cannot be done at scale without automation. Good examples are customer service and lead generation. In the past, customer service chatbots had a limited repertoire of answers, and users quickly ran into dead ends if their question fell outside the scope. The ability to use LLMs that have extensive amounts of underlying data and respond in natural language goes a long way toward overcoming these limitations.
“The LLM produces a natural language response for the customer or a customer support agent,” said Gabriele Monti, founder, Language Media LTD (language-media.co.uk). This is a significant step up for customers who previously had interacted with chatbots that never seemed to understand the question. To enrich the underlying knowledgebase, Monti suggested fine-tuning the LLM model with data from past support tickets as well as using retrieval-augmented generation, which allows the LLM to access verified and relevant content.