Look Left clients create technology that powers the world’s digital experiences — from ingress and DNS to observability and databases. These aren’t run-of-the-mill consumer topics. They’re complicated, nuanced and highly technical. Even large language models (LLMs) like ChatGPT that are trained on the entire internets knowledge base, can struggle to create compelling and accurate content about them. That doesn’t mean that B2B tech companies can’t benefit from generative AI (GAI) tools; it just means they need to teach the tech to get the best results.
Before we get into how to educate GAI tools, let’s set the stage. GAI tools can produce a range of content — from social images and copy to entire websites. Although these tips can be applied across content types, we’re going to focus on how to prompt LLMs to create impactful blog posts for B2B tech.
What to consider when you use AI to write blog posts
It seems like a dozen new GAI tools are announced every day. However, a few that were early to market (and have the trust and brainpower of major brands behind them) have the largest adoption. A leader among them is ChatGPT, which is based on the GPT LLM. How it works is simple — a user types in a prompt or question in natural language and ChatGPT produces a response by predicting the most likely next word. Because these predictions are based on the information it has, users need to be aware of a few things.
Timeliness and factual accuracy
In certain cases, LLMs have a definitive end date and may not include the most up-to-date information. GPT-3, for example, was trained on data up to June 2021. This is already being solved with updated models and plugins that allow certain tools to search the internet. Regardless of the tool you’re using, editing your AI-generated content is critical to ensuring accuracy for your B2B audience.
Remember, LLMs respond with the most likely next word, not necessarily the correct answer to a question, which is why it’s important to fact-check responses. For example, a few members of the Look Left team put our own names into ChatGPT and were all identified as the founder of Look Left Marketing. Funnily enough, the only person who wasn’t identified as such was the actual founder, Bryan Scanlon. Tools like the Open AI Bing integration produce source links, which make fact-checking a lot easier.
In a recent article from MIT, researchers asked ChatGPT if it’s biased. Its response was, “Yes, language models can have biases, because the training data reflects the biases present in society from which that data was collected. For example, gender and racial biases are prevalent in many real-world datasets, and if a language model is trained on that, it can perpetuate and amplify these biases in its predictions.”
These researchers, among others like Armilla, are working to prevent bias in AI, but for now, check AI responses (and your inherent biases) before publishing.
GPT is trained on the entire public internet. That means it considers information from academic papers written by PhDs as well as tween TikTok videos. It doesn’t, however, inherently know your brand’s tone of voice, messaging and keyword strategy, or audience. Tools like Writer can be trained with this information, or it can be added to your prompt when using a tool like ChatGPT.
Prompt engineering tips to get quality AI content
The job search for “prompt engineer” returns more than 300,000 results on LinkedIn, but you can become a power user of GAI even if it’s not in your job description. These simple tips will help you help AI generate quality content.
At a minimum, all prompts should contain details on the content type, topic, intended audience, desired tone and word count. You can even adjust the “temperature” from 0 to 1 in ChatGPT based on your desired output. A lower temperature will return a more mainstream response (consider 0 for a blog about compliance), while a higher temperature will return a more creative response (consider .5+ for an anecdotal blog about an experience using a product).
Here’s an example of a prompt: Write a 600- to 800-word blog post on the benefits of a telemetry pipeline. This should be written for a highly technical software developer audience.
Use existing content to inspire and train
While certain GAI tools can’t access the internet or easily crawl URLs, some can. If you’re using Bard for example, experiment with adding depth to your prompts by sharing one to three URLs of topically-related content you like. If you’re using ChatGPT or other GAI tools that can’t access the internet, copy and paste key snippets of text from materials your organization has already written to guide it.
Use prompt chains and a content outline
Don’t expect an LLM to produce high-quality technical content that spans multiple subtopics from a single prompt. Instead, consider using a series of smaller prompts that let it return more specific responses. A great way to do this is by referring back to your content outline to create unique prompts for each blog section. Before you move on to the next section, tell the tool if you need content revised, expanded, truncated, etc., and continue to respond until you get your desired output.
Become a power user of generative AI tools
As with any technology, process or relationship, the more time you invest, the better the results. In a recent roundtable at VentureBeat Transform, someone compared LLMs to a new colleague. When you first start working together, you don’t know one another’s strengths, experience or communication style. Over time, you learn what questions to ask and how to ask them to get to the answer you’re looking for. Think of GAI tools like a colleague. Collaborate with them on projects and invest time to get to know them. With the right prompts, LLMs may just become your work bestie.