Almost overnight, artificial intelligence has jumped out of the pages of science fiction and into our business reality. Though not exactly like Data of Star Trek fame or the walking, talking androids dreamt up by Asimov – for the time being, at least – generative AI (GenAI) still represents a massive shift in the status quo.
As a seasoned marketing strategist who’s been working closely with generative AI technology for the past year, I’ve witnessed TechTarget’s journey from experimenting with GenAI to achieving real business results. In this post, I’ll share our experience and the best practices we’ve learned along the way.
How TechTarget’s Content Marketing team is using GenAI
To ascertain the right fit for our AI tech stack and strategy, my team started by brainstorming use cases: What processes would benefit from generative AI? Where can it boost scale and efficiency? After weighing the risks and rewards (and after a whole lot of testing), we now use generative AI in the following ways:
| Subject line generation | We leverage generative AI to produce subject lines for marketing emails at scale. With a little bit of instruction, GenAI can construct strong subject lines in a brand’s unique voice – for us, this means adhering to five predefined subject line “styles” that also observe our character length guidelines. |
| Content summarization | GenAI can scrape long pieces of written content and quickly distill its key points or suggest a strong summary or abstract to include in promotional channels. |
| Content classification | We use GenAI prompts to quickly classify entire content libraries by attributes like topic, type, target persona, vertical industry focus, content age and more. In our experience, limiting the number of potential classification choices yields the most accurate results – for example, we defined just 20 of our more common content types and fed this list into our GenAI tool. |
| Plagiarism scanning | AI-based content checkers and plagiarism detection workflows are an essential tool in any marketer’s AI toolbox – they offer speed and efficiency while also providing a legal safety net. |
| Language translation | When it comes to translation, generative AI is becoming increasingly mature (though we still maintain a robust QA process, confirming output accuracy via human in-language experts). |
| Formatting | GenAI is an endlessly useful tool for formatting data. While many word processors have traditionally included rudimentary formatting tools (Microsoft Word’s “replace” function has saved me from manual drudgery on more than one occasion when I decided against a particular phrasing), GenAI’s sophistication kicks data formatting up a notch. For example, given a video transcript with timestamps embedded throughout, I no longer need to delete those timestamps by hand – AI can take care of it for me. |
| SEO strategy and remediation | GenAI delivers a ton of value in search engine optimization (SEO) – it can do the heavy lifting of analyzing a piece of content and suggesting SEO keywords in order to boost its visibility. GenAI can even breathe new life into articles by identifying outdated or ineffective SEO terms and offering more compelling replacements. |
Five tips for writing better AI prompts
Prompting isn’t just a means to produce a GenAI output – it’s a skill that can be learned and enhanced. My team has substantially increased the value we realize from GenAI by developing this skill. Here are five of our AI prompt-writing best practices.
|
1. Be specific about what you’re looking for: Vague prompts will elicit vague results – and chances are, they won’t be the results you want. For a more precise output, include details like target audience, output length, specific topical focus, etc. |
❌ PROMPT: Write a subject line for an email about cybersecurity ✔ PROMPT: In 75 characters or fewer, write an email subject line for a C-suite audience that promotes a blog about risk management |
|
2. Tell the AI who it is: If your GenAI tool knows what you expect it to be, it’s more likely to generate the output you’re looking for. After all, you’d likely have very different expectations when reading an article written by a scientist versus one penned by a comedian. In a similar vein, defining the AI’s job within your prompt will set the tone for the output. You have room to play around with these definitions: you may find better results by telling the AI “you are an award-winning research assistant”, or even “you are a research assistant whose job depends on analyzing data successfully.” |
✔ PROMPT: You are a research assistant focused on analyzing datasets and identifying trends within that data |
|
3. Tell the AI what to do, rather than what not to do: “Don’t think about white bears.” Did you just think of one? According to a phenomenon called ironic process theory, being told to consciously suppress a thought generally has the opposite effect. In our experience, GenAI tools behave the same way – that is, GenAI struggles to interpret negative instructions (language like “don’t”, “never”, or “avoid”). Try using positive language to instruct the AI instead. |
❌ PROMPT: Don’t use complicated words in the output ✔ PROMPT: Use simple, easy-to-understand language in the output |
|
4. Give examples of the type of output you want: To help ensure that the output matches your expectations, you can include an example of strong output within your prompt. By providing a concrete precedent from which the GenAI tool can draw, you’ve established a reference point and increased the likelihood that the AI produces something in line with what you want. What’s more, you can give an example of a weak output to avoid. |
✔ PROMPT: Write an email subject line promoting an article about today’s threat landscape to an audience of cybersecurity professionals. [Good example] Threat intelligence report: Tracking evolving threats [Example to avoid] #ThreatIntelligence Report: Tracking evolving Threats!!! |
|
5. Ask the AI itself for prompt suggestions: No one knows AI like AI itself. So, if you’re stuck on how to write a particular prompt or want more specific advice on how to make a prompt stronger, you can simply ask your AI tool to do the work for you. |
✔ PROMPT: What sort of prompt would get [your AI tool] to edit down a passage of text so that the output is under 1000 total characters, while also retaining all of the key points? |
Other GenAI tips and nuggets of wisdom
Our team has also learned some valuable generative AI lessons outside of the realm of prompting. Here are some additional tips to help strengthen your GenAI output.
- Adjust the temperature: In the context of AI prompting, “temperature” refers to a parameter that defines the probability threshold that a given token will be generated next. In practice, this setting impacts the creativity of the output – a higher temperature will result in a higher degree of randomness and creativity, whereas a lower temperature will lead to more predictable results. When using a GenAI tool, play around with the temperature to find the sweet spot (at TechTarget, we tend to favor a temperature that hovers around .7).
- Consider the Large Language Model (LLM): Different LLMs exist – OpenAI’s ChatGPT, Anthropic’s Claude, Google’s LaMDA and many more. And different iterations of these LLMs also exist (ChatGPT 3.5 and ChatGPT 4, for example). Bear in mind that these different models can have nuances between them, may have been trained on different datasets, and may contain different limitations that may impact which LLM you want to choose for a specific use case.
- Match the prompt language with the language of the desired output: In our experience, tailoring the prompt to the specific language you’re working with generates the best results. For example, if you want an output in Spanish, write your prompt in Spanish.
- Reinforce your desired character length: Based on our testing, GenAI tools tend to adhere more strictly to a desired character length or word count limit if that limit is stated twice or more within a prompt.
- Be polite: As it turns out, respectful language isn’t just for the socially conscious. GenAI responds positively to “please” and “thank you,” too. It makes sense – after all, generative AI was trained on human conversation, and a polite question is more likely to elicit a polite and helpful response. So, remember to mind your “Ps and Qs” when interacting with a GenAI tool (a good habit to practice in general).
Making the case for “human in the loop”
In building your own generative AI strategy, it’s important to remember that generative AI is atool to be leveraged, and not a replacement for your human workforce. GenAI needs to work in concert with humans in order to arrive at the best possible decisions and to achieve your desired business outcomes. This process is called human in the loop (HITL).
AI is a tool designed to produce content – it’s not a substitute for market expertise, nor for human intuition. That said, one thing is becoming clear: AI is a great enabler of productivity and efficiency. So, while AI can’t replace humans, the workforce that uses AI can certainly replace the workforce that doesn’t.
Another major reason to consider HITL is due to the existence of AI hallucinations – a phenomenon characterized by the generation of false or misleading information. Some AI hallucinations are clearly nonsensical, but others appear all too plausible.
Here’s a real example of an AI hallucination we’ve run into at TechTarget, while trying to brainstorm potential subject lines for one of our Coffee Chats (a type of informal meeting with an industry expert). When fed the title “Coffee break: Experts talk DevSecOps,” AI suggested the following:
OUTPUT: “Is your coffee habit putting your company at risk?”
The GenAI tool latched onto the coffee part of “Coffee Chat”, and, ignoring the rest of the context of the request, suggested a title wholly irrelevant to what we were looking for. I wondered if this was a one-off hallucination, so I reran the prompt:
OUTPUT: “How to brew the perfect cup of coffee”
Apply a critical (human) eye when evaluating GenAI outputs – it can save you from headaches or even litigation down the road.
Though it comes with a learning curve, Generative AI is a powerful tool in the content marketing toolbox
GenAI can be used to support a variety of use cases, like subject line generation, formatting, translation and more. Reflecting on your most important GenAI use cases can help right-size your strategy. Be wary of GenAI’s capacity for producing AI hallucinations. For best results, we recommend a human in the loop approach and consistent prompt optimization.
