November 12, 2024
November 12, 2024
Generative AI is a powerful technology that content and communications teams can employ for their research, editing, creative ideation and more. Tools like ChatGPT are disruptors and when used carefully can be true force multipliers – not the job-stealers some have fretted about. But with the rapid rise of generative AI comes the need for best practices to maximize its potential and strict guardrails to navigate its pitfalls.
We tested a few of the leading generative AI tools – ChatGPT, Jasper and Canva Magic Write – to understand their applications for content creation, public relations and strategic communications. Overall, we were impressed by the benefits they provide in multiple areas, from research and ideation to writing and editing. We’re sharing our findings here to help you get started with generative AI in your own organizations.
This is not an exhaustive guide, and ongoing experimentation will be important as improved iterations of these tools are released. In fact, just as people started to get acquainted with the initial ChatGPT release (version 3.5), OpenAI quickly issued GPT-4 with added features like image recognition. That will reportedly be followed by a feature reminiscent of incognito mode on a web browser to help assuage privacy concerns. That is to say: Generative AI is moving astonishingly fast.
Nevertheless, PR and communications professionals should be using these tools now to understand their strengths and apply them to their work where appropriate.
The Tools We Tested
We tested and assessed the utility of three popular tools: ChatGPT, Jasper and Canva’s Magic Write. Each is impressive in its own right and can help today’s communicators, creatives and marketers.
ChatGPT, with its basic user interface, is very intuitive and responsive and can distill complex prompts to deliver useful outputs in seconds. A key benefit is the ability to build a “dialogue” with the platform and turn prompts into a roadmap for brainstorms or further edits. While none of the copy these tools generate should be taken at face value, ChatGPT is navigating the current hype cycle with a determination to improve its product and make an impact now.
Jasper, a generative AI platform for business marketers, uses a template model that allows you to choose from formats such as social media posts, ad copy, blog outlines and product descriptions. The tool makes regenerating or editing outputs simple and is likely as helpful as prominent search engines. In addition to templates, it has a “chat” function – akin to ChatGPT – which can be equally helpful in generating content. We particularly recommend Jasper’s template for social media copy, as its structured outputs are helpful for creators and marketers in a hurry.
Canva, a broader design tool that’s already simplifying the creation process for users worldwide, makes its Magic Write AI tool very easy to navigate (it’s embedded as a clickable tool within a Canva Doc). Business and creative teams can use the feature to brainstorm ideas and draft items like social media posts, press releases, business plans, product descriptions, and more. In addition to the copy generated by Magic Write, Canva users can incorporate unique design elements on the same Canva Doc. This helps teams collaborating on projects requiring text and graphics. We used Magic Write to help us formulate ideas, research topics and understand the motivations of key audiences, technical and nontechnical alike.
<split-lines>"[Canva] helps teams collaborating on projects requiring text and graphics. We used Magic Write to help us formulate ideas, research topics and understand the motivations of key audiences, technical and nontechnical alike."<split-lines>
Writing Quality Prompts
One key point to call out for all these tools is the importance of a strong prompt. That means being precise with language so the chatbot can understand what you’re looking for. Word choice is important, so choose language that will steer the AI towards the desired outcome, and recraft prompts when needed.
Successful prompts, as we elaborate on below, depend on three critical factors:
- Provide sufficient information – brevity isn’t ideal. Prompt the AI with details you already know or understand about your topic.
- Clearly identify the audience you want to communicate with, which will impact the tone and voice of what’s generated.
- Always pose clear, direct questions that the AI can process and work from.
We recommend having several backup prompts prepared. You’ll likely need these options because out the gate the AI can have a pretty wide aperture, producing verbose or even unrelated content.
Ideation: AI as Your Thought Partner
Our tests led us to conclude that generative AI tools will not replace original ideas and thought leadership. The chatbots don’t always grasp the nuances of a topic and they often regurgitate existing thinking versus providing original insights. Still, the technology can produce blocks of informative research, analogies and text to help jumpstart the creative process. The tools also allow writers to create commodity content quickly and get a head start on pieces that require deeper critical thinking.
For example, we asked each tool to generate arguments specific to computer science and quantum mechanics, develop sample headlines for blog posts, summarize pain points for key vertical audiences, and test content structure across different social media channels. What we found in each case is that the technology can be markedly helpful with careful guidance.
One effective, thought-starter prompt for an AI chatbot might be:
For this example, the AI generated solid concepts that a user could iterate on or use to jumpstart a brainstorm. The prompt was effective because it posed a clear question with ample detail and identified the target audiences.
While its initial thinking and recommendations may not be groundbreaking, the AI copy is likely to spur further thinking or generate workable ideas. Used for ideation, users become more like “content directors” or editors – improving prompts until the output can stand alone or meet the desired criteria. Users intent on forming bold, original arguments may be disappointed – but they should use it as more of a sounding board.
It’s important to note here that these tools are generally trained on data up until late-2021. That means the chatbots have no knowledge of the world after that date. In addition, the tools generally continue to learn by storing enormous amounts of user data, raising intellectual property and security concerns. Be hyper-aware of this and develop organizational policies around what information can be shared with the tools.
Tailor Your Message to Different Audiences
Capturing “voice” to address the right audience is critical for communications and marketing. For instance, news moments and thought leadership may need to resonate with executives, engineers, consumers or other audiences, and each requires their own voice.
We found the tools to be powerful aides in this regard, able to convert text to suit multiple audiences. For example, you can feed ChatGPT an existing draft – such as a blog written for senior decision-makers – and request a rewrite that will resonate with developers or data scientists. Each tool can also simplify and explain complex technical concepts to make them accessible for non-technical audiences. We tried this to help brainstorm concise social media language that can still relay technical subjects and each tool provided workable options in a matter of seconds.
Ultimately, achieving the right voice depends on the right prompt. Here are a two prompts that worked well for targeting specific audiences and one that missed the mark:
- “Write five headlines for an op-ed targeting C-suite technical decision-makers about improving adoption of a new tech tool within your organization. Include five-10 potential media outlets that accept contributed content for this topic, and a rationale as to why those outlets were selected.”
- This prompt is successful as it contains ample detail, identifies a clear audience, and remains focused. The outputs here would match the level of detail in the initial query.
- “Outline the definition of ‘property technology,’ or ‘proptech’ and explain the benefits of its use for an audience of non-technical property managers/landlords. Write two analogies to more easily reach this audience: First, describe the type of ‘ally’ that software can be to those running rental businesses today. And second, describe the impact of vertical SaaS technology overall, and what it means for landlords long term.”
- This prompt is detailed, informative, structured and zeroes in on a specific audience/customer type for a proptech company.
- “List three content ideas for a real estate audience. Incorporate the largest industry trends since Q3 2022.”
- This is an example of a poor prompt. For starters, it’s too vague, extremely broad (the real estate industry is massive – with different stakeholders and audiences) and, unfortunately, too recent. That’s because ChatGPT’s engine only indexes web information stored by or before late-2021.
A Research Tool to Empower Scribes
In some respects, we view these AI tools as dynamic and interactive search engines, and they can really shine when they’re employed for research. Generative AI can achieve an incredible level of technical depth or help break down new/challenging subject matter – even on quick, simple “what is X”-type prompts. For communications practitioners constantly tasked with understanding new subject areas, this type of asset could prove as valuable as any search engine – with the important caveat that today, the information may not be current.
Here are three examples of effective research use cases:
- “What is U.S. Cyber Command, who is its commander, and what are some of the missions it runs both overseas and domestically? How might that have an impact on the future ‘health’ of the government’s cybersecurity standing?”
- “Describe the reasoning behind the supply chain bottlenecks that ensued after the outbreak of COVID-19. What sort of impacts have those bottlenecks had worldwide (including for international trade as well as commerce, e-commerce purchases, and pricing/inflation)?”
- “What are three areas of technology the White House has closely examined over the past decade? What have been some of the most impactful technological policies issued out of Washington, D.C., and what might we expect in terms of policy or regulations around Big Tech or cybersecurity in the 2020s?”
One important note is to always exercise caution. You need to verify outputs much as a dedicated staffer in a newsroom, copy editor or comms lead does for traditional content creation. For the most part, the tests we ran produced accurate information, but there is always a chance for errors. Make a habit of checking all informative/factual elements, much as you would with a platform like Wikipedia.
It’s worth repeating that ChatGPT and Canva’s datasets only index web information up to late-2021, with others, including Jasper, also not completely current. So, for users/researchers hoping to explore the latest news, events and information, these tools may not be helpful in their current form.
<split-lines>"Make a habit of checking all informative/factual elements, much as you would with a platform like Wikipedia."<split-lines>
Unlock Time-Savings for Basic Content Work
Our main takeaway here is that commodity content creation – particularly content based on previously documented information – can be comfortably written by generative AI. This creates real time savings (as much as 40%) for communicators and writers. These folks will be able to focus more on ideation, analysis and editing (think of a film director), versus what can be tedious research and initial drafting.
Left to their own devices, these technologies will not automatically produce colorful commentary and bold new arguments. We still believe that’s better left to human intelligence; but used intelligently and carefully these tools yield significant benefits.
Suffice to say that we are in the midst of a decades-long AI revolution that now seems only likely to accelerate. We hope this early guide is helpful for you, and happy experimenting!
