Prompting with Purpose: How Strategic Narrative Will Change the Way You Use ChatGPT
Yes, you can write ChatGPT prompts that tell your brand’s story. Here’s how.
by Harry Meadows
If you’ve done any scrolling on Twitter or LinkedIn recently, chances are you’ve felt like the ground is shifting under your feet. Since it launched in November, 2022, the potential of OpenAI’s ChatGPT to transform the way we work and learn has sent shockwaves through public and business discourse. As a result, it’s become more apparent than ever that the marriage of human ingenuity and innovative technology are essential for any organization to thrive. Whether you like it or not, generative AI is here to stay, and it’s only going to get more powerful.
And ChatGPT is just the start. In the last few days alone, Google announced that it’s ending the waitlist and expanding the capabilities of its own virtual assistant, Bard, to compete with ChatGPT. Microsoft has expanded access to Bing Chat, which, like ChatGPT’s paid version, is powered by GPT-4, the most advanced and capable large language model as of May 2023.
As AI tools proliferate, the sheer volume of content produced will explode. If you think you have an overflowing inbox now or too much content in your LinkedIn feed, wait until most of it is AI-produced. It will be easier than ever for a great company to get lost amid all the noise.
Woden has been working with GPT-4 to develop strategies that leverage the abilities of generative AI, while staying true to our belief that brands must tell an authentic story that makes them essential. We’ve largely used ChatGPT, but much of the analysis and advice that follows will be applicable to other platforms as well.
One of our key takeaways: the nature of generative AI means it’s more important than ever for companies to define their strategic narrative, and align themselves behind that message. Using AI tools in a way that is consistent with your brand’s story and unique voice will be critical to engaging prospects and customers in the future.
We’ve gone beyond the incessant hype to explore the possibilities and limitations of generative AI and uncover how to leverage ChatGPT’s capabilities to tell your brand’s story. Here are the ways common prompting strategies fall short, and how to apply strategic narrative to your ChatGPT prompts.
Strengths: what generative AI is already great at
In order to understand how to prompt ChatGPT into becoming a competent storyteller, we need to understand what it actually is, and what its limitations are. There are some things ChatGPT is great at even without exacting prompts:
• Delivering pure volume of options and text: where it might take a creative person hours to come up with 15 usable calls-to-action or taglines, ChatGPT can generate hundreds of options in minutes — freeing up the same human’s time to instead use elements from those outputs, draw inspiration, and make editorial decisions
• Writing and checking basic code
• Summarizing articles: even if you don’t tell it what to look for, ChatGPT can take long, complex, even technical texts and create bulleted executive summaries
• Copyediting for grammar and spelling, or even style
Answering simple questions and breaking down established concepts: if you asked ChatGPT to summarize the plot of King Lear or The Challenger Sale in eighth grade English the outputs are likely to be accurate and useful — although only to a non-expert
Limitations every user should be aware of
ChatGPT’s very real limitations become apparent when you start looking for it to provide specific outcomes. ChatGPT’s outputs can surprise and delight as long as we remember they were produced by a computer, but they are unmemorable by human standards. Anyone using it in a general, unfocused way will end up with a content mill for generic, mediocre writing.
• ChatGPT’s writing can be dry, generic, and lacking in substance: it produces sentences that look good at first glance, but whose meaning fall apart when subjected to any level of scrutiny
• It has a limited sense of word economy and style, excessively deploying adjectives, adverbs, and the passive voice
• It recycles common phrases, especially in contexts like being asked to write marketing copy — ChatGPT will almost certainly try to say “Unlock the power of [your product here]”
• Sometimes it lies, becoming an unreliable narrator capable of making you look foolish if you don’t fact-check
• It constantly reformulates concepts to more awkward grammatical forms (for example, in our testing GPT-4 changed “essential” to “essentiality” so often we had to explicitly tell it not to)
• Its default tone and voice dwell in the uncanny valley: outputs look familiarly human, but the more AI-generated content you read, the easier it is to identify
There are structural reasons for all of this: ChatGPT has been trained using massive amounts of recorded human conversations and written text to trial-and-error its way to a functional understanding of syntax and semantics. Hamstrung by the fact that it’s not actually intelligent, ChatGPT applies its training along with the conversational inputs a user provides to predict the next word and then the next phrase based on the associations those words and phrases have with one another in the training data. This is a completely different process than deciding what to write based on context, understanding, and experience, as humans do.
The reason ChatGPT needs this backdoor into language is that it can’t actually comprehend anything it reads or writes. It just passes every input through the model of language it acquired during its training to make relevant predictions. Because it is trained to predict based on a large volume of real language, its responses will reflect a grammatically and stylistically clean version of the most average and commonly-used expression of every idea and phrase: a regression to the mean.
The prompts people are using: are they doing enough?
Whether or not you believe that AI is sophisticated enough to use in your business, the fact is that your competitors — and everyone else — are already working hard to find ways to streamline workflows, brainstorm, and write content and copy using ChatGPT. That’s why a litany of voices offering similar advice and a cottage industry of people selling simple prompts have sprung up. Unfortunately, these approaches tend to fall victim to the same limitations we’ve already mentioned: the outputs are deeply average — when you need to be anything but.
Most prompts Woden found in free and paid libraries were brief, and usually offered only one or two clarifying rules or pieces of guidance for ChatGPT. Here’s the results to one available from a free resource for social media messaging:
ChatGPT’s output is predictably generic, even cringeworthy. It followed the instructions, but used phrases and structures that make your eyes glaze over: “Attention all…”, “Look no further…”, “Ready to take [BLANK] to the next level? Do [BLANK].”
Ironically, this instagram post does the opposite of what it’s advising our prospect: it uses a structure that’s generic because it’s commonly used, so not only is it unlikely to stand out, it captures nothing unique about the brand’s own story. Notably, even instructing it to be “captivating” doesn’t exactly get the job done.
Even a (very slightly modified) prompt from a popular paid repository falls short:
Your [product/service] can evoke powerful emotions in your [ideal customer persona]. Develop a [type of text] that leverages storytelling techniques to make them feel [emotion] and persuade them to take [desired action]. Use vivid imagery and strong language to make your message more impactful.
This prompt is more sophisticated than the first one: it includes a space to enter the job title from your buyer persona, and encourages ChatGPT to use “storytelling techniques” and drive a conversion by being emotionally compelling. That’s promising! We used an example of a common type of B2B software platform, and identified a specific buyer persona.
But this output still falls flat. ChatGPT makes several attempts at an emotional appeal — mentioning joy, empowerment, feeling supported — but can’t string those concepts together. It uses what can charitably be described as “storytelling techniques” but fails to map them to a substantive customer journey, or real pain points. Most importantly, it can’t do what effective stories do: move and persuade by making the brand feel essential. What’s even worse is that the voice and tone are just off. Has even the most dedicated HR guru ever felt thrilled when working with employee benefits software?
These types of prompts are useful in a limited way — and they’re easy to write. The outputs can be improved by carrying on a conversation with ChatGPT until it produces text you feel comfortable using or an idea you want to iterate on. But that defeats the purpose: in the time it would have taken you to just write a cold email yourself, you spend an undefined period of time arguing with a machine to generate one email, or one LinkedIn post that doesn’t really match your brand’s voice, and that certainly won’t express its story.
Why prompt with brand story?
ChatGPT exemplifies the universal principle of computing: garbage in, garbage out. Woden was frustrated with the outcomes of simple ChatGPT prompts like the ones we tried above. When it comes to marketing and sales conversions, the content that you put in front of your prospects needs to be decidedly not garbage: it must tell a compelling story that positions your brand as essential.
Storytelling is the oldest and surest way of communicating complex, nuanced ideas. A well-told story reveals a truth that can’t emerge — or stick in peoples’ minds — in any other format.
Your brand story tells people who you are, and shows them why they should care. A great brand story centers your customer as its hero, builds urgency around their pain points and struggles, and positions your product or service as essential to solving their problem — enabling an efficient sales funnel.
We figured that there must be an attainable and useful middle ground between hyper-generic prompts and a fully custom chatbot — a way to prompt ChatGPT strategically and push its limits. So we set out to define and then build the prompts and sequences of prompts that could get ChatGPT to write and brainstorm great story-driven content.
Outputs from ChatGPT that meet our standard do the following:
• Talk about your company in a distinct way that shows why it’s essential. Message discipline is key: ChatGPT will have to know the core belief that drives your company, team, and product, and hammer that message home at every opportunity. Beyond that, it will need the context to articulate, in a way that keeps the customer at the heart of the story, what makes your product great, and what it makes possible by resolving their pain points.
• Identify the right customer, and speak to their pain points. A cold email won’t convert unless it’s written for a specific prospect. Nail down your buyer persona, and the pain points come easily. Getting specific about this is key: messaging that’s too focused on technical features and benefits can mask a product’s true value; a great product and team won’t sell if customers see them as a nice-to-have.
• Speak in your brand’s unique voice. Some brands write like Vice or Pitchfork: casual, direct, and a bit smug. Others sound more like the Wall Street Journal. What matters most is that voice and tone align with your brand’s story (and are unique to you). This is the ineffable side of brand personality, and can be tricky to replicate, so relevant examples are key when prompting ChatGPT.
ChatGPT doesn’t understand your vision, and it can’t empathize with your customers. A defined, strategic brand story centers on your customer, and describes a journey that transforms their pain points into a state of resolution through your guidance and their use of your product. This is not only key to your company being seen as an essential brand — it’s also the context you need to give ChatGPT in your prompts.
The story of every essential brand follows the same approach. Not only is it the framework for an efficient sales funnel — it’s the foundation of great ChatGPT outputs. Download our free guide on how to craft your strategic narrative.
Getting useful, story-driven outputs from ChatGPT is a game-changer: it mitigates generative AI’s limitations as a producer of volume and summaries, and helps it reach its potential as a true extension of your business development team. We built a prompt around our own brand story, then asked ChatGPT a simple question: “Why does every company need a story?” Here’s what it said:
Everything it said above is aligned with our story: it stuck to the message, and even made some substantive points about the importance of story as a “strategic compass” (its term, not ours) — and it did it faster than a human Wodenworker could.
Here’s how you can build a prompt around strategic narrative that gets authentic outputs
Set specific expectations with a great “system prompt.”
One of ChatGPT’s defining features is that it can carry on a conversation that refers back to instructions and context from previous messages. Microsoft recommends using a “system message”, a set of instructions at the start of the prompt that instructs ChatGPT to roleplay a more focused kind of virtual assistant. For example, “You are an AI who summarizes historical events in the poetic style and format used by John Donne. Tell me about the Battle of Shiloh in 14 lines.”
We had success using a “system prompt”, which stands alone at the start of each new conversation with ChatGPT — usually ending with a yes/no question, for example: “Do you understand?” Right now, our working system prompt directs ChatGPT to “become” Woden, although it works for any brand. The prompt articulates key elements of Woden’s StoryKernel — our brand story — expressed to ChatGPT as qualities and descriptions of the world in which it operates.
We use the Hero’s Journey narrative arc to construct StoryKernels; not only is it the platform for the most powerful and memorable stories, but it creates a focused message — one that maps almost perfectly to the customer journey. We made sure to build it into our system prompt in a way that would encourage ChatGPT to actually use storytelling in its outputs.
Here’s an example of how we integrated a paragraph of our strategic narrative that specifies the details of our product, instructing ChatGPT to position it as a useful gift that will help our clients achieve sales funnel efficiency and a great framework for messaging:
The StoryKernel is perfectly formulated to prompt ChatGPT. Its tight structure and alignment with our customer journey makes it easy to build prompts, and is simple enough for ChatGPT to use. All StoryKernels use the same framework: so any Woden client, or anyone with a well-defined story, could build a prompt like this one very easily.
Here’s what we included in our prompt:
• Our strategic narrative (StoryKernel), mapped on to the customer journey. We set it up in the prompt to map on to a general customer journey, which means that you can ask ChatGPT to focus on different sections of the strategic narrative depending on which stage of the sales funnel a prospect is in. Crucially, this already includes a highly specific buyer persona, including the motivations and pain points of our clients.
• Our mission statement, and a short explanation of its meaning.
• Our story-aligned brand personality words, how we define them, and an explanation of how they should influence ChatGPT’s outputs.
• The names of two well-known publications that we draw inspiration from for the tone and voice of our writing, and examples of representative text taken right from those publications.
• Some basic rules for ChatGPT to follow in every response. You’ll find that some of these are necessary to encourage robust use of brand story, while others will just draw hard lines on styles you find yourself consistently clarifying while talking to ChatGPT. For example, one of ours is “Woden does not compose sentences in the passive voice more than ten percent of the time.”
If, like Woden, your organization has a written articulation of its strategic story and brand pillars, that story should be in your system prompt, serving as grounding data for any and all outputs you ask ChatGPT for.
Want the prompts we used to turn ChatGPT into a brand storyteller? Send an email to firstname.lastname@example.org, and our AI will respond to you.
How this changes ChatGPT’s outputs
We noted above that one of ChatGPT’s best uses is in brainstorming large volumes of content. Its ability to do this in a focused way (where you’ll throw out fewer of the outputs) improves dramatically with a good system prompt. We asked in a blank, new chat, for some options for the title of this piece:
Here’s what the outputs look like after entering the system prompt:
Here’s ChatGPT’s effort at some basic email copy without the system prompt — only using a brief system message, followed by the instruction.
Here’s what it came up with in a separate chat where we did enter the system prompt:
The instructional prompt for both of these outputs were nearly identical, but in the second screenshot the output is much more substantive, and concise: it’s on-message, mentions specific outcomes, and messages to a clear pain point. It provides a good structure and outline, as well as some phrases that you could tweak and then build into your email copy. In a pinch, you could make a couple of quick edits and send this out.
Be specific in your instructions, and handle each request with a focused, goal-directed conversation.
The more ChatGPT knows what to expect — and what you expect of it — the better the outputs will be. For example, specify a format, and word count when asking ChatGPT to write content; a prompt to “write some boilerplate copy” may send ChatGPT off on a hallucinatory 500-word odyssey, whereas instructing it to write a 50-word LinkedIn description will lead to a focused, useful response.
One good way to think about this is to ask yourself if there’s anything you’re already expecting from a response before asking ChatGPT. If you’re writing for an executive audience and need it to be economical and professional, tell it exactly that — and provide an example.
If you have a well-written strategic narrative that works in your brand’s voice, you can use that as an example for tone and voice. It’s even better to give ChatGPT an explicit example, either from a published piece of your own content, or a publication that has a similar tone and voice to your brand. Providing examples like this is known as few-shot learning — and it works even better if you present the examples as a dialogue between you and the AI.
Prompt in sequence, building on a foundation of strategic narrative.
The best part about using a strategic system prompt is that ChatGPT will apply the focused training you’ve provided in a deeper sequence of prompts — which you can then use to craft, copy or brainstorm very quickly, without needing to edit heavily for your brand’s voice, tone, or narrative.
Using a well-constructed system prompt frees you from having to provide deep context and tactical instructions at the same time, which can confuse ChatGPT. To test this, Woden challenged ChatGPT to write personalized cold emails to our prospects, consistent with our brand story, at scale.
We began with the system prompt to make sure that ChatGPT wouldn’t need further coaching on strategic narrative and brand personality in the more tactical prompts that followed.
The next step was to perform a trick that we’ve only successfully used with GPT-4: asking ChatGPT to prompt itself. Here’s what this looked like:
We won’t show you the outputs that ChatGPT produced with our real prospects (we were successful in returning up to 20 personalized emails at a time), but here’s one with a fictional persona in the CSV format we used in the prompt:
The response is much better than we expected. Leaving aside the fact that GPT-4 can prompt itself, the result pulled the data perfectly from the text-converted .csv file, personalized the best it could, and stayed on message, which is critical when the word count has to be so low.
ChatGPT is capable of generating serviceable, story-aligned messaging with minimal effort needed to write new prompts. In this case, once the prompts were perfected, all we’d need to do would be to format and export the original spreadsheet, then paste it into ChatGPT.
Using ChatGPT (or Bard) is very different from the other software solutions you use in your business — there are few buttons to click, and no dashboards to view. ChatGPT is designed to be conversational, but has meaningful limitations.
Think of it as your most literal employee: it’s not especially creative, can’t really think or make decisions by itself, and needs a lot of instruction. You have to intentionally train it on the specifics of your company’s strategy and narrative so it can stay aligned with your mission and vision — just as you would a newer member of your team.
But its writing is solid, and when on-task, saves the time and energy of the rest of the team so that they can focus on doing higher-level work and driving your business forward.
When striving to become an essential brand, the last thing you want your written content to be is generic. Generative AI can help your brand tell its story — but only when its outputs are aligned with your strategic narrative.
Velocity is both speed and direction: the power of generative AI to get your brand moving faster is exciting — but you have to know where you’re going first. Strategic narrative sets the direction — the right ChatGPT prompts will accelerate you towards it.
Harry Meadows is the StoryAccelerator Lead at Woden. Want to stay connected? Read our extensive guide on how to craft your organization’s narrative, or send us an email at email@example.com to start the journey to uncover what makes you essential.