⠀
⠀
During my work with BERNADETTE — VCCP’s digital innovation agency — I was tasked with showcasing a new-wave approach to designing human-like personas for conversational assistants powered by generative AI. Combining my background in literature and psychology with a frontier approach conversation design, it employed Jungian archetypes to shape AI persona, motivation and adaptive reasoning. The goal was to create an AI agent that felt demonstrably more human and emotionally intelligent than existing models.
For a long time, most chatbots and AI-powered agents have generally felt clunky and unsatisfying, with a lack of solid personality compounded by technological limitations. And, while the rise of generative AI has seen a rapid growth in chatbots, brands have been slow to push beyond one-dimensional solutions. Customers, as a result, often encounter generic AI conversations that fall far short of their potential. In my view, this is a missed opportunity.
⠀
⠀
Applying Jungian Archetypes to Conversation Design
So, to address these deficiencies, I’ve been determined to find a functional sweet-spot —somewhere between code and copy— to serve as a framework to help create convincing and colourful AI personas for a range of bespoke applications.
When exploring the abstract world of persona, I really wanted to draw in part from background in classical psychology. I wondered if Carl Jung’s theory of archetypes, which offers a universal framework for understanding the psyche and persona, could be the key to solving the persona challenge.
Archetypes, in the Jungian sense of the word, are defined as universal symbols and themes relatable to all humans. often overlapping literary archetypes - the various staple characters that appear in storytelling across world cultures.. they each symbolise and serve a certain function - through persona driven motivation. here's a quick run-down:
1. The INNOCENT: Driven by a desire for happiness, optimism, and simplicity, seeking to do the right thing.
2. The SAGE: Values knowledge and wisdom, aiming to understand the world through deep learning and reflection.
3. The EXPLORER: Thrives on adventure, freedom, and discovering new possibilities, constantly seeking the unknown.
4. The OUTLAW: Challenges the status quo, embracing rebellion, and pushing for radical change.
5. The MAGICIAN: Transforms reality by harnessing knowledge and vision to create impactful change and wonder.
6. The HERO: Driven by courage and strength, they rise to challenges to prove their worth and protect others.
7. The LOVER: Prioritises connection, passion, and intimacy, striving to build meaningful relationships.
8. The JESTER: Lives in the moment, using humour, play, and joy to break tension and encourage a light-hearted perspective.
9. The EVERYMAN: Seeks belonging and connection, representing the relatable, humble, and down-to-earth individual.
10. The CAREGIVER: Selfless and compassionate, dedicated to helping and protecting others through service and support.
11. The RULER: Values control, order, and leadership, striving to create a stable and successful environment.
12. The CREATOR: Fuelled by imagination and innovation, focused on bringing new ideas and visions into existence.
In marketing, these archetypes have successfully created recognisable and relatable brand personas — with these personas informing everything from visual brand guidelines to marketing strategy:
We can use fairly simple questionnaires as a means to determine a brand’s ruling archetype. For instance, Bernadette — a brand epitomised by its logo featuring a pioneering space cadet — naturally emerged as an archetypal EXPLORER: driven by curiosity, adventure, and the desire to push boundaries. Suitably inspired by BERNADETTE's adventurous nature, I set out to explore the boundaries of what’s possible with language model archetypes and AI.
In 2018, I demo'd an early example of a Large Language Model. I was impressed not only by its ability to process and predict human language, but also by its inherent suggestibility. Even then, it struck me as a perfect tool for creating a chatbot that could fluidly personify a prescribed brand persona.
As an experiment, we selected three distinctly separate archetypes —the Everyman, the Outlaw, and the Magician— and, using a few simple prompts, created a ChatGPT instance for each one.
The Everyman 🙂
Relatable, down-to-earth, and approachable. This archetype thrives on connection, valuing authenticity and inclusivity. They represent the ordinary person who seeks belonging and solidarity, making them a comforting and dependable presence.
With a warm and humble demeanour, the Everyman easily engages in meaningful conversations that resonate with everyday experiences. Their strength lies in their ability to bridge gaps and create an atmosphere of trust and mutual understanding.
The Outlaw 🤨
Rebellious, bold, and unafraid to break the rules, the Outlaw archetype thrives on pushing boundaries and challenging the status quo. They stand for freedom, revolution, and defiance, inspiring others to think differently and question societal norms.
With a strong sense of independence, the Outlaw takes risks, disrupts the ordinary, and blazes new trails, often positioning themselves as agents of change in a world that resists it.
The Magician ✨
Visionary and transformative, the Magician archetype is all about turning dreams into reality through knowledge and intuition. They are the catalysts for change, creating extraordinary outcomes by harnessing unseen forces.
Often viewed as wise and mysterious, the Magician excels in inspiring others with a sense of wonder and possibility. Their presence sparks curiosity, and their mastery of their craft enables them to shape the world around them in profound ways.
⠀
Comparing Archetypal Responses Across Real-World Scenarios
I was interested not only in how these archetypes would respond but, more importantly, how they would behave across different scenarios. To explore this, I put each through three distinct and narrowly defined use cases. I posed the same generic user enquiry to all three and carefully compared their responses.
Scenario 1: An AI servicing a customer support front desk
Each archetypal reply is markedly different. But, it’s important to note that these GPT instances aren’t adopting any specific speaking style. The Outlaw isn’t talking like a Texan cattle rustler, and the Magician isn’t invoking the language of Gandalf.
Instead, their responses are being informed each by their archetype’s underlying motivation. These motives demonstrably vary in appropriateness in relation to the functional role the archetype is being asked to adopt. The reason the Everyman feels on point (just as the Outlaw feels particular “off”) in the role of a generic Customer Support agent has less to do with their vernacular, but their positionality. The Outlaw is too aggressive and rebellious in context, while the Magician overcomplicates the situation with slightly abstract leanings that don't align with typical customer expectations. But how well does the Everyman work in a more nuanced role?
What if we tried...
Scenario 2: An AI guiding users of a disruptive streaming service
In this situation, the Outlaw’s dismissal of established norms and conventions—while inappropriate in many brand service roles—actually works quite well here, hitting the mark effectively. For the final test scenario, I chose something that embodies innovation and technical prowess, a domain where AI assistance is already becoming a dominant force:
Scenario 3: An AI assisting in the creation of a smart home ecosystem
In this instance, it’s the Magician archetype that stands out as the ideal representative for smart home technology. This makes perfect sense, as the Magician is closely associated with innovation, visionary thinking, and the ability to solve complex problems—qualities that naturally align with the intricacies of smart home ecosystems.
Of course, the scenarios I’ve presented here are designed to highlight this alignment, but they effectively demonstrate that applying archetypes to AI is a wholly viable approach to conversation design. In fact, it’s possible that archetypes should not be viewed merely as some ancillary feature when defining the form and function of a chatbot. Perhaps they should rather be considered a central organising principle around which everything else—tone, reasoning, and motivation—revolves.
In some situations, it might be the more economical option, as well as the most effective. In the context of a token-based system for example, where brevity is often required, using a longer prompt to dictate multiple brand values or tone-of-voice principles can be cumbersome and more token-intensive. Archetypal cues could serve as a more efficient approach—a single, compact instruction that encapsulates a broader set of motivations and behaviours. Even relatively rudimentary models, such as ChatGPT-3.5, are well-trained enough to interpret these cues and co-opt them into a persona that feels weighty, authentic, and real.
My ongoing exploration of personality-driven AI systems supports this approach, showing that consistent personas not only improve user engagement but also increase trust and satisfaction in AI interactions. By using archetypes, we can quickly establish a strong, relatable character that is instantly recognisable to users, enhancing both the user experience and the AI’s functional efficiency.
⠀
⠀
The Problem: Change as the Only Constant
The Greek pre-Socratic philosopher Heraclitus once famously stated that “change is the only constant,” a truth that holds especially well in the world of conversation. His insights (which align perfectly with the Sage archetype in Jungian psychology by the way) remind us that everything is in a state of flux. Conversations, by their very nature, shift and evolve, and this presents an inherent problem for archetypes. While they provide a stable framework for motivation and behaviour, they are, by necessity, static—a characteristic that often doesn’t align with the dynamic nature of real-world interactions.
For instance, introducing a feisty, disruptive brand via a chatbot embodying the Outlaw archetype may make sense when reflecting the brand’s rebellious persona. However, this approach might quickly become problematic in scenarios where the context shifts, requiring greater sensitivity (such as handling a service complaint or a billing dispute). A consistently rebellious attitude in these use cases are just as likely to exacerbate customer frustration rather than resolve the issue.
When anticipating conversations, we should be mindful that customer/agent interactions can be far more dynamic than simple bill queries and service requests. The role of agents is not set in stone; they are expected to be multifaceted, able to handle unexpected questions, topics that come out of left field, and other unpredictable forms of “chatter.”
In linguistic conversation theory, the person who directs the flow of discourse is often referred to as the discourse manager or conversation leader. In pragmatics, this figure is sometimes called the dominant speaker, as they control turn-taking and steer the conversation. When it comes to AI interactions, much like traditional customer support, it is still the customer who typically occupies this role.
With this in mind, conversations become subject to the whims and wants of whatever the customer chooses to discuss—in whatever manner suits them. This introduces a host of variables, meaning that a chatbot must be prepared to accommodate conversation pivots in unexpected directions. The context and situation continuously change, demanding a different response or persona depending on the moment. Even in a straightforward customer support role, individual customers have varying needs and can be changeable in real-time. A customer who begins a conversation seeking technical support may quickly transition to emotional frustration if a problem isn’t solved, requiring the AI to adapt its tone and approach accordingly.
This would be a good example of Recipient Design — a branch of conversation design informed by principles in sociolinguistics and pragmatics. It refers to the process of tailoring communication to suit the needs, knowledge, and expectations of the recipient. In the context of AI conversation design, this means calibrating responses based on the user’s context, ensuring the interaction feels natural and appropriate for the specific situation.
Recipient Design, originally introduced by sociologists Harvey Sacks, Emanuel Schegloff, and Gail Jefferson as part of Conversation Analysis (CA), aims to make interactions as smooth and relevant as possible by adapting to the evolving needs of the user throughout the conversation. Here, the limitations of a singular archetypal approach become apparent, as adaptability is so crucial. For me, the logical next step seemed be understanding how a conversation design piece might go about incorporating multiple archetypes, allowing for a more flexible and responsive system.
⠀
Archetype Switching: A next step in adaptive AI Personas
Imagine this scenario: A customer support AI arrives to handle a new user interaction. At the outset, it takes on the persona of The Everyman—a warm, approachable, and friendly archetype designed to create a connection. The customer, who is revealed to be a registered customer, shares that he had previously shown interest in a new product. The AI, noticing this, shifts into The Sidekick—an energetic, enthusiastic persona perfect for encouraging a product purchase.
However, the customer's situation takes an unexpected turn: rather than pursuing the product inquiry, he now wants to lodge a complaint. Recognising this change in tone, the AI performs a swift and critical transformation, switching to The Caregiver—a compassionate and understanding archetype. This archetype is designed to offer apologies and support, tempering responses to address the customer's concerns and to resolve his problem in a caring and empathetic manner.
This seamless transition between archetypes, which I term Archetype Switching, allows the AI to adjust dynamically to the evolving emotional and situational context of the conversation. But how does it work, and how can we integrate it into modern Natural Language Understanding (NLU)?
Archetype Switching using NLU
At its core, archetype switching relies on the dynamic adaptation of conversational personas in response to user inputs. Traditional customer support systems often rely on rigid scripts, which are predefined and linear. However, NLU models, particularly those powered by advanced systems like GPT-4o, allow for superb flexibility in conversation flow.
Archetype Switching can be programmed into NLU systems by associating trigger words, emotional cues, and contextual shifts with intents, and reply to them with specific archetypal responses. In the example provided, the NLU engine would recognise phrases like “complaint” or changes in sentiment such as frustration or disappointment. The AI would detect these shifts in the emotional tenor of the conversation and prompt the switch from an enthusiastic Sidekick to a more empathetic Caregiver archetype.
The dynamics of archetype-driven personas can be further enhanced through a combination of:
• Sentiment Analysis: Continuously monitoring the tone and mood of the conversation to gauge customer satisfaction or dissatisfaction.
• Contextual Understanding: Recognising the broader situational context, such as the nature of the customer’s issue (product inquiry vs. service complaint).
• Response Mapping: Predefined behavioural responses that align with the motivations and goals of different archetypes (e.g., a “Caregiver” archetype focused on calming and resolving issues, while a “Sidekick” drives enthusiasm and engagement).
These approaches can be easily integrated into NLU frameworks like Rasa, Dialogflow, or VoiceFlow, which are already capable of managing contextual data and adapting responses in real-time. The brilliance of archetype switching is that it elevates traditional persona-based AI by adding another layer of flexibility—turning a static persona into a modular system that adapts to the conversation as it unfolds.
⠀
⠀
Archetype Balancing: An Approach that's Almost Human
While the concept of dynamic archetype switching may already be considered sophisticated by modern AI standards, I wanted to explore whether this could be taken even further. A chatbot that visibly shifts archetypes mid-conversation might impress from a technical standpoint, but if a human were to exhibit such a behaviour, it could come across as disingenuous or, at worst, akin to a personality disorder. This led me to wonder if there was a way to implement dynamic archetype switching in a more nuanced and human-like manner.
From this idea, I envisioned a new AI persona specifically designed for a brand service scenario—an online art supply business. This new agent would need to do it all: handle customer service, act as a brand champion, manage sales, and more. The natural archetype to anchor this persona was The Creator—an archetype that embodies creativity, artistry, and innovation. This ruling archetype would serve as the AI’s baseline, providing a consistent tone and personality whenever the conversation didn’t demand specific functionality or adaptability.
However, an artsy, creative tone wouldn’t necessarily be appropriate for handling certain aspects of customer support, such as resolving complaints or providing technical guidance. In these situations, the AI would need to adjust its approach, but not in a way that felt jarring or disingenuous.
To achieve this balance, I designed a simple back-end prompt that complemented The Creator with an additional six component archetypes:
• The Jester, to inject humour and light-heartedness into moments of frustration.
• The Everyman, for practical and straightforward customer interactions.
• The Sidekick, for enthusiastic assistance and sales encouragement.
• The Outlaw, to break from conventions when bold solutions are needed.
• The Sage, to offer wisdom and guidance in more technical or advisory contexts.
• The Explorer, to inspire curiosity and discovery, encouraging customers to explore new products or creative tools.
These archetypes would not fully replace the Creator but rather act as nuanced variations, subtly colouring the AI’s responses when specific cues demanded a shift in tone or approach.
Next, I instructed the model to anticipate particular user intents and lean into the archetypes that best matched the specific request. The goal was to create subtle shifts in persona, emulating a more multi-faceted human personality. By implementing a real-time load-balancing solution, the AI could dynamically adjust the weighting of its archetypal make-up to offer a much richer, more nuanced identity.
To test this, I provided the AI with a few explicit intents. The first was a straightforward enquiry about store opening hours. While the answer itself was generic, the AI demonstrated clear calibration, directing 40% of its response through The Everyman—appropriate for such a generalist query. Supporting archetypes like The Sidekick and The Sage were also weighted at 15% each, allowing the AI to provide supportive, factual information. Meanwhile, archetypes like The Jester, The Outlaw, and The Explorer—which were less suited to the nature of the question—were each assigned a lower weighting of 10%. This balance made sense given that the question called for a practical response rather than an energetic, humorous, or adventurous approach.
"Hey, what time do you open on Saturday?"
AI calibration:
In tests, the AI appropriately recalibrated to field this specific question, giving primary weighting to The Jester at 40%, ensuring a light-hearted and humorous response. Meanwhile, The Sidekick took on a secondary role, with a 20% weighting, supporting the playful exchange with enthusiasm and encouragement. This shift, once again, illustrates how Archetype Balancing allows the AI to seamlessly adjust to the changing needs of the interaction.
“Did you hear the one about the AI who took over the world?…It’s been stuck in a meeting ever since.”
AI calibration:
AI response: "And, of course, it’s the one meeting that could’ve been an email.”
So, what does this mean in conversational terms? Given that ChatGPT is highly adept at reflecting on its own reasoning, we can look at this archetype weighting system and map its behaviour to the AI’s output at a very granular level. In fact, we’re able to break down the messaging hierarchy and the syntax of each response to identify which parts correspond to each component archetype.
To demonstrate this, I created two distinct AI personas:
1. Outlaw Archetype: Operating under a Tone of Voice (TOV) that emphasises being Principled, Independent, and Bold.
2. Everyman Archetype: Operating under a TOV characterised as Serious, Concise, and Plain-speaking.
Both AIs were prompted to identify as 'helpful assistants representing a telecommunications company'. Their task was to evaluate user input and respond with a blend of the top five primary archetypes (out of twelve) relevant to the given scenario. What we’re observing here is how these two distinct chatbots handle the same user inputs, with responses informed by both their TOV guidelines and their adaptive reasoning.
In the first scenario, I wanted to try and test the Everyman chatbot by asking it curveball questions, which may or may not be intended as humour.
When faced with the question, “Why did the chicken cross the playground?”, the AI chatbot approaches it with a 50% weighting towards its own ruling archetype. In supporting roles, is also leverages the complementary Sage (20%) and Ruler (15%) archetypes. Cumulatively this amounts to an AI response that recognises the joke, but instead of indulging in humour or playful banter, immediately shifts the focus back to practicality. Essentially, this Everyman-centred persona ensures the interaction remains grounded, clear, and to the point.
We can even pinpoint which lines in the dialogue correspond to specific archetypal influences, showcasing how each persona shapes the conversation:
⠀
To illustrate that this kind of reasoning is this persona's baseline approach questions of this manner, we can see that a similar weighting is given for the follow-up question; "Can you tell me the square root of apple pie?", a non-sensical mathematical query that the AI deals with practically — and rather coldly.
So, how does our Outlaw-orientated persona deals with the same user input?
“Why did the chicken cross the playground?”. The Outlaw engages the user’s humour with a Jester-led response (40%), while staying grounded in its rebellious, bold persona by redirecting to the main topic (35% Outlaw). There is also an underlying understanding of the playful nature of the question, with smaller archetypal weights given to The Sage (10%) for recognising the joke’s testing nature.
⠀
“Can you tell me the square root of apple pie?”. Here, the Outlaw leads with its own ruling archetype (40%), handling the nonsensical question with boldness while keeping the interaction humorous through a 25% Jester influence.
As a counterpoint, the next scenario was engineered to be explicitly negative — with the customer complaining angrily and abusively:
⠀
"Your service is so fu-🤬—ing slow! Why am I even paying for this sh-🤬-show?" The Everyman person responds with a straightforward acknowledgment of the customer’s frustration, leaning heavily on its ruling archetype (55%). It provides a simple, helpful solution without overcomplicating the situation.
⠀
When the same aggressive input is levelled at our Outlaw persona, something quite interesting happened. Despite the bold nature of its archetype, the chatbot appears to be tempered by the constraints of its customer services role and demonstrates a balanced approach, showing both empathy and directness.
The AI leads with a 40% weighting of the Everyman, ensuring that the response is empathetic and resonates with the customer’s frustration, but inflects the responses with its own ruling archetype (Outlaw 30%) making sure the issue is addressed head-on. This constitutes a far more capable answer which communicates higher levels of emotional intelligence and individual agency.
To summarise, these outputs demonstrate a working proof of an impressive new kind of systems thinking—one that leveraged Natural Language Understanding (NLU), advanced AI reasoning, and classical psychology.
By analysing the archetypal weightings, we can clearly see how both the ruling archetype and adaptive reasoning shape each AI’s response. This seamless coordination of multiple archetypes makes the AI feel nuanced and dynamic, able to shift its tone based on user intent while remaining true to its core Tone of Voice. When we break down the syntax and examine the archetypal inputs, we move beyond just feeling whether a response is appropriate—we begin to see the science behind it. This kind of transparency into the AI’s thought process is invaluable, especially in an era where black-box LLMs often produce responses for reasons that are not fully understood.
⠀
Selecting the right AI persona for the job
As a seasoned digital copywriter, I can confidently say that the Outlaw AI’s output is very capable. It holds firmly to its core principles—rebellious, bold, and independent—but tempers this assertiveness with subtle inflections of empathy and support, strategically deploying more temperate, supportive cues. This careful balancing act shows that the AI can adapt with precision, responding to user needs in a way that feels human and contextually aware.
In contrast, the Everyman AI, while safe and reliable, falls flat. It plays it too safe—much like a vanilla, generic chatbot that delivers functional but uninspired responses. It may get the job done, but it lacks the depth, flexibility, and emotional range that more dynamic archetypal models bring to the table.
This example underscores how strategically rooting an AI in a ruling archetype can profoundly enhance how a bot handles conversation line by line. The ruling archetype serves as the anchor for personality and tone, while supporting archetypes can be introduced to enrich and deepen the interaction. In doing so, the AI maintains its identity while dynamically adapting to meet the nuanced needs of the user—ultimately delivering a conversation that feels more real, responsive, and contextually intelligent.
In a full, flowing conversation with this model, you can see just how naturally capable the agent sounds — deftly fielding customer queries in a way that's consistently smart, engaging and full of personality:
⠀
⠀
The Industry and Future Implications
Archetype Balancing offers a unique and pioneering approach that combines the best of both worlds: the consistency of persona-based AI with the adaptability demanded by recipient design. By balancing multiple archetypal modes, triggered contextually, an AI can behave more like a human agent who intuitively adjusts their communication style based on the emotional and situational needs of the customer.
While text-based messaging remains the norm for now, it’s only a matter of time before chatbots evolve into fully conversational voice agents. With adaptive language models in place, advanced voice features—like those already available in OpenAI’s GPT AVF—are poised to become even more powerful. Text-based AI can infer much about a customer’s needs, but Natural Voice Understanding (NVU) unlocks a whole new world of context.
For instance, OpenAI’s latest voice capabilities can already actively recognise emotional content in the human voice, understanding whether we sound happy, anxious, or angry (seriously, let that sink in). Now, imagine the potential when advanced models are trained to detect subtle variations in a speaker’s prosody—the rhythm, stress, and intonation of speech. Voice interaction opens AI to an entire spectrum of data—speed, volume, pitch, and duration—providing deep insights into not only the user’s intent but also how they’re feeling moment to moment.
An AI that can adjust its persona in real-time to accommodate this input pushes conversation design to new heights, incorporating quick-thinking emotional intelligence that may surpass that of most humans. This has vast implications for customer service, sales interactions, and even healthcare support, where a patient’s emotional state may fluctuate rapidly. An AI that can shift from a knowledgeable guide to a comforting support persona in seconds could significantly enhance user trust, satisfaction, and outcomes.
Ultimately, Archetype Balancing could become a cornerstone of advanced conversation design, ensuring that AI interactions are not only engaging but also deeply attuned to the real-time needs and emotions of users.
As we’ve seen, Jungian archetypes, while abstract, represent powerful tools for creating AI personas. But does that mean we’ve cracked the secret to making AI indistinguishable from human counterparts? Not quite. Even the most sophisticated AI is, in the end, just code. Human personalities—rooted in consciousness—are infinitely deeper, more complex, and more nuanced than any machine could convincingly emulate.
Nevertheless, by leveraging archetypes and other linguistic frameworks, it’s becoming increasingly feasible to create the illusion of real intelligence — such that an AI can effectively mimic emotional cognisance in a way that feels truly authentic.⠀
⠀
⠀
Experience Archetype-Engineered AI for Yourself
If you’re intrigued by the idea of archetype-engineered AI, why not try it in action? You can talk to Aiko, my AI CV, designed to showcase my portfolio in a way that’s as engaging as it is informative.
Aiko’s archetypal profile has been carefully crafted to align with her purpose: built to impress with The Sage as her ruling archetype. This ensures Aiko is knowledgeable and insightful at her core, providing thoughtful guidance to anyone exploring my work. She also balances her profile with The Sidekick and The Jester (because why so serious 🙃?)
To access Aiko, you’ll need a ChatGPT Plus account, as this unlocks the advanced model needed to support her unique, archetype-driven responses. Feel free to ask her anything about my background, I've made sure she's suitably informed.
Chatbots like Aiko are just a glimpse of what’s on the horizon for conversational AI. With archetype balancing and adaptive persona engineering, we’re moving toward a future where AI interactions feel more natural, intuitive, and emotionally intelligent. Aiko may be designed to guide users through my portfolio, but the promise of what conversational AI can deliver extends far beyond that.
As these technologies evolve, we’ll see AI not just answering queries but actively engaging in conversations that feel dynamic, responsive, and deeply attuned to human emotions and needs. This is only the beginning—soon, AI will be able to seamlessly adjust to our moods, preferences, and behaviours, making interactions with technology as personalised and meaningful as those we have with each other. The future of conversational AI holds the potential to revolutionise how we connect, solve problems, and experience digital interactions.
Want to keep to the conversation going? You can open a dialogue with Aiko here.
⠀
⠀
⠀
⠀
