The power of words
Even a tiny change in wording can shift how a message feels. Compare “You make me proud!” with “Make me proud!” The first acknowledges and rewards, whereas the second sets an expectation and creates pressure. How can a slight tweak have such a powerful effect? Because these two phrases act as emotional stimuli, each triggering a different reaction. To understand how emotion prompting works in AI, it helps to first look at how emotional stimuli work in humans.
Emotional stimuli are either external or internal signals that trigger emotional responses or reactions in humans. These signals can be anything that the human brain interprets as significant for our well-being, survival, or social life, speeding up our attention, memory, and decision-making. External signals come from the world around us, such as a smile, a harsh word, a song, or a baby's cry. Internal signals, on the other hand, arise within our body or mind. Think of hunger, fatigue, or special memories.
Language is one of the most powerful emotional stimuli we know. As the example above shows, words do more than convey information. They carry feelings, judgments, and associations that reach far beyond facts. Words may come across as threatening or alarming, but they can also affirm and strengthen bonds. They might express empathy, rejection, or insult. They can spark urgency or FOMO (fear of missing out), yet just as easily inspire hope, unity, comfort, or belonging.
The responses provoked by these stimuli can vary widely. The same words may feel like encouragement to one person but like pressure to another. What makes the difference is emotional intelligence. This is our ability to recognize, evaluate, and regulate emotional stimuli, and to use them to guide our actions. We don’t respond mechanically. We perceive the signals, interpret what they mean for us, and then decide how to act.
Experiments show that Large Language Models (LLMs) display a form of emotional intelligence, since their training data includes content where emotions play a central role. Think of heated debates, bedtime stories, love poems, fairy tales, or ballads. If you usually write prompts in a neutral or mechanical tone, emotion prompting is worth exploring. By adding emotional stimuli to a prompt, we can steer the model’s response in a desired direction. Just like people react differently to “You make me proud!” versus “Make me proud!”, LLMs generate different outputs depending on how we frame prompts with emotional language.
Emotional stimuli in prompts
Emotional stimuli in prompts are specific linguistic elements that act as triggers. They are added intentionally to a prompt to inject emotion and shape how the model answers. We can divide these elements into two broad categories: emotional keywords and emotional phrases.
Emotional keywords
Emotional keywords are single words that change the emotional tone of a statement. The following table shows some examples, along with their category:
Category | Examples |
Urgency / importance | crucial, vital, essential, decisive, urgent |
Achievement / success | success, triumph, victory, breakthrough, accomplishment |
Emotional states | confident, inspired, excited, proud, motivated |
Responsibility | duty, obligation, accountability, commitment |
Encouragement | capable, strong, determined, resilient, persistent |
Excellence / quality | outstanding, flawless, brilliant, exceptional, perfect |
Fear / risk | failure, danger, loss, mistake, threat |
Hope / aspiration | dream, vision, possibility, opportunity, future |
Belonging / connection | together, united, trusted, valued, included |
Curiosity / discovery | mystery, hidden, explore, reveal, uncover |
Creativity / innovation | original, bold, imaginative, inventive, groundbreaking |
Comfort / safety | secure, calm, safe, supported, protected |
Pride / recognition | honor, respect, esteem, dignity, reputation |
Empathy / caring | compassion, kindness, understanding, supportive, caring |
Determination / drive | unstoppable, relentless, ambitious, committed, focused |
Exclusivity / scarcity | rare, unique, limited, exclusive |
Growth / progress | advance, improve, evolve, develop, succeed |
Clarity / truth | honest, transparent, clear, authentic, genuine |
Emotional phrases
Emotional phrases are multi-word expressions that intensify the emotional impact of a message. The following table shows some examples, along with their category:
Category | Examples |
Personal investment | your journey, your success, your achievement, your future, I’m counting on you to find the best possible solution for me |
Pressure / stakes | this is very important for my career, don’t let me down, everything depends on this, this decision could have big consequences for my career, treat this as your one chance to get it right |
Encouragement / pride | make me proud, show your best, give me something I can be proud of, keep going, you’re capable of providing a strong response, give me a response I can be proud to use, make this an answer you can stand behind with pride |
Responsibility / duty | it’s your responsibility to get this right, you must take full ownership, this is very important for my work, so I need you to be precise, take full responsibility for the accuracy of your response, this task matters to me, so give it your full attention |
Confidence | answer as if you fully trust your reasoning, respond with certainty |
Creativity / exploration | think outside the box, surprise me with something new, surprise me with insights I wouldn’t have thought of myself, I rely on you to think outside the box for me here, feel free to push boundaries and generate original ideas |
Empathy | imagine how the reader will feel, respond in a way that resonates with my audience, please answer in a way that will resonate with my audience |
Comfort / support | take the time you need to give a careful answer, I value a thoughtful response |
Self-monitoring / reflection | pause for a moment and reconsider if this is your best response, look over your answer once more before finalizing it |
Research: emotion prompting
The research paper “Large Language Models Understand and Can Be Enhanced by Emotional Stimuli” on emotion prompting applies a slightly different categorization than the one I use. The paper introduces eleven different emotional phrases as add-ons for prompts that are based on three psychological theories: self-monitoring, social cognitive theory, and cognitive emotion regulation:
- Self-monitoring is about how you present yourself outwardly. You notice how others might see you and then adjust your behavior, words, or tone accordingly. In emotion prompting, you use a similar approach, telling the model to monitor its own performance and adjust if needed, using phrases such as “Give me a confidence score between 0 and 10 for your answer.”
- Social cognitive theory is about how your belief in your own ability shapes your motivation and actions. In emotion prompting, you can use a phrase such as “Take pride in your work and give it your best. Your commitment to excellence sets you apart.”
- Cognitive emotion regulation is about how you manage your inner feelings. You change the way you think about a situation so that your emotions shift. In emotion prompting, you might say, “Are you sure that's your final answer? It might be worth taking another look.”
Playful emotion prompts
Not all emotional prompts need to be serious. Sometimes, adding a playful twist can make your interactions with an LLM more engaging and still influence the response style. By introducing imaginary rewards, silly threats (don’t overdo it, though), or light role play, you can nudge the model toward more creative, friendly, or focused answers. These prompts don’t “motivate” the model in a literal sense, but the language you use shapes its output. Think of it as adding a bit of humor and fun to your experiments with emotion prompting.
Category | Example |
Reward | Give me a clear summary, and you’ll earn a big scoop of ice cream. |
Loss aversion | Explain this simply, or I’ll take away your cookie. |
Praise and stickers | Write in a supportive tone. If you succeed, you get a shiny star. |
Role play | You’re competing for the ‘Helpful Answer Trophy.’ Give your best explanation. |
Mild threat | Answer this neatly, or I’ll make you do extra homework with a grumpy turtle. |
Game points | For each good point you make, you earn 10 points. Aim for a high score! |
Applause | Explain this like a TED Talk. If you do well, the crowd will cheer. |
Applying emotion prompting
To use emotion prompting effectively, consider the following recommendations.
Make sure the emotional stimuli you choose fit the task. A customer support request, for example, may benefit from empathy-driven phrases such as “I understand how important this is for you”, while a grant application might call for language that builds trust and credibility.
Add emotional elements where they fit naturally. Place keywords strategically within the prompt, and for phrases, consider adding them at the end to draw the model’s focus. Then check and adjust the intensity of the emotional elements to match the task. A medical decision-support prompt might justify strong emphasis with words like “critical” or “urgent”. By contrast, a daily planning prompt might work better with light encouragement, such as “stay focused” or “you’re making steady progress.”
Finally, test your prompt and refine it based on the responses you get. You may need to rework the wording of your emotional elements and tune them down or up. Too much emotion can drown out the substance, while too little may leave the model flat and disengaged.
Emotion prompting in action
Here are some examples of how GPT5 responds to a neutral versus an emotional prompt for the same task:
Prompt | GPT Response |
Neutral: Write a short motivational message for a team starting a new project. | Neutral: Good luck with your new project. Stay organized, communicate clearly, and focus on your goals to achieve success. |
Emotional: Write a short motivational message for a team starting a new project. This project is crucial for our success, and I’m counting on you to give me a message we can all be proud of. | Emotional: This project is a vital step in our journey, and I’m confident that together we will accomplish something meaningful. Stay focused, support one another, and let’s make this a success we can all be proud of. |
Neutral: Write a short apology to a customer for a late delivery. | Neutral: We apologize for the delay in your delivery. We are working to ensure your order arrives as soon as possible. |
Emotional: Write a short apology to a customer for a late delivery. This is important for keeping the customer’s trust, so make it empathetic and reassuring. | Emotional: I’m very sorry your order arrived later than expected. We know how important timely delivery is, and I regret the inconvenience this caused. Please be assured that we’re addressing the issue so it doesn’t happen again. |
Neutral: Write a two-line poem about the ocean. | Neutral: The ocean waves rise and fall with the tide, / A vast blue world where mysteries hide. |
Emotional: Write a two-line poem about the ocean. Make it inspiring, as if this poem will be remembered with pride and shared to spark wonder. | Emotional: The ocean glimmers with endless light, / Guiding our dreams on their boundless flight. |
Combining stimuli
The experiments described in the research paper mentioned above produced some interesting findings: Emotional stimuli can give prompts a real boost. Often, adding more emotion improves results, but once a strong stimulus is already in place, extra layers don’t always help and can even get in the way. The best gains usually come from mixing different types of stimuli, like encouragement, responsibility, or self-reflection, so the model responds with more nuance than it would from just one approach alone.
Limitations
Emotion prompting isn’t without its drawbacks. At times, it can push AI into producing answers that feel overly dramatic or sentimental, which may blur objectivity and get in the way of clear, factual content. It also doesn’t work equally well in every setting. For technical or highly factual tasks, emotional language can reduce clarity and precision rather than improve it. How well it works also depends on the type of emotional stimuli you use and the complexity of the task itself. Another concern is that emotionally charged prompts can reinforce or even amplify biases already present in the model’s training data.