How to talk to AI

Freelance writer Theo Green explains when to keep it conversational and when to get technical

 

Artificial Intelligence tools, like large language models, are continually becoming more capable as well as more widespread. As they do, people naturally begin to wonder: what’s the best, most effective way to interact with these systems?

 

Should we speak to an AI as though it were a human; polite, conversational, using "please" and "thank you"? Or should we treat it as a machine: using highly structured, precise prompts to glean the best information possible from the model?

 

The truth is that both approaches have their merits, and the optimal choice depends on your goals, context and personal style. One method isn’t always better than the other.

 

 

Human-like interactions: conversational and polite

There are several benefits of using a conversational style of prompts:

1. Ease of use and accessibility: Conversational language comes effortlessly to most people, meaning that writing prompts that use normal speech comes easily to most people. When you phrase a prompt as you would a question to a colleague or friend, you lower the barrier to entry, and therefore the understanding of the information provided, meaning it will be easier to read and digest. This approach lowers the technical level of the AI’s response, making it accessible even to users without technical backgrounds.
2. Context-rich queries: Conversational style prompts often include context, backstory, or clarifying details organically within the prompt itself, without a real need to tailor or sculpt the prompt into something more specific. For example, "Could you summarise last week’s marketing campaign performance and highlight which channels did best? Thanks!" conveys not just the task, but the broader scenario. This helps the AI generate more nuanced, context-aware responses which fit the situation and the requirements needed from the prompt.
3. Encouraging empathy and rapport: Using polite language, such as saying “please” and “thank you”, may not influence the AI’s algorithm greatly, but it sets a tone for the user. Treating AI like a human, with courtesy, can help the user write prompts that enable the AI to provide a more “human” answer, rather than just delivering raw technical data or an unnuanced description.
4. Flexible exploration: Conversational prompts invite follow-ups and dynamic exploration. You might start with, "I’m not sure how to approach a competitor analysis; can you help?" and then refine your question based on the answer. A dialogue that feels realistic and dynamic will encourage you towards iterative exploration. It can also help you to zero in on the outputs you want the AI to deliver, enabling you to create better and more effective drafts and outlines.

There are several instances when using a more conversational and casual tone is effective for AI tools. These include:

• Brainstorming and ideation: When you want creative, free-flowing ideas, such as story concepts, marketing slogans, or design inspiration, an informal style helps the AI tap into broader associations. It also allows the AI to avoid casting out things that you might have found helpful, even if the AI didn’t.
• Learning and tutoring: Students or novices asking questions in a friendly, conversational tone are likely to receive explanations at an appropriate difficulty level, with analogies and examples that they can understand, rather than a wall of hard-to-understand content.
• Customer support chatbots: Businesses deploying AI-driven chat interfaces often benefit from a warm, conversational tone, enhancing user experience and perceived empathy, as well as clear, concise, and easy-to-follow replies to their questions.

 

Machine-like interactions: structured and precise

There are several benefits to more machine-like prompts. When you want specific, detailed answers to something, and you already have a good deal of background knowledge, a structured and precise prompt will help, as it will when you are asking a complex question that requires a precise answer.

1. Predictability and consistency: Highly structured prompts, often using complex “prompt engineering” techniques, use clear instructions, bullet points or proven templates to eliminate ambiguity. These prompts provide the AI with a rigid set of rules to follow, enabling it to know exactly what information to include and what to disregard. This can generate a more precise answer that is more exactly tailored to the question. It enables the AI to be far more precise in its output.
2. Efficiency and speed: Structured prompts can deliver high-quality results in a single pass if designed correctly, reducing the need for multiple edits or clarification rounds. This technique can be excellent for drafts or outlines, but it requires very specific and well-worded prompts.
3. Reproducibility: By codifying prompts into templates or standardised formats that can be used when some key words have been added in, teams can maintain consistent voice, style and structure across documents, something that can be critical for branding, compliance or reporting.
4. Complex, technical tasks: For tasks that require precision, such as code generation, data analysis instructions or policy drafting, structured prompts allow the AI to be guided. The precision enables it to follow exact specifications, reducing the risk of misinterpretation and increasing the ability the AI model to be concise.

Structured prompting can be of pivotal use in several situations, such as:

• Professional documentation: Crafting legal briefs, technical manuals or compliance reports where structure and accuracy are paramount, and a certain level of background knowledge on the subject is expected.
• Data-driven tasks: When requesting summaries of datasets, statistical analysis or when using data-specific templates (e.g. financial models), a precise prompt ensures completeness. The results are more likely to be correct and accurate, increasing the chances that they can be relied upon.
• Coding and debugging: Asking the AI to generate or fix code often works best with explicit directives, test cases and error messages. AI still isn’t perfect for code writing or debugging, and its outputs will generally require a human to double-check, but it can still be a useful tool.

 

Writing good prompts

Be flexible

The dichotomy between conversational and structured prompting is not absolute. In practice, many interactions benefit from a hybrid strategy, using one method then the other to tailor the response of the AI to what you really want it to be.

 

One useful technique that uses a hybrid strategy is to begin conversationally and then use more structured prompts. Start with a broad, friendly prompt to establish the context and goals of your request. Then, once you receive an initial response, switch to structured follow-ups, aiming to refine length, tone or format. Finally, explain the ideal format of whatever it is you are creating.

 

Another technique is to use layered prompts where you embed a structured outline within a conversational message. For instance:

 

Hey AI, could you help me draft my newsletter? I need: 

1. a 50-word intro;

2. three bullet points on product updates;

3. a closing call to action

 

This allows less rigid prompt crafting, but still gives the AI model enough information to get an accurate idea of the task you want it to perform.

 

Use best practice

There are several tips you should remember when writing prompts:

1. Know your objective: Think deeply about what you want to achieve. For instance, identify whether you need creativity, concision, detail or consistency. You can then align your prompt with your objectives
2. Use context wisely: Even a structured prompt can include background context in a few sentences. Context informs the AI’s choices without sacrificing clarity. It can be pivotal in making sure the AI knows what to include and what to disregard in the response it gives to you.
3. Be explicit about tone of voice: If politeness matters, specify it. "Use a respectful tone with please and thank you". If it doesn’t, then "Be direct and concise" may be a better prompt. Sharing information about the tone needed can save a lot of time editing later on.

Iterate based on results

View AI interaction as a dialogue. Don’t expect perfection on the first pass. It’s essential to be able to evaluate the responses you’re getting from the AI. Use follow-up prompts to adjust and tailor the draft you’re working on until it fits with what you were looking for. If the responses are too generic or off-target, refine your prompt and use more specific language and precise instructions. If they’re too rigid or lack creativity, loosen the structure and add conversational flair or ask the AI to adjust the tone to a more friendly style.

 

Remember: it’s not alive!

AI lacks genuine understanding. Human oversight is almost always necessary to catch biases, factual errors or inappropriate content. However precise or structured you make your prompt, the AI can always miss something, misunderstand your intentions, or analyse data incorrectly. Outputs need to be double-checked.

 

 

Human or machine?

There is no right or wrong way. You can talk to an AI like you would a human, conversationally and politely. Or you can treat it as a machine, using strict, structured prompts. Which course you take should depend on your goals, context and preferences. And while both approaches have their strengths and weaknesses, they can often be used together.

 

By understanding the strengths of each approach and combining them where appropriate, users can, with some practice, unlock the full potential of AI systems.

 

---

 

Theo Green is a freelance writer

 

Main image courtesy of iStockPhoto.com and demaerre