Talking to AI effectively – the power of the right words in your prompts

Talking to AI effectively – the power of the right words in your prompts

Cet article est également disponible en Français

Language models like ChatGPT, Gemini, or Claude have become powerful allies for writing, coding, classifying, and analyzing. But one often underestimated factor is the choice of words, especially action verbs, in your prompts.

This article is for content creators, developers, analysts, communicators, or curious individuals using AI to generate text. It answers a simple yet crucial question: which word should you use to get exactly what you expect?

We’ll explore how the choice of terms in a prompt can radically transform the quality of the response, structuring the article as follows:

  • Framing and origin of the problem
  • Three axes of reflection around lexical choice
  • Objections and limitations
  • Practical applications and broader implications

I. Framing: Why the Right Word is Essential

Definition and Origin

"Prompt engineering" refers to the art of crafting effective queries to guide a LLM (Large Language Model). While complex techniques (zero-shot, CoT, ReAct...) are often discussed, it all starts with simple words. LLMs are highly sensitive to phrasing. Between "summarize" and "synthesize," or "analyze" and "critique," the result can be radically different.

Concrete Tension

Imagine this prompt:

"Give me an analysis of this text."

And this one:

"Summarize this text by highlighting the main ideas and secondary arguments."

The first is vague. The second is precise, oriented, and structures the expected response. Everything lies in the verb ("summarize") and the object complements. The right word, at the right time, is the essence of an effective prompt.


II. How Language Drives the Response

1. The Power of Action Verbs

LLMs better understand prompts starting with a clear action verb. It’s no coincidence that Google recommends a list of verbs in its internal guides:

  • Classify (instead of "tell me what you think")
  • Summarize (instead of "what is this text about?")
  • Explain, Compare, Translate, Reorganize, Extract, List, Formulate, etc.

Each verb activates a specific processing mode. For example:

  • "Summarize" expects a neutral condensation.
  • "Explain" allows subjectivity.
  • "Compare" calls for a relational analysis.

2. Tone, Role, and Style: Underestimated Modulators

Beyond the verb, the tone of the prompt influences the output. Saying:

  • "Write an article" produces a generic output.
  • "Draft a dynamic and engaging blog post" refines the style.
  • "Act as a tech journalist and write an article" assigns a role to the model, enhancing coherence.

Word choice is thus closely tied to context, intent, and tone.

3. Strategic Lexicon: Table of Effective Verbs

Here is an enriched table with typical objects on which each intention and its effective verbs operate. It articulates:

  • The intention (what you want to do),
  • Effective verbs (how you do it),
  • Cognitive objects (what you act upon).
IntentionEffective VerbsTypical Cognitive Objects
SynthesizeSummarize, Condense, Structure, SimplifyIdeas, long texts, discussions, scattered data
CompareCompare, Differentiate, Oppose, EvaluateConcepts, options, approaches, results, versions
CategorizeSort, Classify, Group, OrganizeData, cases, examples, objects, profiles
CreateDraft, Generate, Invent, Write, Assemble, StructureContent (texts, stories, messages), formats (tables, diagrams), novel objects
TranslateTranslate, Rephrase, Adapt, Rewrite, TranscodeLanguages, registers, formats, target audiences
AnalyzeAnalyze, Decode, Interpret, Decompose, Deduce, EvaluateProblems, behaviors, structures, dynamics, discourses
ObserveDescribe, Measure, Identify, Spot, DelimitFacts, phenomena, signals, trends, metrics
ExtractList, Select, Retrieve, Extract, Parse, FilterRelevant elements, named entities, key passages, metadata
ArgueDefend, Justify, Critique, Contrast, Counter-argue, Demonstrate, IllustrateTheses, hypotheses, decisions, positions, reasoning

These verbs are cognitive "modulators" you apply to the AI.

⚠️ What if the AI suggested the right words itself?

This is the principle of Automatic Prompt Engineering (APE): you ask an AI to generate multiple prompt formulations for the same task. The goal is to automatically test several variants (e.g., "explain," "clarify," "summarize"...) and measure which works best. This approach is particularly useful for creating robust conversational assistants or reusable production prompts. APE takes the idea that the right keyword makes all the difference to the extreme.

Example of APE in Practice:

Objective: Obtain a short product description for an e-commerce page.

APE Prompt: "Generate 10 prompt variants to ask a LLM to write a product description for a rock t-shirt, size M."

AI Output:

  1. Write a catchy description for a rock t-shirt, size M.
  2. Provide a sales-oriented text to highlight this size M rock t-shirt for fans.
  3. Summarize the key points of this t-shirt for an online product page.

The user can then test all three, measure the impact on results (clicks, conversions, etc.), and choose the most effective prompt.


III. Common Objections and Perspective

Frequent Counterarguments

  • "AI understands even vague requests": true, but at the cost of increased uncertainty in the response.
  • "You can always rephrase": sure, but each iteration costs time, especially in production.

Measured Responses

Lexical precision isn’t a formalist obsession. It’s an efficiency strategy. Better to spend 30 seconds thinking about your verb than 5 minutes correcting an irrelevant response.


IV. Practical Applications and Broader Framework

Concrete Consequences

  • In Business: Briefing an AI clearly saves time and improves accuracy.
  • In Education: Training in explicit and structured formulation.
  • In Development: Standardizing AI calls with optimized prompt templates.

Knowing how to craft a precise prompt is becoming as essential as mastering internet search. It’s the language of human-machine collaboration.


Conclusion

The right word can change everything. In prompt engineering, every term is a lever. Choosing the right verb, clarifying your intent, adopting a precise tone... these are all steps to obtaining relevant, useful, and actionable responses.

Rather than blindly multiplying attempts, adopt a conscious, strategic, structured approach. Because speaking well to an AI is, above all, thinking well about your language.


🧠 Key Takeaways

  • ✅ Action verbs structure the AI's response.
  • ✅ A good prompt often starts with a clear and direct word.
  • ✅ Choosing your words is steering the AI's thought process before it produces anything.

Propels your product to the upper level 🚀

Do you need help to assess your needs and choose the best option for your product? Otterly is there to accompany you and propel your product to the market!