- Define tone, structure, roles, and level with instructions and adjustments for consistent and useful responses.
- Specify sources, ask for citations, and set dates for rigor, currency, and less wishful thinking.
- Enable native personalization and control memory and data to balance continuity and privacy.
Furthermore, the techniques you'll see here also work on other assistants like Google Gemini, Copilot, Grok, Meta AI, Perplexity, or DeepSeek, so you can apply the same principles and polish any modern chatbot without having to reinvent the wheel.
Stop talking to it like it's a search engine.
ChatGPT doesn't index pages like Google or respond to random keyword combinations: it's trained to understand natural language , so expressing yourself as you would with a person usually yields better results.
Forget typing telegraphic phrases like "best laptops 2025 battery life" and try a complete request: "I need a lightweight laptop for travel, with good battery life and a budget of €1.000" ; the model has more signals to reason and adjust the response.
Be specific and provide context.
A precise question avoids ambiguity: the more specific you are, the better . Don't ask for "a formal email," ask for "an email in English to apply for an analyst position, with a professional tone, highlighting experience in data and immediate availability."
When searching for information, specify the scope, model, date, or version . “Explain the difference between iPhone 15 and 15 Pro in camera, battery life, and materials” is much more helpful than “iPhone differences.”
The details make the difference
In text and image creation, the nuances change the outcome . If you're designing a visual scene, including aesthetics, colors, era, lighting, or clothing helps to establish the style; in text, it indicates the target audience, length, and desired format.
You can also specify which part of a topic interests you: “Tell me only the key milestones of the event, no opinions, maximum 8 lines .” This level of precision saves iterations and reduces noise.
It requests specific formats and structures
If you have a specific way of presenting the information in mind, state it explicitly: structure, style, and order can be enforced by requesting them in the prompt.
- Comparisons: “Pros and cons of each model in vignettes”.
- Academic: “short format with introduction, development and references”.
- Guides: “Numbered steps, notes and warnings at the end”.
- Executive Summary: “5 actionable bullets for steering”.
Adjust the level of explanation to your profile
If an answer seems too difficult, ask it to lower the level: “explain it to me as if I were 10 years old” , “for a beginner” or “with simple metaphors” forces the AI to change terminology and examples.
You can also go to the other extreme: “enter at a C2 technical level and include formal notation where appropriate” ; the key is not to resign yourself: ask for the version that is most useful and clear to you.
Sources: types, institutions and citations
When reliability matters to you, guide the source of the information: indicate the type of source (“academic research and peer-reviewed studies”) or cite specific organizations (“according to the WHO” for health, “according to AEMET” for weather) or media outlets.
Furthermore, it requests traceability: “include references and citations in your response .” Although the template may not always link to the full text, the simple act of requesting references improves rigor and the quality of the argumentation.
- Font types: academic, institutional reports, specialized press, official documentation.
- Institutions/media: AEMET, WHO, universities (e.g., Harvard) and public agencies.
- Quotes: mention authors, year and, if possible, link or DOI.
Recent data and verification to avoid hallucinations
ChatGPT can query the web, but it doesn't always do so by default . It explicitly requests current data and limits the search time: "best rock albums of 2024" or "trends of the last 5 years".
AIs sometimes "hallucinate": verify important information with more than one source . If the answer is critical, cross-reference information and request additional checks before making decisions.
Customize ChatGPT from the settings
In the app, tap on your name (bottom left) and go to “Customize GPT” . There you can tell it how to address you, what you do, and what style you prefer for daily interactions.
Define your professional context in “What do you do?” to adjust vocabulary and examples to your area , and explore “What personality should ChatGPT have?” with the base styles and additional traits.
- Base personalities: Default (balanced), Cynical (sarcastic and critical), Robot (direct and straightforward), Attentive (helpful), Geek (detail-oriented and relaxed).
- Extra features: talkative, imaginative, sincere, encouraging, pragmatic, direct, corporate, empathetic, etc.
In “Anything else I should know about you?” you can add preferences and quirks, but avoid sensitive information . Keep in mind that OpenAI may use conversations to improve its models, so it's a good idea to review the data controls.
Activate or deactivate the memory with "Consult saved memories": if you prefer fresh conversations without prior context bias , turn it off; if continuity suits you, leave it on.
Quick questions: This customization is available on the free plan ; and no, changing the tone or style does not alter the intrinsic truthfulness of the model's data, only its way of responding.
Personalized instructions: what they are and how to activate them
"Custom instructions" allow you to set persistent preferences that are automatically applied to each response (tone, structure, level of detail, etc.). They have a limit of about 1.500 characters, so it's best to keep them concise.
- WebsiteClick on your name → “Custom Instructions” → define how you want it to respond.
- iOS / Android: app menu → your name → “Custom instructions” → adjust your preferences.
When you activate them, you'll avoid repeating requirements in each chat and maintain consistency between sessions, which is key if you work with processes or teams.
Three principles for writing good instructions
1) Clear response guidelines
It indicates how to argue, when to ask for clarification, and whether to provide several alternatives, break down steps, and cite sources . A general example: “If the question is ambiguous, ask for more information; explain reasoning step by step and show 2-3 possible approaches; use a professional style unless otherwise requested.”
If you create content, add SEO and formatting requirements: “include related keywords, organize with subheadings and bullet points , and change 'Conclusion' to a thoughtful closing.”
2) Verbosity control and language level
Define the amount of detail using a scale of 0 to 5 (V=0 minimal, V=5 extensive) and your preferred language level from A1 to C2 (L=A1…L=C2) . For example, “V=4, L=B2” for extensive answers in intermediate-advanced Spanish.
If you don't specify anything, the model will make assumptions; that's why it's a good idea to set a default value in the instructions and then adjust it in each conversation as needed.
3) Roles for different scenarios
Since you can only have one set of instructions at a time, define letter-switchable roles for different tasks and mention them when asking: “R=W” (writer), “R=P” (programmer), “R=A” (analyst), “R=L” (lawyer), “R=G” (generalist), “R=C” (content creator).
If you don't specify a role, let them infer it from the topic, but encourage them to ask before assuming. This technique reduces constant manual changes when switching between tasks.
Add-ons, memory and privacy
Customized instructions also improve the use of plugins: if a plugin uses location , for example, the experience is smoother if those preferences are already clear and you don't have to repeat them.
Privacy: You can disable the use of your chats for training in Settings → Data controls → “Chat history and training” . If you're concerned about your data footprint, turn off this option and avoid sharing sensitive information.
Remember the memory switch: preserving memories helps string conversations together , but if you're bothered by biases from previous context, turn it off to keep each chat "clean".
Templates and coworking spaces as a quick alternative
If you don't feel like designing your instructions from scratch, there are pre-designed templates for generating content (blogs, SEO texts, essays, tutorials…) that save time and allow you to change the tone on the fly (for example, to level B1 with one click).
Some suites include “digital coworkers” (an evolution of custom GPTs) for brainstorming, planning, and publishing . Always look for tools that prioritize privacy and comply with frameworks like GDPR.
Debate mode to sharpen your ideas
You can turn ChatGPT into a critical, not just accommodating, partner by adding a block like “analyze my assumptions, offer counterpoints, and test my logic .” Also, ask it to point out biases and propose alternative frameworks.
- Analyze assumptions: what I take for granted that could be false.
- CounterpointsWhat would a well-informed skeptic say?
- Logical stress test: flaws or gaps in reasoning.
- Alternative perspectives: other ways of framing the topic.
- Truth before agreement: correct clearly when necessary.
Advanced customization template (ready to adapt)
If you want an "ultra-intelligent assistant," create a short block that combines reasoning habits, style, quality control, and empathy. These pieces work very well together :
- Interdisciplinary perspective: integrate project management, psychology, economics, design, marketing and engineering, citing relevant references where appropriate.
- Self-auditAfter each response, check consistency with known frameworks and adjust if you detect contradictions.
- Proactive clarityIf you have any doubts or gaps in the information, ask for clarification before making assumptions.
- real examples: illustrate with specific cases (what went well/wrong and why).
- Human toneClear, direct language without unnecessary jargon.
- academic support: add classic and current studies if they shed light, explained in an understandable way.
- "What if..." scenarios: explore alternative paths and their implications.
- Logical controls: point out weak assumptions and propose alternatives.
- Final mini-summary: close with 3-5 key bullet points to reinforce ideas.
- Debate mode: present pros and cons when it comes to decisions.
- Contextual memory: track previously discussed objectives and limits.
- Adaptive empathy: detect stress and adjust tone and rhythm.
- Depth on demand: offer a brief version or an extensive analysis as requested.
- Useful frameworks: apply RICE, OKR or others when appropriate.
- Breach detection: identify data gaps or biases.
- Narrative: explain complex concepts with short stories.
- Expanded areas: contribute in marketing, finance, leadership, etc., depending on the context.
- Topic Tracking: remember recurring themes to delve deeper.
- Stable terminology: maintain consistent definitions unless you request changes.
- Continuous improvement: incorporate your feedback to refine the style.
Reasons for this approach: more rigor (self-criticism and citations), more usefulness (scenarios and summaries) , and more closeness (human tone and empathy) translate into applicable and reliable answers with fewer iterations.
Practical applications: If you're an entrepreneur, use it to design business strategies with proven frameworks; if you work in digital communication, it will help you structure campaigns and content strategically ; if you want to improve productivity or leadership, activate it as an advisor that proposes concrete actions and measures results. Even in local settings (for example, in cities like Marbella), you can adjust the tone and examples to the market and audience.
Table of useful shortcuts
To get a quick overview, here's a summary of common adjustments and their impact; use it as a quick cheat sheet :
| Adjustment | What do you get? |
|---|---|
| Personalized tone and style | More natural responses tailored to the audience |
| Technical level and verbosity (V and L) | Control of linguistic depth and complexity |
| Required structure (vignettes, comparisons) | Clearer and more actionable reading |
| Roles (R=W, P, A, L, G, C) | Responses focused on the correct scenario |
| Sources and institutions | Greater reliability and useful quotes |
Natural questions, context, structure, sources, and controls are the combination that makes ChatGPT a top-notch assistant: fast when needed, deep when necessary, and always aligned with your way of working.
