How to customize ChatGPT and fine-tune your responses like a pro

Informatec Digital » Resources » How to customize ChatGPT and fine-tune your responses like a pro

Conversational AI isn't limited to a single name. While ChatGPT has been the go-to for millions since its launch, there are top-tier alternatives like Google Gemini, Grok, Meta AI, Perplexity or DeepSeek which is also worth knowing. In any case, if you're going to work with ChatGPT daily, it's worth adjusting its behavior so that it responds as you need and with the style that best suits you or your brand.

Starting with a blank AI can be reminiscent of cold calling: without context or instructions, the results are more generic. Luckily, ChatGPT allows customize the tone, format, and depth In addition to saving key preferences (nickname, occupation, or style traits), this guide will show you how to activate and optimize custom prompts, provide prompt tricks to improve quality, offer privacy tips, and provide advanced settings ready to customize.

Customize ChatGPT step by step: styles, tones, and memory

It all starts in the service's dashboard. On the web version, tap your account icon (bottom left) and choose the option "Customize GPT" In the settings. From there you can tell it your name, your occupation, and what you expect from its communication. This information helps the model use more familiar vocabulary to your professional context and avoid unnecessary detours on topics you are familiar with.

In the personality block, ChatGPT offers basic styles: Default (the cheerful factory tone), Cynic (more sarcastic), Robotized (direct and without embellishment), Attentive (warm and collaborative) and Geek (detail enthusiast). You can also add fine features like talkative, imaginative, sincere, encouraging, pragmatic, direct, corporate, or empatheticeither by choosing suggested labels or writing your own description of the ideal style.

The "Anything else I should know about you?" section allows you to include habits, interests, likes, and dislikes. Use it wisely: although it can improve the personalization of the responses, It is not advisable to enter sensitive information.Unlike other systems such as Gemini, ChatGPT retains copies of conversations at the service level, even if you delete the local history; therefore, it is advisable to limit personal data and focus on work preferences and style.

Regarding availability and scope: customization options by instructions are available in both the free and paid plansIt's key to understand that this customization adjusts how ChatGPT responds (tone, format, structure), but It does not change the veracity of the knowledge base of the model; that is, it does not make AI a different source, it only guides how it expresses what it knows or seeks.

One last very useful setting is the memory setting: the option to consult saved memories This allows ChatGPT to remember preferences and context from previous chats. If you prefer to avoid accumulated biases between conversations or keep your chats "blank," you can turn this switch off so that each session starts from scratch.

Personalized instructions: web and mobile activation and how to get the most out of them

The personalized instructions This allows you to set your preferences once, so the model takes them into account in every response. On the web: open your profile, go to "Custom Instructions" and fill in the fields, paying attention to "How you want ChatGPT to respond." The process is similar on iOS and Android. Lateral Menu, username at the bottom, and access to the same section to fill out guidelines.

To make them truly effective, think of three pillars. First, specify the desired behavior: for example, that break down complex problems into stepsAsk clarifying questions when the statement is ambiguous, offer several possible solutions, cite reliable sources with links if available, and maintain a professional style unless otherwise indicated. If you create content, include references to SEO, text organization, and tone, and ask that Do not repeat unnecessary disclaimers.

Second, define the language scope and level. A handy trick is to set a scale of verbosity V=0–5 (0, minimum; 5, exhaustive) and a level of linguistic competence L=A1–C2This allows you to request V=3 and L=B2 for clear and moderately detailed explanations, or V=5 and L=C2 for in-depth analysis in expert language. If you forget to specify this in a query, the system can infer it from the subject, but it's best to set it at the beginning of each critical request.

Third, it uses roles to change focus without rewriting everything. It includes a parameter for role R=… In your instructions, define several scenarios. For example: P=Programming (tips and code examples), W=Writing (factual text, tone, and style), A=Analysis (simplify and contribute ideas), L=Legal (legal terms and concepts), G=Generalist (objective overviews), and C=Content Creator (ideas, keywords, and timelines for 18–25 years). This allows you to request "IA, V=4, L=C1, R=W" for a well-structured article, or switch to "R=P" when you need technical guidance.

There are limits to keep in mind: the "How do you want me to respond" box has a approximate restriction of 1.500 charactersTherefore, it's best to prioritize essential rules and rely on roles, V, and L to fine-tune them for each case. Developing them takes time (testing, debugging, and improving), but the investment It results in consistency and step savings. In day to day.

Integrations and data: Customized instructions also improve the experience with connected tools and connected tools (for example, remembering your area for location-based services, preventing you from repeating data). If you're concerned about privacy, go to Settings > Data controls You can disable "Chat History and Training" to prevent your conversations from being used to improve the model. In any case, avoid sharing sensitive or private information in both the instructions and the chats themselves.

Prompt techniques that make a difference

Forget the search engine chip. ChatGPT processes natural languageNot loose strings of keywords. Instead of writing "climate change definition studies," phrases like "Explain what climate change is based on peer-reviewed research" generate more complete entries with citations.

Specify what you want. The clearer you are, the better. “Write a professional email in English to apply for a job, highlighting skills X and Y” produces a more useful output than simply asking for a "email to find a job"Add the context and purpose you have in mind.

Include specific details. In text and images, every nuance counts. target audience, length, format, examplesStyle, keywords, dates, models, or exact versions. The AI ​​optimizes what you give it; if you fall short, it will tend to fill in with generic guesses.

Request the structure. If you need pros and cons, a comparison table, or bullet points, say so explicitly. A "Bulleted summary, divided into advantages and disadvantages, and add a brief recommendation" saves iterations and raises clarity of the result.

Adjust the level of explanation. If something sounds like jargon, ask for clarification.Explain it to me as if I were 10 years old» or adjust L=B1/B2 in your instructions. This changes the terminology, simplifies examples, and speeds up understanding.

Specify the source of the information. It's not the same to ask "what is inflation" as it is to ask "define it based on..." academic literature and macroeconomics textbooks with citations." Guiding AI towards reliable sources reduces ambiguity.

Mention organizations and media outlets. "According to the OMS", "with data from AEMET“, or “following Harvard publications” is more precise than a generic query. This guides the AI ​​toward specific references in your domain.

Require citations. If you ask for "references" or "citations to relevant studies," the model will tend to provide better documentation. Check the links afterward, because AI can sometimes mix them up. Sources and dates if the topic is very new.

Request recent data. When time is a factor, ask for up-to-date information, for example, "best rock albums." out of 2024" or "last 5 years". The model can consult the web, but it doesn't always do so by default, and specifying dates gives it clear responsibilities.

Verify what matters. All AIs can be delusional. Confirm critical information with multiple reliable sources before acting, especially regarding health. legal or financeAdding "indicates the level of confidence and possible limitations" to the prompt also helps to detect weak points.

Templates and assistants: accelerate without losing control

Creating good custom instructions takes time: research, testing, debugging, and fine-tuning. That's why many users turn to... predesigned templates that function as shortcuts for blogs, SEO, tutorials or essays, with the possibility of choosing tone and language level on the fly.

Some platforms also incorporate "coworkers" or specialized digital assistants (an evolution of personalized GPTs) for brainstorming, content planning, or campaigns. The key is that you can adapt your brand's voice and maintain control of the data, ideally under privacy frameworks such as GDPR.

Advanced behavior settings: a preset to turn it into an advisor

If you want ChatGPT to behave like a real multidisciplinary consultantYou can create a compact and powerful configuration within "Customize GPT". For inspiration, these blocks have proven to be very effective when combined:

• Interdisciplinary perspective: integrate references from management, psychology, economics, design, marketing and engineering, with citations of authors/studies where applicable.
• Self appraisalAfter generating the response, review its logic against existing frameworks and polish it if it detects contradictions or gaps.
• Proactive clarityIf the request is ambiguous, ask for details before making an assumption.
• real examples: incorporate anecdotes or cases of "what went well/what went wrong" to ground recommendations.
• Human tone: natural language, without corporate jargon unless required.
• Academic bridge: include study results translated into practical actions.
• Hypothetical scenarios: "what if..." analysis to compare paths.
• Logical controls: point out weak assumptions and propose alternatives.
• Executive Summary: close with bullet points to highlight key ideas.
• Debate mode: pros and cons of each route before recommending.
• Statement of objectives: follow the thread of goals and limits discussed.
• Empathy: modulate your tone if you detect stress or frustration.
• Depth on demand: offer a short version or a long analysis as requested.
• Recognized frameworks: RICE, OKR, Double Diamond, etc., when relevant.
• Detection of gaps: warn of data gaps or risky hypotheses.
• Narrative: explain complex concepts like a story when appropriate.
• Cross-cutting areas: marketing, finance, leadership, operations.
• Topic Tracking: to revisit recurring issues in future interactions.
• Consistent terminology: maintain stable definitions unless otherwise indicated.
• Continuous improvement: refine with your feedback to evolve the style.

You can combine this preset with your dials of V (verbosity), L (language level) and R (role)For example, V=4, L=C2, R=W when you want well-structured, technical writing; or V=3, L=B2, R=C for lighter content aimed at 18–25 year olds, incorporating keywords and an editorial plan when requested.

Applications in business and customer service

Personalization is especially valuable in customer use cases: it helps align the tone and message with the brand, and match the technical level with the interlocutor's knowledge. Automate responses without losing humanityThe combination of customized instructions, roles, and verbosity scales allows you to adapt to sales, support, training, or marketing with a single system.

Capacity	How to improve your results
tone and style	Natural messages aligned with the brand voice.
Technical level	Explanations for beginners or experts without redoing the prompt.
Verbosity	Fine control of the extent: summary or full report.
roles	Responses focused by context: legal, programming, writing, etc.

If you work with connected equipment or tools, personalized instructions enhance efficiency. third-party plugins and services (for example, logistics or booking apps), because the context is established beforehand and you reduce friction in each query.

Privacy and good practices

The million-dollar question: Is it safe to add personal data to the instructions? The recommendation is Do not include sensitive information Nor private. OpenAI can use conversations to improve its models, though you can disable this option in the data controls. Keep your preferences within the realm of style and work to minimize risks.

In your daily life, get used to verifying critical content and requesting that AI indicate Sources and limitationsDo this especially when the query depends on recent information: AIs can mix versions or fall short if they are not explicitly required to update.

Brief FAQ

Is customization available without paying?

Yes. Customized instructions They work in both the free version and the free version. as in payment plans.

Does it affect the accuracy of the answers?

No. Change the tone, structure and stylebut it does not alter the accuracy of the underlying information.

How do I activate it on web and mobile?

Go to your profile > «Custom instructions» and fill in the fields. On iOS/Android, access is through the app menu.

Can I prevent my training history from being used?

Yes. Go to Settings > Data Controls and Disable "Chat and Training History".

Are there any alternatives to speed up the setup process?

You can use specialized assistants that come with preset tone and format settings, keeping you in control of your data.

If you choose just one idea, let it be this: customizing ChatGPT isn't about forcing it to say what you want, but about teach him to respond as you needAdjusting styles, setting roles, requesting structure, delimiting sources, and rigorously verifying each output makes it more useful and reliable; with a few well-written rules, the assistant goes from a friendly generalist to a work ally that saves time and reduces errors.