Configure ChatGPT for Your Profession

An assistant tailored to your professional reality

The previous two lessons gave you the tools: custom instructions, which you write, and memory, which builds itself. This lesson combines them to transform ChatGPT into a true professional assistant, calibrated to your daily needs. The difference between a user who has done this work and one who hasn’t is obvious from the very first response: one gets a deliverable that’s almost ready to use, while the other gets generic text that needs a complete rewrite.

Start by defining your professional profile

Before you touch any settings, take five minutes—with a real notebook, not just mentally—to answer five questions. What is your exact role? Not just “manager,” but “sales manager for 12 field representatives in the food industry.” It’s this level of detail that allows ChatGPT to choose the right angle. What are your recurring deliverables—weekly reports, client emails, product sheets, sales proposals? These are what ChatGPT will produce most often, so it needs to know their nature. What tools do you use (CRM, office suite, specialized software) so the advice is actionable in your actual environment? What are your constraints—industry regulations, communication guidelines, language? And finally, what tone do you adopt, knowing it’s often dual: formal with clients, direct internally?

This questionnaire isn’t a formality. Every answer becomes a line in your instructions, and every unanswered question becomes a blind spot in ChatGPT’s responses.

Three profiles, three configurations

Let’s see how this work translates in practice for three very different professions.

For a corporate lawyer, the “Who I am” section could be: “Senior corporate lawyer at a 30-attorney firm in Paris. Specialized in M&A and corporate law. Clients: French SMEs and mid-cap companies.” And the “How I want responses” section: “Structure your answers in numbered points. Cite applicable legal articles, flagging those I need to verify. Distinguish between French and European law where relevant. Flag if recent legislative changes could alter the answer, searching the web if needed. Use precise legal French without unnecessary jargon.” Note that every instruction stems from a professional requirement: citing articles, for example, because a legal assertion without a source is unusable in a firm.

For a middle school teacher, the context would be: “Science teacher in middle school, grades 6 and 9. Lyon school district. I need activity ideas, differentiated exercises, and help with assessments.” And the format: “Adapt vocabulary to the middle school level. Propose activities feasible within 55 minutes. Always include a clear learning objective. Suggest variations for struggling students and advanced learners.” The 55-minute constraint perfectly illustrates the principle: it’s a detail a non-teacher would never think to specify, yet it’s precisely what makes the proposals usable as-is.

For a full-stack developer, finally: “Full-stack TypeScript developer, Next.js + PostgreSQL + Prisma stack. Working at a 15-person startup. Projects follow Agile methodology.” With this format: “Provide functional code, not pseudo-code. Use strict TypeScript. Prefer solutions with the most maintained libraries. Explain performance trade-offs where relevant. No obvious comments in the code.”

The 5-request test

Once you’ve configured your instructions, don’t consider them validated until you’ve stress-tested them. Submit five typical requests from your daily routine to ChatGPT: a writing task (email, report, memo), an information search in your field, a data or situation analysis, a creative request (brainstorming, proposal), and a repetitive task you’d like to automate. If the responses don’t meet your expectations (too long, wrong tone, ignored context), go back to Settings > Personalization and adjust the faulty instruction; a base style or characteristic might also be misconfigured. It’s an iterative process: two or three rounds are usually enough to achieve a solid configuration.

Aligning instructions, memory, and prompts

The optimal strategy layers three components, with a fourth added for recurring folders. Custom instructions carry your core profile, stable over time: profession, tone, format. Memory captures evolving details: current projects, client names, deadlines. Contextual prompts, finally, bring the specifics unique to a given conversation. This division avoids two symmetric pitfalls: putting everything in instructions (which become bloated and outdated) or repeating everything in every prompt (which negates the benefit of personalization).

A word of caution to finish. The three examples above are starting points, not off-the-shelf solutions—copy-pasting them without adaptation reproduces the very problem you’re trying to solve. If your sector imposes confidentiality rules (healthcare, finance, legal), integrate them explicitly into your instructions. And if you use ChatGPT for both personal and professional purposes, consider two accounts or alternate instruction profiles; otherwise, the two contexts will contaminate each other.

The fourth layer is the project: for a client or case that recurs constantly, its instructions and files take precedence within the project over your general instructions (see the Writing Blocks, Memory, and Projects course).

Your turn

Your professional configuration. Answer the five questions from the start of the lesson in writing, turn each answer into an instruction, then draft your instructions in two parts in Settings > Personalization, staying within your plan’s limit. Next, run the five-request test and adjust at least twice.

Official sources: ChatGPT Custom Instructions (OpenAI Help Center) · Memory in ChatGPT · Projects in ChatGPT (project instructions)

Key takeaways

• A precise professional profile transforms ChatGPT from a generic tool into a specialized assistant
• First answer the five questions (role, deliverables, tools, constraints, tone): each answer becomes an instruction
• Combine three layers: instructions (stable framework), memory (evolving details), prompts (specific details)
• Validate with the 5 typical request test, then iterate
• Adapt templates to your reality—there is no universal configuration