Do You Really Know How to Write Effective Prompts for GPT-5?

Most people still treat a prompt as just a command. It’s not. A good prompt is an operational contract between you and the model — it defines roles, boundaries, thinking patterns, and acceptance criteria. Below, I break down the anatomy of a high-performance prompt into 6 components, with a reusable template and a real example.

---

1) Role — state who the model is

Define the role precisely, not with fluff. The better the job description, the less ambiguity.

• Ex: “Act as a senior [[UX Writer]] specializing in B2B products in the SaaS sector.”
• When useful, add principles (e.g.: “clarity > style”, “avoid jargon”, “professional and direct tone”).

Goal: reduce degrees of freedom without stifling creativity.

---

2) Task — state what needs to be delivered

Be specific in the verb (create, rewrite, prioritize, compare, estimate). Bring in visible scope + restrictions.

• Ex: “List 3 options of medium-length hikes, within 2h of [[San Francisco]], excluding obvious tourist spots.”

Goal: turn a vague intention into executable work.

---

3) Context — rules, data, and quality criteria

Without context, the model guesses. With context, it optimizes.

• Domain rules, input data, persona, audience, objectives, and quality criteria (e.g.: geographic accuracy, terminology consistency).
• Include what not to do (e.g.: “do not mention [[Mount Tam]] or [[Golden Gate Park]]”).

Goal: reduce noise and align expectations.

---

4) Reasoning — how the model should think

Request steps, internal checks, and cross-referencing sources when applicable.

• Ex: “Check trail names with reliable sources and validate time/distance before responding.”
• If the task is critical, ask for self-checking (“review inconsistencies before finishing”).

Goal: make reasoning auditable and less prone to hallucinations.

---

5) Output format — predictable and ready-to-use

Define the output container (e.g.: [[Markdown]] table, JSON, hierarchical bullets) and mandatory columns/keys.

• Ex: Table with Name, Address, Time, Distance, Summary.
• For automation, prefer JSON with a fixed schema.

Goal: reduce rework and guarantee fit into your pipeline.

---

6) Final conditions — acceptance and verification criteria

Close the contract: what characterizes “done”? What are the success conditions?

• Ex: “The task is complete when there are 3 verified, unique, medium-length trails, within the time/distance, excluding popular locations.”

Goal: turn subjectivity into a Definition of Done.

---

Ready-to-use template (copy/paste)

Use this block as the basis for any task with GPT-5.
Adjust domain, restrictions, and format to your use case.

Act as <Role/Specialty> for <Audience/Sector>. Prioritize <Principles>.

Task:

- <Verb+deliverable> with <scope> and <main restrictions>.
- Include <quantity/variations> where applicable.

Context and Rules:

- <Input data / sources / exclusions / quality criteria>.
- <What to avoid>.

Reasoning:

- Follow steps: <1>, <2>, <3>...
- Check <consistencies / data / sources>.
- If gaps are found, suggest hypotheses and indicate uncertainty.

Output format:

- Deliver in <Markdown/JSON/Table>.
- Required fields: <...>.

Final conditions:

- Consider the task complete when <acceptance criteria>.