The old prompt mindset is too small
In 2023, many people treated prompt engineering as the art of writing one perfect sentence. That was useful for early chatbots, but it is not enough anymore. In 2026, AI systems can read images, PDFs, charts, screenshots, code, audio, and long documents. Some can use tools, search, calculate, create files, or run multi-step tasks.
That changes the skill. Prompt engineering is no longer only question writing. It is workflow design.
The prompt anatomy I teach
A strong modern prompt usually includes six parts:
Role: who the AI should act as - tutor, reviewer, marketer, analyst, facilitator, or product strategist.
Goal: the outcome you need, not just the topic you want discussed.
Context: audience, business type, files, constraints, level, examples, and local situation.
Constraints: length, tone, format, boundaries, privacy rules, and what not to do.
Output format: table, checklist, script, slide outline, rubric, decision memo, or step-by-step plan.
Verification: ask for assumptions, checks, sources, uncertainty, or points requiring human review.
R-A-C-E + V
For students, founders, and teams, I often simplify the structure into R-A-C-E + V:
R - Role: Act as a practical business mentor for a small Nepali business.
A - Action: create, critique, compare, summarize, plan, or test.
C - Context: the customer, offer, market, file, screenshot, budget, or current workflow.
E - Expectation: the output style, format, tone, length, and level of detail.
V - Verify: check facts, flag assumptions, list risks, or identify what needs human review.
Weak prompt, stronger prompt
A weak business prompt says: “Give me marketing ideas.”
A stronger prompt says: “Act as a practical marketer for a Nepali homemade achar brand targeting working women in Kathmandu. Create a 7-day Facebook content plan with caption idea, visual idea, proof element, and CTA. Keep the tone natural, not luxury-brand English. Flag any claim that needs proof before publishing.”
The second prompt works better because it carries business judgment inside it: customer, product, channel, tone, proof, and quality control.
Prompt repair is a core skill
When AI gives a generic answer, do not simply accept it. Repair the prompt:
Add the customer.
Add the offer.
Add the location or market context.
Add the desired tone.
Add the output format.
Ask for alternatives, trade-offs, or a critique.
A useful repair line is: “This is too generic. Rewrite it for [customer], [location], [offer], in a [tone] tone, as a [format]. Add what must be verified before use.”
Verification is not optional
AI can sound confident even when it is wrong. It can invent names, dates, sources, features, policies, prices, and local facts. That is why every serious prompt workflow needs a checking rule.
For current facts, ask for dates and sources. For numbers, ask for assumptions. For public claims, ask what evidence is required. For private data, ask whether the information should be uploaded at all.
The practical takeaway
Good prompting is clear thinking made visible. The best users are not the ones who memorize tricks. They are the ones who can define the work, give context, set boundaries, verify output, and improve the result through a loop.