Business Insider, December 9, 2025: After OpenAI employees published six tips for 'using ChatGPT more intelligently,' a reporter tested each one and recorded the notable changes and limitations.
The tests began with the two most basic tips: specify identity and audience, and define the output format. The reporter prompted the model: 'You are a growth manager; write a playful Instagram caption for users aged 18–25, within 45 characters.' ChatGPT returned three drafts that matched the platform's tone, included emoji and calls to action, and required almost no edits. This scenario directly demonstrated how templated prompts improve output efficiency.
Next came constraints and examples: when asked to compress a legal article into three points understandable to general readers, the reporter provided the target audience and example phrasings. The model's three‑point summaries were more consistent in length and style and retained key terms, avoiding multiple rounds of correction.
'Please explain your reasoning step by step' was listed as the third tip. While debugging a Python script, the reporter asked the model to explain errors line by line; ChatGPT provided stepwise troubleshooting paths and repair suggestions, ultimately shortening debugging time. The author also noted that enabling step‑by‑step explanations increases token usage and that complex tasks can hit response length limits.
The fourth and fifth tips were to provide sources and self‑critique: when the reporter requested sources and an uncertainty statement, the model could annotate some references, but in a few cases it still generated plausible‑seeming citations without verifiable origins (so‑called 'hallucinations'), requiring user verification. Having the model produce an initial draft and then play the role of 'reviewer' to suggest improvements noticeably enhanced the text's logic and readability.
The final tip emphasizes iteration and fine‑tuning: different prompt wordings for the same task produced markedly different results; narrowing the instruction scope often yielded more directly usable outputs than vague requests.
Overall assessment: these methods are not magic, but when prompt engineering is turned into a reproducible workflow, ChatGPT performs more like a practical tool in content creation, product planning, and code repair. The reporter warns that for high‑confidence facts or academic citations, human verification is still necessary; for younger users, these tips lower the barrier to use and make AI feel more like a collaborator than a 'black box'.