AI methodological note: prompt engineering process
Table of Contents
AI tools used
ChatGPT (OpenAI)
- Version: Free.
- Usage: Brainstorming and developing the article's angle, identifying complementary arguments, structuring the article, refining English writing, suggesting SEO keywords and drafting the methodological note.
- Link: chatgpt.com
TinyPNG
- Version: Free.
- Usage: Compressing and optimizing images before uploading them to WordPress to reduce file size and improve website loading performance.
- Link: tinypng.com
Exploration, idea development & creativity
The idea originated from Emma Vidal Madjar's article on the MBA DMB blog about AI bias and the reputational risks for brands. Its connection to my professional thesis on AI and international communication made the topic particularly relevant. I used ChatGPT to explore different editorial angles and develop my initial ideas, including automation bias, unconscious human bias and Shadow AI. I selected these themes because they complemented the original article while bringing a new perspective on corporate responsibility.
Planning & organization
Rather than relying on a single generic prompt, I used ChatGPT progressively to transform my initial ideas into a structured rebound article. The process involved several steps:
- Initial exploration: I provided the original MBA DMB article and asked ChatGPT to summarize its main arguments. This helped identify potential directions for a complementary article.
- Angle development: I shared my initial reflections on human bias, the influence of prompts, automation bias and Shadow AI. ChatGPT helped connect these ideas and formulate a central question: If AI systems reproduce existing biases and humans are not always capable of recognizing them, how can companies ensure that AI-generated communication does not damage their brand reputation?
- Research and structure: I identified relevant academic studies, press articles and examples of advertising controversies. ChatGPT helped organize the arguments into four sections: the origins of AI bias, human limitations, Shadow AI and solutions for responsible AI governance.
- SEO and final structure: I requested a structure adapted to a 1 000-word article, including a table of contents, SEO-friendly headings, keywords and a meta description.
Drafting & articulation
ChatGPT was used as a writing assistant to help articulate my ideas in English, connect research findings and improve the clarity and flow of the arguments. It helped transform my initial notes into a coherent article while maintaining a professional and accessible editorial tone.
- Tone & vocabulary: Professional, analytical and accessible, targeting marketing and communication professionals, brand managers and digital specialists.
- Content development: AI helped formulate transitions, explain technical concepts such as automation bias and Shadow AI, and integrate research findings into the article.
- SEO optimization: ChatGPT suggested a primary keyword (AI bias and brand reputation), secondary keywords, an SEO title, a URL slug and a meta description.
- Target impact & KPIs:
- Visibility: Improve organic search visibility through relevant keywords and structured headings.
- Engagement: Encourage discussion among communication professionals about AI responsibility and human supervision.
- Readability: Deliver an informative article of approximately 1,000 words, with a clear structure and concrete examples.
Ethics, sources & hallucinations
- Source verification: ChatGPT helped identify and summarize relevant research, but its suggestions were treated as starting points rather than automatically reliable information. Figures, study findings, publication dates and examples must be checked against the original sources before publication.
- Critical review: Particular attention was given to distinguishing AI-generated advertising from advertising controversies caused by human creative decisions. The Converse controversy was therefore used to illustrate human and organizational blind spots, not as evidence of an AI-generated campaign.
- Attribution: The original MBA DMB article, academic studies, press articles and institutional sources are referenced in the main article. AI assistance in brainstorming, structuring and drafting is disclosed in this methodological note.
- Limitations: ChatGPT can produce inaccurate figures, misinterpret research findings or suggest unsupported conclusions. It also cannot independently guarantee that an article is free from cultural bias.
- Human review and source verification remain essential.
- Data privacy: No personal or confidential company information was shared during the interactions.
Prompts & interaction examples
The process involved several exchanges in which I progressively refined the subject and the expected output. Here are three examples:
Developing the editorial angle
- User prompt: "Here are some ideas for the subject of the rebound article: AI bias creates reputational risks for brands, but how can companies become more aware of these biases? Human reviewers may also have unconscious biases, and employees sometimes trust AI recommendations too much. Can you combine these ideas and propose a plan?"
- AI contribution: ChatGPT connected the different ideas and helped identify a broader argument: AI bias is not only a technological issue, but also a human and organizational challenge. It suggested introducing automation bias and Shadow AI to move beyond the original article's discussion.
Structuring the argument and integrating research
- User prompt: "Please write the article within 1,000 words, taking into account SEO. Use the central question: If AI systems reproduce existing biases and humans are not always capable of recognizing them, how can companies ensure that their cross cultural communication does not damage their brand reputation?"
- AI contribution: ChatGPT proposed a four-part structure covering the origins of AI bias, human limitations, Shadow AI and practical solutions. It also suggested integrating academic research and real advertising controversies to support the argument.
Improving the accuracy of an example
- User prompt: "Please use the Converse controversy as an example if pertinent to show how people can still fail to recognize cultural bias, thus affecting the brand's reputation."
- AI contribution: This exchange helped clarify the distinction between AI-generated bias and human cultural blind spots. The Converse example was integrated to demonstrate that problematic representations can pass through human approval processes, reinforcing the article's argument that human supervision alone is not sufficient.
Discover the article "Is Human supervision enough to prevent cultural bias in AI ?" by clicking the link below.
