Comment écrire avec l'IA ? – Formation Complète
AI Summary
The speaker shares their extensive experience using AI for content creation, from marketing articles and administrative documents to books and novels. Having trained clients since 2018, well before ChatGPT's widespread adoption, they have identified key best practices and common pitfalls. This video aims to provide a comprehensive guide to writing effectively with AI.
The training is divided into two parts: first, creating professional, office-style content, and second, writing books and novels. Before diving in, the speaker mentions their private community, "Le QG de l'IA" (The AI HQ), where members share best practices for AI writing. The formation presented is partly derived from a roundtable discussion within this community.
The core principle for effective AI writing is that the best results don't come from a single "magic prompt." Instead, a structured process is required. AI does not replace human intention, objectives, style, or editorial responsibility. The author remains accountable for all published content. Writing with AI is an iterative process that involves guiding, feeding, and correcting the AI, using specific AI tools for different stages. The human author's intention, voice, ideas, and quality judgment are paramount.
For professional content (marketing, administrative, general business), the golden rule is to apply "prompt chaining" or "prompt engineering." The key is to avoid asking the AI for a final piece of content in one go; instead, the process must be broken down. This applies to all types of content, from administrative reports to books and biographies.
A five-step process is outlined for creating quality content:
1. **Clarify your project and objective:** Define the subject, theme, target audience, and search intent (especially for online content). Understanding what users search for and their underlying intentions (information, entertainment, purchase) is crucial. AI can assist in this clarification, including developing personas.
2. **Structure your content:** Use AI to create a detailed plan and logical progression. This ensures that the audience's attention is held and key information is effectively conveyed. This applies to articles, books, and training materials.
3. **Draft the content:** This is not a one-time drafting process. Content should be worked on section by section, or even part by part, with introductions often written last. For a blog post, the process can involve up to ten distinct steps.
4. **Refine and optimize:** Enhance the content to meet objectives. This involves enriching it with examples of preferred content, evidence to illustrate objectives, and elements of your brand voice.
5. **Review:** Use a quality checklist and ask AI to critique the content, perhaps using the "devil's advocate" technique.
A significant challenge with AI is the potential for bias. To mitigate this, using multiple AI tools is recommended. The speaker's workflow involves:
* **ChatGPT** for initial clarification, structuring, and producing the first draft.
* **Google Gemini** (optional) for adding creativity, authenticity, storytelling, challenging approaches, and leveraging recent web sources due to its effective search capabilities.
* **Claude** for finalization and improvement, as it's considered the best AI writer for enhancing quality, refining style, and ensuring fluidity and coherence with the author's voice.
After initial drafting with ChatGPT, Gemini might be used for creative enhancement. The improved draft is then passed to Claude for refinement according to the author's specific writing method, voice, values, and expression. Finally, ChatGPT can be used again for a final review, employing techniques like the devil's advocate. For most content, a combination of two AIs (e.g., ChatGPT and Claude) is sufficient to avoid bias and allow the AIs to challenge each other.
**Developing a Brand Voice:**
To ensure AI-generated content reflects a specific brand voice, a four-step method is suggested:
1. **Gather examples:** Collect your own written content (articles, social media posts, emails, books) or content from authors you admire, in the desired format.
2. **AI analysis:** Ask the AI to create a detailed writing brief or specification sheet based on these examples, identifying tone, rhythm, structure, recurring expressions, pedagogy, emotional density, and storytelling techniques.
3. **Transform into instructions:** Convert the AI's analysis into clear instructions.
4. **Iterate and refine:** Test and improve the AI's writing machine over time.
This analysis can be transformed into a PDF document to be used within AI projects (like custom GPTs or Claude projects) to train the AI to write in a specific style. A good style guide should include an editorial charter, target persona details, examples of existing content, brand values, desired tone and register, editorial prohibitions, and format objectives.
**Handling Sensitive Content:**
For sensitive documents like legal, administrative, or HR materials, a more rigorous review process is necessary after the enrichment stage. This involves:
1. **Initial AI draft and refinement.**
2. **Cross-audit:** Use a second AI to identify flaws, potential objections, risks, ambiguities, omissions, or factual errors.
3. **Content rework:** Incorporate feedback from the audit into the content.
4. **Source verification:** Double-check sources, legal texts, and official references, either with AI or manually.
5. **Human review:** Always include a human review for sensitive documents, as ultimate responsibility lies with the author.
**The Devil's Advocate Technique:**
This technique involves asking the AI to adopt the persona of the content's recipient to critique it, identify weak arguments, and raise objections. This feedback is then used to improve the content. The process can be repeated with different AIs for diverse perspectives and to avoid bias. Examples include:
* A sales proposal challenged by an demanding buyer.
* A job description critiqued from the perspective of an ideal candidate.
* A blog post reviewed by an expert in the field or the target persona.
The speaker then promotes their private community, "Le QG de l'IA," offering a special price for a limited time. The community provides training, ready-to-use workflows, private live sessions, and a network of over 900 members for support and collaboration. It addresses the common issue of AI content feeling generic, lukewarm, or lacking authenticity.
**Five Antidotes to Generic AI Content:**
1. **Provide ample material:** Feed the AI with more examples, documents, and instructions. The AI produces richer content when given more input, including your own drafts or preferred content.
2. **Assign a role:** Use "persona prompting" effectively. Instead of just asking the AI to "act as," use specific keywords that guide it accurately and concisely, respecting token limits. This is universally recommended by AI developers.
3. **Break down tasks:** AI is more creative when working on distinct stages (clarification, structuring, drafting sections, optimizing) rather than generating content in one block.
4. **Intentional review:** Ask the AI to critique content not just for general improvement, but to identify generic passages, strengthen examples, simplify long sentences, and address specific areas you've identified as weak.
5. **Share human input:** Authenticity comes from your unique perspective, intentions, and the human element you inject into the AI-assisted process.
**Workflow for Professional Content:**
1. **Prepare content:** Clarify intention and share relevant material with the AI.
2. **Structure and draft:** Use ChatGPT for planning, structuring, and the initial draft.
3. **Critique and research:** Employ another AI (Gemini, Perplexity, Notebook LM) to critique, challenge, and conduct supplementary research.
4. **Finalize and refine:** Use Claude for optimization and polishing, incorporating your brand voice through its project or skill features.
5. **Final critique:** Use the devil's advocate technique.
6. **Beta reading:** Consider AI-assisted beta reading and a final human review.
**Writing Books, Novels, and Biographies with AI:**
AI makes the lengthy process of writing a book more accessible but does not replace the author. AI can help structure, accelerate drafting, reformulate, enrich, beta-read, and analyze style. The author's intention, story, characters, plot, personal perspective, narrative coherence, and lived experiences remain the author's domain. AI can assist in refining character profiles and narrative arcs.
**Key benefits of using AI for books include:**
* Making long projects accessible.
* Generating multiple testable versions.
* Providing immediate feedback.
* Overcoming writer's block.
* Enabling a much faster writing pace.
**Starting a Book with AI:**
Six categories of material should be shared with the AI:
1. **Purpose of the book:** Why is it being written? (Professional, thematic, novel).
2. **Ideal reader:** Who are they? What do they expect, like, and read?
3. **Central emotion or idea:** What should the book convey? How should the reader be changed?
4. **Key scenes (for novels):** Outline significant plot points.
5. **Narrative voice:** Desired tone, perspective, and style.
**Book Writing Protocol:**
1. **Global structure:** Define the book's overall structure.
2. **Chapter-by-chapter writing:** Do not write in one block. Work through chapters sequentially.
3. **Define style and narration rules:** Share examples of authors you admire and character profiles. AI can help develop character arcs, mandatory scenes, length constraints, review criteria, and editorial prohibitions.
4. **Character development:** Define objectives, wounds, contradictions, evolution, language, relationships, secrets, and flaws. The "Hero's Journey" is a useful framework.
5. **World-building:** Establish context, era, location, geography, world rules, and cultural constraints.
6. **Plot development:** Outline the initial situation, inciting incident, obstacles, plot twists, climax, resolution, and falling action.
7. **Scene development:** Detail the inciting incident, progressive tension, narrative objective of each scene, and its function within the overall arc.
For the drafting phase, providing at least 1000 words of your own writing or writing in the desired style is recommended, especially when using Claude. Analyze your style with AI to create a detailed stylistic profile (sentence rhythm, description level, psychological depth, narration type, dialogue approach, emotional density, recurring imagery, tone shifts, language level, suspense techniques). Integrate this profile into an AI project or skill for consistent application.
**Seven Steps for Writing a Book with AI:**
1. **Clarify intention:** Book type, ideal reader, promise, theme.
2. **Reference dossier:** Create a synopsis, define characters, universe, and style.
3. **AI protocol:** Follow the established AI process.
4. **Detailed structure:** Outline arcs and key scenes.
5. **Chapter-by-chapter writing:** Draft, challenge AI, refine, and consolidate each chapter.
6. **Beta reading:** Use AI (acting as editors, target readers, critics) and potentially multiple AIs for diverse perspectives.
7. **Finalization:** Incorporate AI feedback, potentially human rewriting of specific sentences, and a final human proofread before publication.
**10 Fundamental Principles for AI Writing:**
1. **AI writes better when guided:** Provide material, examples, and your own writings.
2. **Human raw material is essential:** Inject humanity for authentic writing.
3. **Process over prompt:** The overall workflow is more critical than a single prompt.
4. **Break down the work:** Use prompt chaining and segmented processes.
5. **Style matters:** Develop an authentic style through examples and a technical style sheet.
6. **Cross-reference AIs:** Use multiple AIs to improve quality and avoid bias.
7. **Critique is as important as writing:** Challenge AI writing with other AIs.
8. **Human responsibility remains:** Authenticity and unique criteria come from human intention.
9. **Your raw material makes the difference:** Whether professional or literary content, your input is key.
10. **Know when to stop optimizing:** Avoid over-optimization, as it becomes counterproductive. Limit AI reviews to two or three, plus human review.
Practical prompts include using AI for style guides, the devil's advocate technique, beta reading (analyzing strengths, weaknesses, and offering rewrite suggestions), and optimizing content according to your brand voice.
In conclusion, AI accelerates writing but doesn't replace vision. Creativity and raw material are what differentiate content. Otherwise, it