{"id":4360,"date":"2026-01-11T11:28:37","date_gmt":"2026-01-11T10:28:37","guid":{"rendered":"https:\/\/laurentchani.fr\/?p=4360"},"modified":"2026-01-11T11:36:09","modified_gmt":"2026-01-11T10:36:09","slug":"state-of-the-art-of-authors-prompts","status":"publish","type":"post","link":"https:\/\/laurentchani.fr\/en\/etat-de-lart-des-prompts-dauteurs\/","title":{"rendered":"State of the Art of Authorial AI Prompts"},"content":{"rendered":"<figure class=\"wp-block-image aligncenter size-medium\"><img loading=\"lazy\" decoding=\"async\" width=\"300\" height=\"196\" src=\"https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-99-P-300x196.jpg\" alt=\"\" class=\"wp-image-4361\" srcset=\"https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-99-P-300x196.jpg 300w, https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-99-P-768x502.jpg 768w, https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-99-P-18x12.jpg 18w, https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-99-P.jpg 784w\" sizes=\"auto, (max-width: 300px) 100vw, 300px\" \/><\/figure>\n\n\n\n<p>Generative AI has changed the way we correct and improve a text. But for a novel, the real challenge isn&#039;t &quot;having the best prompt&quot;: it&#039;s building a <strong>reliable workflow<\/strong>, traceable and compatible with the <strong>author&#039;s voice<\/strong>. This article summarizes the best practices observed online (tools, methods, safeguards) and proposes a reproducible approach.<\/p>\n\n\n\n<p><strong>1) The turning point: from \u201cmagic prompt\u201d to \u201cprocess\u201d<\/strong><\/p>\n\n\n\n<p>On the internet, you can find hundreds of proofreading prompts. Most fail for one simple reason: they mix everything up (spelling, style, rewriting, consistency) in one pass. The result: the text becomes \u201cclean,\u201d but often <strong>standardized<\/strong>. For a novel, the priority is the opposite: <strong>preserve the intention<\/strong>, and only automate what is controllable.<\/p>\n\n\n\n<p><strong>2) The prevailing model: a chain in passes<\/strong><\/p>\n\n\n\n<p>The state of the art converges towards a chain in <strong>modules<\/strong>, From the most objective to the most subjective:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Pass A \u2014 Strict Correction<\/strong> (spelling \/ grammar \/ punctuation \/ typography), without reformulation.<\/li>\n\n\n\n<li><strong>Pass B \u2014 Coherence<\/strong> (names, variants, local chronology, vague referents).<\/li>\n\n\n\n<li><strong>Pass C \u2014 Controlled Fluidity<\/strong> (micro-adjustments, but with explicit prohibitions: do not change the voice, do not change the meaning).<\/li>\n\n\n\n<li><strong>Pass D \u2014 Orthotypography \/ \u201chouse style\u201d<\/strong> (quotation marks, non-breaking spaces, dashes, numbers, units, capitalization).<\/li>\n\n\n\n<li><strong>Pass E \u2014 Validation<\/strong> (comparison of versions, change log, human decision).<\/li>\n<\/ul>\n\n\n\n<p>That&#039;s the difference between &quot;using AI&quot; and <strong>industrialize quality<\/strong>.<\/p>\n\n\n\n<p><strong>3) Safeguard #1: \u201czero reformulation\u201d when you want control<\/strong><\/p>\n\n\n\n<p>The most profitable prompt for a novelist is often\u2026 the most restrictive:<\/p>\n\n\n\n<p>\u201cCorrect only what is objectively wrong. Do not rephrase. Maintain rhythm and voice. Return a LOG of corrections.\u201d<br>This type of prompt limits creative drift and makes the result <strong>auditable<\/strong>.<\/p>\n\n\n\n<p><strong>4) Traceability: the professional database (Word, Track Changes, Compare)<\/strong><\/p>\n\n\n\n<p>Effective workflows rely on:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Change tracking<\/strong> (to validate\/reject),<\/li>\n\n\n\n<li><strong>Compare \/ Merge<\/strong> (to measure the difference between V0 and V1),<\/li>\n\n\n\n<li>A <strong>changelog<\/strong> (LOG), ideally categorized (typo \/ grammar \/ consistency \/ style).<br>Without traceability, an AI becomes a \u201cblack box editor\u201d. With traceability, it becomes a <strong>assistant controlled<\/strong>.<\/li>\n<\/ul>\n\n\n\n<p><strong>5) Confidentiality: work with extracts (and anonymize if necessary)<\/strong><\/p>\n\n\n\n<p>A novel can contain sensitive elements (people, places, emails, private facts). The most robust practice:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>work by <strong>excerpts<\/strong> (500\u20131500 words),<\/li>\n\n\n\n<li><strong>anonymize<\/strong> names\/contact details if needed (NAME_1, PLACE_1\u2026),<\/li>\n\n\n\n<li>Then reintroduce the real elements.<br>It&#039;s the ideal compromise: quality + caution.<\/li>\n<\/ul>\n\n\n\n<p><strong>6) High-end orthotypography: the finishing touch that makes all the difference<\/strong><\/p>\n\n\n\n<p>After linguistic correction, the typographic finishing touches make the &quot;editing&quot;:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>quotation marks \u00ab\u2026\u00bb and non-breaking spaces,<\/li>\n\n\n\n<li>dashes (dialogues \/ parenthetical remarks),<\/li>\n\n\n\n<li>ellipses (...),<\/li>\n\n\n\n<li>numbers (10,000), units (10 %),<\/li>\n\n\n\n<li>consistency of capitals (Level 0 \/ level 0), etc.<br>The key point: to solidify these choices in a <strong>style sheet<\/strong> (\u201chouse style\u201d) \u2014 otherwise you spend your time arbitrating.<\/li>\n<\/ul>\n\n\n\n<p><strong>7) Benchmark: how to know if a prompt is really good<\/strong><\/p>\n\n\n\n<p>The state of the art \u201cpro\u201d uses a <strong>stable test suite<\/strong> : 10 typical extracts (dialogue, action, description, technical passage, emotion, etc.). We run different prompts\/tools on the same extracts, then we take notes:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>fidelity to meaning,<\/li>\n\n\n\n<li>fidelity to the voice,<\/li>\n\n\n\n<li>linguistic quality,<\/li>\n\n\n\n<li>consistency,<\/li>\n\n\n\n<li>rate of change (too high = danger).<br>This avoids the trap of: \u201cit looks better\u201d without proof.<\/li>\n<\/ul>\n\n\n\n<p><strong>8) Three prompt templates (ready to adapt)<\/strong><\/p>\n\n\n\n<p><strong>A \u2014 Strict correction (objective)<\/strong><\/p>\n\n\n\n<p>\u201cYou are a proofreader. Only correct French spelling\/grammar\/punctuation\/typography. Do not rewrite. Preserve paragraphs and rhythm. Output: corrected text + LOG (error \u2192 correction \u2192 category).\u201d<br><strong>B \u2014 Consistency audit (without rewriting)<\/strong><br>\u201cYou are a consistency reviewer. List of vague references, local contradictions, variants of names\/objects\/units. Do not rewrite. Output: table (problem \u2192 extract \u2192 minimal correction).\u201d<br><strong>C \u2014 Controlled polishing (risk managed)<\/strong><br>\u201cImproves fluency without changing the voice. Prohibited: changing meaning, adding information, standardizing style. Limit: max 10% of modified sentences. Output: text + list of modified sentences.\u201d<\/p>\n\n\n\n<p><strong>Conclusion: true maturity lies in method<\/strong><\/p>\n\n\n\n<p>Yes, the internet is full of prompts. But for a novelist, maturity lies in transforming those ideas into <strong>production line<\/strong> Modularity, logs, validation, custom style, confidentiality, benchmarking. AI then becomes what it should be: a time multiplier\u2026 without stealing your voice.<\/p>","protected":false},"excerpt":{"rendered":"<p>Les IA g\u00e9n\u00e9ratives ont chang\u00e9 la mani\u00e8re de corriger et d\u2019am\u00e9liorer un texte. Mais pour un roman, le vrai enjeu n\u2019est pas \u201cd\u2019avoir le meilleur prompt\u201d : c\u2019est de b\u00e2tir un workflow fiable, tra\u00e7able et compatible avec la voix de l\u2019auteur. Cet article synth\u00e9tise les meilleures pratiques observables en ligne (outils, m\u00e9thodes, garde-fous) et propose [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":4361,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_uag_custom_page_level_css":"","footnotes":""},"categories":[43,10,11,14],"tags":[],"class_list":["post-4360","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-autoedition","category-ia","category-infos-pertinentes","category-processus-decriture"],"blocksy_meta":[],"uagb_featured_image_src":{"full":["https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-99-P.jpg",784,512,false],"thumbnail":["https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-99-P-150x150.jpg",150,150,true],"medium":["https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-99-P-300x196.jpg",300,196,true],"medium_large":["https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-99-P-768x502.jpg",768,502,true],"large":["https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-99-P.jpg",784,512,false],"1536x1536":["https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-99-P.jpg",784,512,false],"2048x2048":["https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-99-P.jpg",784,512,false],"trp-custom-language-flag":["https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-99-P-18x12.jpg",18,12,true],"mailpoet_newsletter_max":["https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-99-P.jpg",784,512,false]},"uagb_author_info":{"display_name":"admin5261","author_link":"https:\/\/laurentchani.fr\/en\/author\/admin5261\/"},"uagb_comment_info":0,"uagb_excerpt":"Les IA g\u00e9n\u00e9ratives ont chang\u00e9 la mani\u00e8re de corriger et d\u2019am\u00e9liorer un texte. Mais pour un roman, le vrai enjeu n\u2019est pas \u201cd\u2019avoir le meilleur prompt\u201d : c\u2019est de b\u00e2tir un workflow fiable, tra\u00e7able et compatible avec la voix de l\u2019auteur. Cet article synth\u00e9tise les meilleures pratiques observables en ligne (outils, m\u00e9thodes, garde-fous) et propose&hellip;","_links":{"self":[{"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/posts\/4360","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/comments?post=4360"}],"version-history":[{"count":3,"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/posts\/4360\/revisions"}],"predecessor-version":[{"id":4364,"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/posts\/4360\/revisions\/4364"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/media\/4361"}],"wp:attachment":[{"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/media?parent=4360"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/categories?post=4360"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/tags?post=4360"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}