[{"data":1,"prerenderedAt":354},["ShallowReactive",2],{"\u002Fen\u002Fdossiers\u002Fchatgpt-visibility":3,"dossiers-catalogue":321},{"id":4,"title":5,"body":6,"description":285,"extension":286,"image":287,"kind":288,"manufacturer":287,"meta":289,"metaTitle":290,"model":287,"navigation":291,"order":292,"path":293,"publishedAt":294,"relatedDossiers":287,"relatedDrones":287,"relatedServices":287,"seo":295,"sources":301,"specs":287,"stem":318,"translationKey":319,"updatedAt":294,"__hash__":320},"content\u002Fen\u002Fdossiers\u002Fchatgpt-visibility.md","ChatGPT visibility: what actually works, measured",{"type":7,"value":8,"toc":273},"minimark",[9,13,18,21,28,34,41,45,48,63,66,70,73,95,98,105,112,120,126,152,159,163,166,173,176,190,194,197,207,217,234,249,253,256,262],[10,11,12],"p",{},"The subject is young, and the volume of assertion about it far exceeds the volume\nof measurement. This piece states only what can be verified, and flags explicitly\nwhat cannot — including where that contradicts the pages currently ranking best on\nthe question.",[14,15,17],"h2",{"id":16},"how-does-chatgpt-pick-its-sources","How does ChatGPT pick its sources?",[10,19,20],{},"Two mechanisms have to be separated first, because they are constantly conflated.",[10,22,23,27],{},[24,25,26],"strong",{},"Training memory."," The model read the web at a given moment. It can mention your\ncompany without ever fetching your site. That is the most fragile position there\nis: it does not update, it does not depend on you, and it disappears with the next\nmodel.",[10,29,30,33],{},[24,31,32],{},"Retrieval."," When the question calls for it, ChatGPT fetches pages and answers\nfrom them. That is the only half where your site does any work, and the only half\nyou have a grip on.",[10,35,36,37,40],{},"The practical consequence is counter-intuitive: ",[24,38,39],{},"a brand can be cited without its\ndomain ever being retrieved",". A serious audit separates the two; most commercial\npitches merge them.",[14,42,44],{"id":43},"do-you-need-to-be-in-bing","Do you need to be in Bing?",[10,46,47],{},"This is the question nearly every page settles too quickly, in one direction or\nthe other.",[10,49,50,51,54,55,58,59,62],{},"ChatGPT search originally ran on Bing's index. Since then OpenAI has operated\n",[24,52,53],{},"its own crawler and its own index",", and publisher agreements have been added on\ntop. The defensible reading today is a ",[24,56,57],{},"blended stack",": the dependency on Bing\nhas ",[24,60,61],{},"loosened rather than disappeared",".",[10,64,65],{},"What that means concretely: registering your site with Bing Webmaster Tools is\nstill worth the half hour, but it is no longer the single lever it was described\nas in 2024. A page selling you \"Bing optimisation\" as the key to ChatGPT\nvisibility is selling an architecture that has changed.",[14,67,69],{"id":68},"why-can-a-site-rank-on-google-and-be-invisible-to-chatgpt","Why can a site rank on Google and be invisible to ChatGPT?",[10,71,72],{},"This is the most important technical point here, and the only genuinely binary\none.",[10,74,75,78,79,82,83,86,87,91,92,62],{},[24,76,77],{},"AI crawlers do not execute JavaScript."," The joint Vercel and MERJ analysis,\ncovering ",[24,80,81],{},"more than 500 million GPTBot fetches",", found ",[24,84,85],{},"no evidence of\nexecution at all",". The crawler does fetch JavaScript files — around 11.5 % of\nrequests — but never runs them. The finding extends to ",[88,89,90],"code",{},"ClaudeBot"," and\n",[88,93,94],{},"PerplexityBot",[10,96,97],{},"The exception is Gemini, which inherits Googlebot's rendering infrastructure.",[10,99,100,101,104],{},"The consequence is blunt: if your page content only appears once JavaScript has\nrun — a catalogue loaded by API call, prices rendered afterwards, an accordion\nwhose contents mount on open — your site can rank normally on Google, whose\ncrawler does render, and be ",[24,102,103],{},"a blank page"," to ChatGPT, Claude and Perplexity.",[10,106,107,108,111],{},"It is checkable in one command, with no paid tool: request the page while\npresenting as ",[88,109,110],{},"GPTBot",", strip scripts and tags, count what is left. A number close\nto zero means invisible.",[14,113,115,116,119],{"id":114},"should-you-create-an-llmstxt-file","Should you create an ",[88,117,118],{},"llms.txt"," file?",[10,121,122,125],{},[24,123,124],{},"No."," And this is where this piece contradicts head-on several of the pages\nranking best on the query, which still recommend it.",[10,127,128,129,132,133,136,137,139,140,143,144,147,148,151],{},"The available measurements converge and are severe. Across ",[24,130,131],{},"more than 500 million\nAI bot visits"," observed over 90 days, ",[24,134,135],{},"408"," targeted an ",[88,138,118],{},". Ahrefs, over\n137,000 sites, finds that ",[24,141,142],{},"97 % of those files received no traffic",". SE Ranking\nmeasures roughly ",[24,145,146],{},"10 % adoption"," across 300,000 domains: one site in ten has\nposted one, and nobody reads it. No major vendor — OpenAI, Google, Anthropic,\nMeta, Mistral — has committed to consuming it, and John Mueller publicly compared\nit to the ",[88,149,150],{},"meta keywords"," tag.",[10,153,154,155,158],{},"What works instead is more mundane, which is precisely why it sells badly: ",[24,156,157],{},"an\nordinary page",", served as HTML readable without JavaScript, declared in the\nsitemap, and linked from the rest of the site. The file nobody reads becomes a URL\nevery crawler reads.",[14,160,162],{"id":161},"what-are-the-percentages-you-read-everywhere-worth","What are the percentages you read everywhere worth?",[10,164,165],{},"Not much, and it helps to know what produces them.",[10,167,168,169,172],{},"The field is saturated with precise-looking figures — this factor supposedly\naccounts for 55 % of citation probability, that one for 14 %. They almost always\ncome from ",[24,170,171],{},"tool vendors",", who have a direct commercial interest in demonstrating\nthat everything has changed and a subscription is required. Some are credited to\nno source at all.",[10,174,175],{},"The most telling detail is that one of the best-ranking pages on this query itself\nwrites that such percentages \"rest on no publicly verifiable study\" — in an\narticle that quotes one. That is not dishonesty; it is the real difficulty of the\nsubject. Something has to be written, and independent measurement does not exist\nyet.",[10,177,178,179,182,183,186,187,189],{},"The rule I apply, and that you can apply without being technical: before accepting\na figure on this subject, ask ",[24,180,181],{},"who produced it"," and ",[24,184,185],{},"what they sell",". The\nrobust facts in this piece — no JavaScript execution, the inertia of ",[88,188,118],{},",\nthe independence of the agents — come from server log analysis and vendor\ndocumentation, not from marketing studies.",[14,191,193],{"id":192},"what-can-actually-be-measured-and-what-cannot","What can actually be measured, and what cannot?",[10,195,196],{},"Two different questions, and only one is within reach without a subscription.",[10,198,199,202,203,206],{},[24,200,201],{},"\"Am I cited?\""," — querying the models dozens of times, engine by engine, is\nexpensive and needs a dedicated tool. And there is ",[24,204,205],{},"no single ranking",": the same\nsite can be highly visible on one engine and absent from another. Any metric\naggregated over \"the AIs\" is misleading.",[10,208,209,212,213,216],{},[24,210,211],{},"\"Can I be?\""," — this is verifiable for free, and it is the half that yields\nactions rather than a score: text actually served to crawlers, agents named in\n",[88,214,215],{},"robots.txt",", coherent structured data, a canonical reference page, an up-to-date\nsitemap.",[10,218,219,220,225,226,229,230,233],{},"One point almost nobody makes: ",[24,221,222,223],{},"two named agents do not obey ",[88,224,215],{},".\nOpenAI writes that its rules \"may not apply\" to ",[88,227,228],{},"ChatGPT-User",", and Perplexity\nthat ",[88,231,232],{},"Perplexity-User"," generally ignores the file — in both cases because a person\ntriggered the fetch. Declaring them is still worth doing, but it is an intention,\nnot a control.",[10,235,236,237,240,241,244,245,248],{},"And an expensive trap: many sites blocked training crawlers in 2023 and 2024 to\nprotect their content, and excluded themselves from ",[24,238,239],{},"citation"," along the way.\nThese are not the same agents, and OpenAI states that each setting is independent\nof the others. A ",[88,242,243],{},"Disallow"," placed under ",[88,246,247],{},"User-agent: *"," blocks all of them.",[14,250,252],{"id":251},"where-should-you-start","Where should you start?",[10,254,255],{},"With measuring the text served without JavaScript. It is free, it takes a minute,\nand it is binary: as long as it fails, everything else is decoration.",[10,257,258,259,261],{},"Then with ",[88,260,215],{},", naming the search agents explicitly so the intent survives\nthe next refactor.",[10,263,264,265,268,269,272],{},"That is what my diagnosis covers — and it tells you whether you ",[24,266,267],{},"can"," be cited,\nnever whether you ",[24,270,271],{},"are",". Anyone promising you the second measurement for free\nhas not made it.",{"title":274,"searchDepth":275,"depth":275,"links":276},"",2,[277,278,279,280,282,283,284],{"id":16,"depth":275,"text":17},{"id":43,"depth":275,"text":44},{"id":68,"depth":275,"text":69},{"id":114,"depth":275,"text":281},"Should you create an llms.txt file?",{"id":161,"depth":275,"text":162},{"id":192,"depth":275,"text":193},{"id":251,"depth":275,"text":252},"How ChatGPT picks its sources, what makes a site invisible to it, and which widely quoted figures are worth nothing.","md",null,"dossier",{},"ChatGPT visibility, measured",true,4,"\u002Fen\u002Fdossiers\u002Fchatgpt-visibility","2026-08-21",{"keywords":296,"title":5,"description":285},[297,298,299,300,118],"ChatGPT visibility","get cited by ChatGPT","generative engine optimisation","AI search visibility",[302,305,308,311,314],{"label":303,"url":304},"Vercel & MERJ — The rise of the AI crawler","https:\u002F\u002Fvercel.com\u002Fblog\u002Fthe-rise-of-the-ai-crawler",{"label":306,"url":307},"OpenAI — Overview of OpenAI crawlers","https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fbots",{"label":309,"url":310},"Google — AI features and your website","https:\u002F\u002Fdevelopers.google.com\u002Fsearch\u002Fdocs\u002Fappearance\u002Fai-features",{"label":312,"url":313},"Search Engine Land — Does llms.txt matter?","https:\u002F\u002Fsearchengineland.com\u002Fdoes-llms-txt-matter-467740",{"label":315,"url":316,"date":317},"Search Engine Land — Inside ChatGPT's retrieval stack","https:\u002F\u002Fsearchengineland.com\u002Fchatgpt-retrieval-stack-index-cache-pages-485036","2026","en\u002Fdossiers\u002Fchatgpt-visibility","chatgpt-visibility","gmUJL5hQ3qZesG-9NxNqs5UeE3fzZSDtaVWPI2U3nzI",[322,327,328,332,337,342,346,350],{"locale":323,"slug":324,"path":325,"title":326,"translationKey":324},"en","application-mapping","\u002Fen\u002Fdossiers\u002Fapplication-mapping","Application mapping: what it must contain",{"locale":323,"slug":319,"path":293,"title":5,"translationKey":319},{"locale":323,"slug":329,"path":330,"title":331,"translationKey":329},"it-obsolescence","\u002Fen\u002Fdossiers\u002Fit-obsolescence","IT obsolescence: what it really costs you",{"locale":323,"slug":333,"path":334,"title":335,"translationKey":336},"technical-debt-ai","\u002Fen\u002Fdossiers\u002Ftechnical-debt-ai","Technical debt and AI: finally clearing the legacy","tech-debt-ai",{"locale":338,"slug":339,"path":340,"title":341,"translationKey":324},"fr","cartographie-applicative","\u002Ffr\u002Fdossiers\u002Fcartographie-applicative","Cartographie applicative : ce qu'elle doit contenir",{"locale":338,"slug":343,"path":344,"title":345,"translationKey":336},"dette-technique-ia","\u002Ffr\u002Fdossiers\u002Fdette-technique-ia","Dette technique et IA : solder enfin le legacy",{"locale":338,"slug":347,"path":348,"title":349,"translationKey":329},"obsolescence-informatique","\u002Ffr\u002Fdossiers\u002Fobsolescence-informatique","Obsolescence informatique : ce qu'elle vous coûte",{"locale":338,"slug":351,"path":352,"title":353,"translationKey":319},"referencement-chatgpt","\u002Ffr\u002Fdossiers\u002Freferencement-chatgpt","Référencement ChatGPT : ce qui marche, mesuré",1787432465271]