{"id":19512,"date":"2026-02-23T12:52:50","date_gmt":"2026-02-23T12:52:50","guid":{"rendered":"https:\/\/smart-impact.ch\/?p=19512"},"modified":"2026-02-20T12:58:33","modified_gmt":"2026-02-20T12:58:33","slug":"sources-fiables-et-ia-comment-demeler-le-vrai-du-faux","status":"publish","type":"post","link":"https:\/\/2025.smart-impact.ch\/en\/articles\/sources-fiables-et-ia-comment-demeler-le-vrai-du-faux\/","title":{"rendered":"Reliable sources and AI: how to sort out the real from the fake?"},"content":{"rendered":"

You probably use ChatGPT or Google Gemini on a daily basis to write e-mails, summarize documents or search for quick information. It's become a reflex for many of us. But have you ever wondered how these tools decide what's true or false?<\/p>\n\n\n\n

In the old world, a reliable source was a book published by a recognized publisher, an article signed by a journalist or a study validated by peers. Today, the definition changes radically. Artificial intelligence doesn't \u201cread\u201d sources like we do; it ingests billions of data points and calculates probabilities.<\/p>\n\n\n\n

This paradigm shift is shaking up our relationship with truth and information. In this article, we'll explore how AI engines are redefining trustworthiness, and how you can navigate this new ecosystem without getting trapped.<\/p>\n\n\n\n

What is a reliable source (before AI)?<\/h2>\n\n\n\n

Historically, assessing the credibility of information was based on tangible, human criteria. These methods were taught at university or journalism schools. It was a solid grid for filtering out the noise.<\/p>\n\n\n\n

Here are the traditional pillars of reliability:<\/p>\n\n\n\n

    \n
  1. The authority<\/strong> Who is the author? Does he or she have the necessary qualifications or experience?<\/li>\n\n\n\n
  2. Objectivity<\/strong> Is the information neutral or does it serve a particular agenda?<\/li>\n\n\n\n
  3. Precision<\/strong> Are the facts verifiable and sourced?<\/li>\n\n\n\n
  4. The news<\/strong> Is the information up to date?<\/li>\n<\/ol>\n\n\n\n

    However, these criteria are showing their limitations in the face of AI. An algorithm doesn't care about the author's degree. It doesn't understand the concept of reputation in the same way as a human. For an AI, a very popular Reddit post can sometimes carry as much statistical weight as a little-cited academic article.<\/p>\n\n\n\n

    Comparison: human vs. algorithmic criteria<\/h3>\n\n\n\n
    Criteria<\/th>Traditional (human) approach<\/th>AI (algorithmic) approach<\/th><\/tr>
    Validity<\/strong><\/td>Based on author\/publisher reputation<\/td>Based on recurring patterns in the data<\/td><\/tr>
    Bias<\/strong><\/td>Critical judgment and context<\/td>Statistical biases inherited from training data<\/td><\/tr>
    Trust<\/strong><\/td>Institutional (Le Monde, EPFL, etc.)<\/td>Probabilistic (The next most likely word)<\/td><\/tr>
    Context<\/strong><\/td>Cultural and nuanced understanding<\/td>Semantic analysis without real awareness<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

    How do AI engines determine reliability?<\/h2>\n\n\n\n

    L\u2019artificial intelligence<\/a> is not looking for \u201ctruth\u201d in the philosophical sense. It looks for statistical consistency. When a language model (LLM) generates a response, it predicts the most plausible sequence of words based on its training.<\/p>\n\n\n\n

    Data analysis and recurrence<\/h3>\n\n\n\n

    Imagine that the AI has read the entire Internet. If 90% of the texts associate \u201csky\u201d with \u201cblue\u201d, the AI will conclude that this is reliable information. Reliability for an AI is often a question of volume and repetition. If false information is repeated enough times on high-traffic sites, it risks being ingested as truth.<\/p>\n\n\n\n

    Human validation (RLHF)<\/h3>\n\n\n\n

    Fortunately, it's not the Wild West. Model builders use techniques such as reinforcement learning from human feedback (RLHF). Humans rate the AI's responses to teach it to favor quality sources and avoid toxic content. This is an essential safeguard, but it is not infallible.<\/p>\n\n\n\n

    The major challenges of AI assessment<\/h2>\n\n\n\n

    While AI is a formidable tool, it poses complex problems when it comes to reliability. The first is the \u201cblack box\u201d effect. We don't always know which precise sources were used to generate a given answer.<\/p>\n\n\n\n

    Algorithmic bias<\/h3>\n\n\n\n

    Algorithms are not neutral. They reflect the data on which they have been trained. If the data contains predominantly Western or English-speaking viewpoints, the AI will tend to consider these perspectives as more \u201creliable\u201d or standard, marginalizing other worldviews.<\/p>\n\n\n\n

    Hallucination and verification<\/h3>\n\n\n\n

    This is the most well-known problem: AI can invent facts with absolute confidence. It can cite sources that don't exist, or attribute quotes to the wrong people. For the user, this makes verification tedious. The generated text can no longer be trusted blindly, even if it looks professional.<\/p>\n\n\n\n

    Graph: public confidence in AI<\/h3>\n\n\n\n

    The graph below illustrates the perceived reliability of AI-generated content according to a recent study (data modeled for the example).<\/p>\n\n\n\n

    Information sector<\/th>Trust in Human Content<\/th>Trust in AI content<\/th><\/tr>
    Health & Medicine<\/strong><\/td>85%<\/td>40%<\/td><\/tr>
    Finance & Economics<\/strong><\/td>78%<\/td>55%<\/td><\/tr>
    General News<\/strong><\/td>65%<\/td>35%<\/td><\/tr>
    Code & Technique<\/strong><\/td>70%<\/td>85%<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

    We can see that while AI is considered very reliable for technical tasks (coding), it is still distrusted for sensitive subjects such as health.<\/p>\n\n\n\n

    READ : How do you structure a site so that it can be cited correctly by AI?<\/a><\/p>\n\n\n\n

    The concrete impact on our company<\/h2>\n\n\n\n

    This redefinition of the reliable source has direct consequences for our daily professional and personal lives.<\/p>\n\n\n\n

    Education and research<\/h3>\n\n\n\n

    Students use AI for their work. The risk is to see the emergence of a generation that no longer knows how to look for information at the primary source, but is content with the synthesis digested by an algorithm. Swiss and European universities need to adapt their curricula to teach AI criticism rather than prohibition.<\/p>\n\n\n\n

    Journalism and media<\/h3>\n\n\n\n

    The media are under pressure. If Google offers an AI-generated direct response (SGE), the user no longer clicks on the newspaper link. This raises an economic question, but also a democratic one: if the primary source disappears for lack of revenue, what will AI train on in the future?<\/p>\n\n\n\n

    Best practices for evaluating sources in the AI era<\/h2>\n\n\n\n

    So how do you get out of it? You can't stop progress, but you can adapt the way you work. Here's a pragmatic approach to using AI without being fooled.<\/p>\n\n\n\n

    1. The triangulation rule<\/h3>\n\n\n\n

    Never rely on a single AI answer for a critical decision. Cross-reference the information. If ChatGPT gives you a key figure for your marketing strategy, ask for its source, then check it on Google or in an official report.<\/p>\n\n\n\n

    2. Understanding the tool's limitations<\/h3>\n\n\n\n

    Use AI for what it does best: synthesize, reformulate, code. Be much more cautious when it comes to precise historical facts, medical data or very recent events (on which it has little hindsight).<\/p>\n\n\n\n

    3. Consult human experts<\/h3>\n\n\n\n

    Human intuition and experience remain irreplaceable when it comes to contextualizing information. AI can tell you which<\/em> happened, but an expert will be able to explain better why<\/em> it's important for your specific business.<\/p>\n\n\n\n

    Towards a new digital hygiene<\/h2>\n\n\n\n

    The notion of a reliable source has not disappeared, but it has become more complex. We are moving from an era of institutional trust to one of continuous verification. AI is a powerful assistant, but it must not become your sole editor.<\/p>\n\n\n\n

    For entrepreneurs and decision-makers, the stakes are high: using the power of AI to save time, while keeping a sharp critical mind to avoid strategic mistakes. The next time you copy and paste an AI answer, take three seconds to ask yourself, \u201cIf this were wrong, what would the consequences be?\u201d<\/p>\n\n\n\n

    Sources and recommended reading<\/h2>\n\n\n\n
      \n
    1. European Commission<\/strong> : Ethical guidelines on AI<\/a> - To understand the regulatory framework for AI trust.<\/li>\n\n\n\n
    2. OpenAI Research<\/a><\/strong> GPT-4 System Card - Technical details on model limits and safety.<\/li>\n\n\n\n
    3. EPFL (\u00c9cole Polytechnique F\u00e9d\u00e9rale de Lausanne)<\/strong> : Center for Digital Trust<\/a> - Research on digital trust and cybersecurity.<\/li>\n\n\n\n
    4. UNESCO<\/strong> : Artificial intelligence in education<\/a> - Analysis of the impact on learning and the reliability of knowledge.<\/li>\n<\/ol>\n\n\n\n

      <\/p>","protected":false},"excerpt":{"rendered":"

      Vous utilisez probablement ChatGPT ou Google Gemini quotidiennement pour r\u00e9diger des e-mails, r\u00e9sumer des documents ou chercher des informations rapides… Read more<\/span><\/a><\/span><\/p>","protected":false},"author":2,"featured_media":19674,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[49,25],"tags":[70,106],"class_list":["post-19512","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai","category-contenu","tag-intelligence-artificielle","tag-redaction"],"acf":[],"_yoast_wpseo_title":"Sources fiables et IA : comment d\u00e9m\u00ealer le vrai du faux ?","_yoast_wpseo_metadesc":"L'IA change notre rapport \u00e0 la v\u00e9rit\u00e9. D\u00e9couvrez comment les algorithmes \u00e9valuent les sources fiables et comment \u00e9viter les pi\u00e8ges.","_yoast_wpseo_focuskw":"sources fiables","_yoast_wpseo_canonical":"","_yoast_wpseo_opengraph-title":"","_yoast_wpseo_opengraph-description":"","_yoast_wpseo_twitter-title":"","_yoast_wpseo_twitter-description":"","_yoast_wpseo_meta-robots-noindex":"","_yoast_wpseo_meta-robots-nofollow":"","_yoast_wpseo_meta-robots-adv":"","_yoast_wpseo_bctitle":"","_yoast_wpseo_metakeywords":"","_links":{"self":[{"href":"https:\/\/2025.smart-impact.ch\/en\/wp-json\/wp\/v2\/posts\/19512","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/2025.smart-impact.ch\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/2025.smart-impact.ch\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/2025.smart-impact.ch\/en\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/2025.smart-impact.ch\/en\/wp-json\/wp\/v2\/comments?post=19512"}],"version-history":[{"count":2,"href":"https:\/\/2025.smart-impact.ch\/en\/wp-json\/wp\/v2\/posts\/19512\/revisions"}],"predecessor-version":[{"id":19675,"href":"https:\/\/2025.smart-impact.ch\/en\/wp-json\/wp\/v2\/posts\/19512\/revisions\/19675"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/2025.smart-impact.ch\/en\/wp-json\/wp\/v2\/media\/19674"}],"wp:attachment":[{"href":"https:\/\/2025.smart-impact.ch\/en\/wp-json\/wp\/v2\/media?parent=19512"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/2025.smart-impact.ch\/en\/wp-json\/wp\/v2\/categories?post=19512"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/2025.smart-impact.ch\/en\/wp-json\/wp\/v2\/tags?post=19512"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}