Impact of fake quotes on reader trust

“Everybody lies. Even your AI.” Predicts confidently. Delivers fiction. Breaks trust. Let’s talk hallucinations — and how to fix them. Because here’s the thing: Fake quotes. Fake sources. Wrong data. Delivered with bullet points, perfect formatting, and a tone that says: “Trust me — I’m smart.” Here’s the uncomfortable truth: Hallucinations aren’t bugs. They’re baked in. Language models don’t “know” facts. They predict the next most likely word — not the next verified truth. Kind of like the Mandela Effect: Millions remember the Monopoly man with a monocle. (He never had one.) False confidence at scale. Now imagine that — inside your product. That’s what models do. They don’t retrieve facts. They generate believable fiction. And they do it fast, polished, and wrong. Not because they want to lie. But because they were never designed to tell the truth. So what can you do? You use a system that works: MODEL — a framework for diagnosing and reducing hallucinations: 𝗠 — Map the context Where is the AI being used? What kind of risk does hallucination pose? 𝗢 — Observe the patterns Are the hallucinations factual, logical, or structural? 𝗗 — Detect the source Is the issue in training data, system prompts, or missing guardrails? 𝗘 — Evaluate the impact Will users notice? Will they trust you again if they don’t? 𝗟 — Limit the risk Add retrieval, validation, or clearly mark generated content. Because if your model can hallucinate, your product can mislead — even if unintentionally. And in a world full of digital Mandela Effects, clarity isn’t a feature — it’s a responsibility. Let’s build systems that don’t just sound smart. Let’s build ones we can trust. 👇 Drop your best Monopoly persona in the comments, and I’ll send step-by-step tips on how to sanity-check your AI’s answers. — 🙋♀️ I’m Caroline — a PM who believes trust isn’t built on polish. It’s built on clarity, checks, and asking the hard questions early.

As a follow up to yesterday, thought I’d highlight the second of the “7 technologies to watch in 2024”. It’s Tuesday, and Tuesdays are as good as any day to discuss the ever-terrifying topic of “Deep Fakes” ⤵️ 👀 The rise of AI in creating realistic but fake media poses risks for misinformation, especially during political conflicts and elections. 🕵️Experts are developing methods to detect these fakes, but face challenges in creating universal and widely-used detection tools. 😬 This is especially terrifying to me. I was reminded of very relevant quotes like “In a time of universal deceit, telling the truth is a revolutionary act” and “I’m not upset that you lied to me, I’m upset that from now on I can’t believe you.” I could leave it at that and anyone still reading might think… clever and relevant quotes! 🔀 But there’s a plot twist: the first quote is often mis-attributed to Orwell, and the second one to Nietzsche. I’m actually not certain either man said those words, and it was difficult to figure out on a quick internet search. Herein lies the irony. It’s always been easy to spread misinformation, now we’re verging into the ridiculous. When the world is filled with fake tales, you can end up electing dictators, starting or escalating wars, disempowering people, and brewing distrust, even in what is legitimately true. This is particularly scary in the world of scientific and healthcare data, where we already have had to deal with our share of mis-information, from censorship of #Darwinian evolution, to #antivaccine papers created with manipulated data. 🙏 My wish for 2024: Forget the blue checkmarks on social media: let’s use #technology create standards for data veracity. [[In thematic spirit, and because I love a tinge of irony for breakfast, I asked Dall-E to create an image for this LinkedIn post]] https://lnkd.in/ee-BRfHQ

A growing problem in academic publishing: AI-generated fake or irrelevant references. As both book editor and journal editor, I am seeing an alarming increase in submissions containing citations that either do not exist or are completely unrelated to the topic. In many cases, it’s clear that AI tools have been used to generate these references without proper verification. While AI can be a powerful assistant, it cannot replace the researcher’s responsibility to verify every source. Fake or irrelevant citations damage more than just a manuscript: They erode the credibility of the author. They waste reviewer and editorial time. They undermine trust in the academic process. My advice: Use AI cautiously and always cross-check every citation in Google Scholar, Scopus, or the publisher’s database. Only include references you have personally read and verified. Treat your reference list as part of your scholarly identity — because it is. Peer reviewers, editors, and readers will check your citations. Make sure what they find strengthens your work, not your rejection letter.

AI writes the internet. Then reads the internet. Then writes it again. What could go wrong? More than 74 percent of new web content is now generated by AI according to Ahrefs. And most of it is trained on other AI content. We’ve seen this before. Remember the SEO doorway pages? I certainly built a few in the early WebPositionGold days. Endless keyword-stuffed clones built for search engines, not people. They ranked for a moment, but no one stayed. They never persuaded, never sold, and never earned belief. Now we’re facing the same problem. Only this time, it’s happening at scale. When AI trains on AI, signal turns to static. Confidence remains, but accuracy slips. The voice sounds human, but the substance disappears. Trust doesn’t vanish all at once. It fades quietly, then fails suddenly. For brands, here’s what’s at stake: Credibility - One hallucinated fact, one fake quote, or one off-tone paragraph can erode years of earned trust. People won’t think it is a formatting error. Differentiation - If your message sounds like everyone else’s, it won’t be remembered. It will be ignored. Search visibility - Google remembers the doorway era. Now it simply de-ranks low-value content without warning. No penalty. No alert. Just silence. Liability - If your company publishes it, your company owns it. That includes compliance, ethics, and accuracy. So how do you stay trusted in a synthetic web? • Share stories machines cannot fake. Behind-the-scenes footage, customer interviews, and real team voices create experiences and texture that models cannot mimic. • Back it up. Proof builds belief. Use citations, screenshots, and source links. • Be unmistakable. Keep your brand voice sharp, intentional, and consistent. Don’t let AI rewrite your identity. • Be open about AI use. It’s not a vulnerability. It’s a commitment to transparency. • Require human signoff. Every piece of content should have a name and a point of view behind it. Before you publish anything, ask: 1. Where did this information come from? 2. Can I verify it through at least two human-edited sources? 3. Does this reflect our voice and values? 4. Is it accurate, legal, and responsible? 5. Who is accountable for it? AI is not the threat. Autopilot is. The brands that will lead tomorrow won’t rely on volume. They’ll rely on clarity, conviction, and trust. Because in a world full of synthetic noise, the most powerful signal is still something authentic.