--- title: Why Do LLMs Lie? | daily.dev description: LLMs hallucinate because they are optimized for plausible text generation rather than factual accuracy. The post breaks down three failure modes — factual... canonical: https://daily.dev/posts/why-do-llms-lie--ntzewhtr7 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Why Do LLMs Lie? | daily.dev og:description: LLMs hallucinate because they are optimized for plausible text generation rather than factual accuracy. The post breaks down three failure modes — factual... og:url: https://daily.dev/posts/why-do-llms-lie--ntzewhtr7 og:image: https://api.daily.dev/og/posts/ntzewhTR7.png og:image:alt: Why Do LLMs Lie? og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Why Do LLMs Lie? **[ByteByteGo](https://daily.dev/sources/bytebytego)** · 14 min read · 2 upvotes · 1 comments ## Summary LLMs hallucinate because they are optimized for plausible text generation rather than factual accuracy. The post breaks down three failure modes — factual hallucinations, faithfulness hallucinations, and fabrications — and explains why token-prediction mechanics and training incentives make models prone to confident guessing. It then covers practical mitigations: RAG to supply grounding evidence, tool use to fetch real customer data, structured prompting to handle uncertainty explicitly, chain-of-thought caveats, and answer verification workflows that check claims and citations before responses reach users. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Why do LLMs hallucinate even when they sound confident? LLMs generate text by predicting the most probable next token, not by verifying facts. Confidence words like 'certainly' are generated language, not evidence of accuracy. Training incentives can also reward guessing over abstaining — if a wrong answer and an admission of uncertainty score equally, the model learns to guess. This produces fluent but factually incorrect outputs with no internal mechanism to separate verified facts from plausible continuations. _Developers shipping LLM features track hallucination research and mitigation patterns on daily.dev._ ### What is the difference between factual hallucination and faithfulness hallucination in LLMs? A factual hallucination contradicts reality — for example, stating a 30-day refund window when the real policy is 14 days. A faithfulness hallucination contradicts the supplied evidence — for example, ignoring a retrieved document that requires an unused account and claiming usage does not matter. Fabrication, such as inventing a confirmation number or a non-existent research paper, falls under factual failures. Both dimensions need separate evaluation. _Teams building RAG pipelines debate these distinctions and share evaluation approaches on daily.dev._ ### How does RAG reduce LLM hallucinations and what are its failure points? RAG retrieves relevant documents and injects them into the model's input before generation, giving the model grounding evidence instead of relying on training memory. Failure points include retrieving a retired or wrong-product policy, splitting related conditions across passage boundaries so only part of a rule is retrieved, and the model misreading a correct passage or adding unsupported promises. Documents need clear product names, effective dates, and approval status to work reliably. _Engineers evaluating RAG architectures for production apps find real-world tradeoffs discussed on daily.dev._ ## Community discussion Top comments from developers on daily.dev. **@kartiknvj** · 0 upvotes > The split between faithfulness hallucinations and outright fabrications is the part most teams skip, and it changes how you catch each one. I've found the citation-verification step catches the confident-but-wrong answers that RAG grounding alone waves through, since the model still paraphrases the retrieved doc into something it never said. Do you gate on citation-faithfulness at response time, or batch-check it after the fact? ## Similar posts on daily.dev - [ChatGPT 5.1 explains why it hallucinates](https://daily.dev/posts/chatgpt-5-1-explains-why-it-hallucinates-5yumdkwxa) · Justin Searls · 6 upvotes · 1 comments - [5 Practical Techniques to Detect and Mitigate LLM Hallucinations Beyond Prompt Engineering](https://daily.dev/posts/5-practical-techniques-to-detect-and-mitigate-llm-hallucinations-beyond-prompt-engineering-c2nv4kkuh) · Machine Learning Mastery · 1 upvotes · 0 comments - [OpenAI Study Investigates the Causes of LLM Hallucinations and Potential Solutions](https://daily.dev/posts/openai-study-investigates-the-causes-of-llm-hallucinations-and-potential-solutions-l15kdmz0z) · InfoQ · 4 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#prompt-engineering](https://daily.dev/tags/prompt-engineering), [#rag](https://daily.dev/tags/rag) [View this post on daily.dev](https://daily.dev/posts/why-do-llms-lie--ntzewhtr7) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Why Do LLMs Lie?","url":"https://daily.dev/posts/why-do-llms-lie--ntzewhtr7","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/why-do-llms-lie--ntzewhtr7"},"datePublished":"2026-09-29T15:36:24.149Z","dateModified":"2026-09-29T15:37:11.903Z","description":"LLMs hallucinate because they are optimized for plausible text generation rather than factual accuracy. The post breaks down three failure modes — factual...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/9b4de0ff3f6c44d9353e5d040723d4e8?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/9b4de0ff3f6c44d9353e5d040723d4e8?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"ByteByteGo","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"ByteByteGo","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/35be29234ee14d01a9cd049c52e12753","url":"https://daily.dev/sources/bytebytego"},"commentCount":1,"discussionUrl":"https://daily.dev/posts/why-do-llms-lie--ntzewhtr7","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":1}],"keywords":"llm,ai-agents,prompt-engineering,rag","timeRequired":"PT14M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"ByteByteGo","item":"https://daily.dev/sources/bytebytego"},{"@type":"ListItem","position":3,"name":"Why Do LLMs Lie?"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/why-do-llms-lie--ntzewhtr7","comment":[{"@type":"Comment","text":"The split between faithfulness hallucinations and outright fabrications is the part most teams skip, and it changes how you catch each one. I’ve found the citation-verification step catches the confident-but-wrong answers that RAG grounding alone waves through, since the model still paraphrases the retrieved doc into something it never said. Do you gate on citation-faithfulness at response time, or batch-check it after the fact?","datePublished":"2026-09-29T19:40:13.155Z","url":"https://daily.dev/posts/ntzewhTR7#c-hcxdtQyZf","author":{"@type":"Person","name":"kartik-nvjk","url":"https://daily.dev/kartiknvj","image":"https://media.daily.dev/image/upload/s--3gGgsVCw--/f_auto/v1781456774/avatars/avatar_TvTVeiMdkRCqWUDullFmy?_a=BAMAMiWQ0"}}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/why-do-llms-lie--ntzewhtr7#faq","mainEntity":[{"@type":"Question","name":"Why do LLMs hallucinate even when they sound confident?","acceptedAnswer":{"@type":"Answer","text":"LLMs generate text by predicting the most probable next token, not by verifying facts. Confidence words like 'certainly' are generated language, not evidence of accuracy. Training incentives can also reward guessing over abstaining — if a wrong answer and an admission of uncertainty score equally, the model learns to guess. This produces fluent but factually incorrect outputs with no internal mechanism to separate verified facts from plausible continuations. Developers shipping LLM features track hallucination research and mitigation patterns on daily.dev."}},{"@type":"Question","name":"What is the difference between factual hallucination and faithfulness hallucination in LLMs?","acceptedAnswer":{"@type":"Answer","text":"A factual hallucination contradicts reality — for example, stating a 30-day refund window when the real policy is 14 days. A faithfulness hallucination contradicts the supplied evidence — for example, ignoring a retrieved document that requires an unused account and claiming usage does not matter. Fabrication, such as inventing a confirmation number or a non-existent research paper, falls under factual failures. Both dimensions need separate evaluation. Teams building RAG pipelines debate these distinctions and share evaluation approaches on daily.dev."}},{"@type":"Question","name":"How does RAG reduce LLM hallucinations and what are its failure points?","acceptedAnswer":{"@type":"Answer","text":"RAG retrieves relevant documents and injects them into the model's input before generation, giving the model grounding evidence instead of relying on training memory. Failure points include retrieving a retired or wrong-product policy, splitting related conditions across passage boundaries so only part of a rule is retrieved, and the model misreading a correct passage or adding unsupported promises. Documents need clear product names, effective dates, and approval status to work reliably. Engineers evaluating RAG architectures for production apps find real-world tradeoffs discussed on daily.dev."}}]} ```