{"id":7643,"date":"2024-09-13T10:30:00","date_gmt":"2024-09-13T14:30:00","guid":{"rendered":"https:\/\/canny.io\/blog\/?p=7643"},"modified":"2026-04-02T19:27:53","modified_gmt":"2026-04-02T23:27:53","slug":"get-chatgpt-to-stop-lying","status":"publish","type":"post","link":"https:\/\/canny.io\/blog\/get-chatgpt-to-stop-lying\/","title":{"rendered":"How to get ChatGPT to stop lying: 7 actionable tips for product&nbsp;managers"},"content":{"rendered":"\n<p>ChatGPT is the rising tide that lifted all AI tools. Many rely on it every day, including <a href=\"https:\/\/canny.io\/blog\/top-ai-tools-product-management\/\">product managers<\/a>. They use <a href=\"https:\/\/canny.io\/blog\/7-product-managers-used-chatgpt-heres-what-happened\/\">ChatGPT<\/a>&nbsp;to:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Research the market, competition, and&nbsp;trends<\/li>\n\n\n\n<li>Create user personas and ideal customer profiles&nbsp;(ICPs)<\/li>\n\n\n\n<li>Organize feature&nbsp;ideas<\/li>\n\n\n\n<li>Analyze user&nbsp;feedback<\/li>\n\n\n\n<li>Prioritize feature&nbsp;requests<\/li>\n\n\n\n<li>Brainstorm user experience optimization&nbsp;ideas<\/li>\n\n\n\n<li>Create project&nbsp;plans<\/li>\n\n\n\n<li>Write support documentation and product release&nbsp;materials<\/li>\n\n\n\n<li>And&nbsp;more<\/li>\n<\/ul>\n\n\n\n<p>However, it often gives false information that defeats the tool&#8217;s&nbsp;purpose.<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full is-resized\"><img decoding=\"async\" width=\"1600\" height=\"900\" data-src=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/chat-gpt-lies.png\" alt=\"\" class=\"wp-image-7712 lazyload\" style=\"--smush-placeholder-width: 1600px; --smush-placeholder-aspect-ratio: 1600\/900;width:600px\" data-srcset=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/chat-gpt-lies.png 1600w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/chat-gpt-lies-768x432.png 768w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/chat-gpt-lies-1536x864.png 1536w\" data-sizes=\"(max-width: 1600px) 100vw, 1600px\" src=\"data:image\/svg+xml;base64,PHN2ZyB3aWR0aD0iMSIgaGVpZ2h0PSIxIiB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciPjwvc3ZnPg==\" \/><\/figure>\n<\/div>\n\n\n<p>For example, you might ask ChatGPT to analyze a collection of customer chats and find feature requests. You can&#8217;t rely on this tool if it misses some or makes up non-existent feature&nbsp;requests.<\/p>\n\n\n\n<p>Eight AI experts shared their ChatGPT misinformation stories with us. They also gave tips for preventing it. Spoiler \u2013 you&nbsp;should:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Clarify your&nbsp;prompts<\/li>\n\n\n\n<li>Manually check the information it gives&nbsp;you<\/li>\n\n\n\n<li>Try specialized AI tools for product&nbsp;managers<\/li>\n<\/ul>\n\n\n\n<p>We&#8217;re sharing seven tips and ten prompts to train GPT to provide only accurate information. But first, let\u2019s dig into ChatGPT and its&nbsp;limitations.<\/p>\n\n\n\n\n\n<h2 class=\"wp-block-heading\">Key takeaways<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>ChatGPT doesn&#8217;t lie on purpose. It fills gaps in its knowledge with confident-sounding&nbsp;guesses.<\/li>\n\n\n\n<li>When accuracy really matters, switch to Thinking mode. It cuts factual errors by 50 to 80% compared to standard&nbsp;responses.<\/li>\n\n\n\n<li>Prompt quality matters more than most people realize. Give ChatGPT a role, background context, and ask for&nbsp;citations.<\/li>\n\n\n\n<li>Go one question at a time. Iterative prompting with course corrections produces more reliable&nbsp;output.<\/li>\n\n\n\n<li>For product work, purpose-built AI tools tend to be more accurate. They&#8217;re trained on your data, not the entire&nbsp;internet.<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Why does ChatGPT give wrong information?<\/strong><\/h2>\n\n\n\n<p>Let\u2019s go back to the beginning. What\u2019s ChatGPT in plain&nbsp;English?<\/p>\n\n\n\n<p>OpenAI built an app called ChatGPT in 2022. They used AI (artificial intelligence) to teach their <a href=\"https:\/\/www.techtarget.com\/whatis\/definition\/large-language-model-LLM\">LLM<\/a> (large language model) to answer your questions. Today, it can also write copy, create images, analyze and translate text, explain concepts, and more. OpenAI took a common chatbot and improved it. That\u2019s the \u201cchat\u201d part of&nbsp;it.<\/p>\n\n\n\n<p>&#8220;<a href=\"https:\/\/zapier.com\/blog\/how-does-chatgpt-work\/\">GPT<\/a>&#8221; stands for generative pre-trained transformer. Today, ChatGPT runs on the GPT-5 family. GPT-5.3 Instant handles everyday tasks for all users. GPT-5.4 Thinking is available on paid plans for complex or high-stakes&nbsp;work.<\/p>\n\n\n\n<p>When ChatGPT exploded, some product experts got pretty concerned about its&nbsp;accuracy.<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p><em>\u201cI fear that we are going to develop products based on completely made-up reports, and nobody (not even the accountable people) will&nbsp;know.\u201d<\/em><\/p>\n<\/blockquote>\n\n\n\n<p><a href=\"https:\/\/www.reddit.com\/r\/ProductManagement\/comments\/11nk3hw\/psa_chatgpt_generates_information_it_has_no\/\">Anonymous Reddit user on&nbsp;r\/ProductManagement<\/a><\/p>\n\n\n\n<p>Some experts have seen the negative outcomes of ChatGPT misinformation&nbsp;first-hand.<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p><em>\u201cUntil you\u2019ve sat aghast at the sight of a confident, detailed, but completely wrong answer, you will have no understanding of the skepticism you need to apply to the guidance it provides. Already losing track of the number of engineers I\u2019ve seen apply <\/em><a href=\"https:\/\/kevinyank.com\/posts\/assume-chatgpt-is-lying\/\"><em>ChatGPT advice<\/em><\/a><em> that turns out to be&nbsp;terrible.\u201d<\/em><\/p>\n<\/blockquote>\n\n\n\n<p><a href=\"https:\/\/www.linkedin.com\/in\/kevinyank\/\">Kevin Yank<\/a>, principal architect, front end, <a href=\"https:\/\/cultureamp.com\/\">Culture&nbsp;Amp<\/a><\/p>\n\n\n\n<p>Companies like <a href=\"https:\/\/www.theverge.com\/2022\/12\/5\/23493932\/chatgpt-ai-generated-answers-temporarily-banned-stack-overflow-llms-dangers\">StackOverflow<\/a> took this very seriously. In 2022, they temporarily banned ChatGPT-generated responses from&nbsp;users.<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p><em>\u201cThe primary problem is that while the answers which ChatGPT produces have a high rate of being incorrect, they typically look like they might be good and the answers are very easy to&nbsp;produce.\u201d&nbsp;<\/em><\/p>\n<\/blockquote>\n\n\n\n<p>Moderator at&nbsp;StackOverflow<\/p>\n\n\n\n<p>ChatGPT has come a long way since its <a href=\"https:\/\/www.forbes.com\/sites\/bernardmarr\/2023\/05\/19\/a-short-history-of-chatgpt-how-we-got-to-where-we-are-today\/\">inception<\/a> but still has some limitations. Here are the main ones you should know&nbsp;about.<\/p>\n\n\n\n<p>1. <strong>Accuracy. <\/strong>ChatGPT may miss important details or misunderstand nuances in language. This may result in incorrect or misleading&nbsp;information.<\/p>\n\n\n\n<ol class=\"wp-block-list\"><\/ol>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Example<strong>:<\/strong> ChatGPT might incorrectly interpret phrases like &#8220;I hate how much I love this feature&#8221; as negative sentiments. The word &#8220;hate&#8221; misleads the positive intent of the&nbsp;statement.<\/li>\n<\/ul>\n\n\n\n<p>2. <strong>Outdated information.<\/strong>\u00a0ChatGPT&#8217;s training data has a cutoff date. The main GPT-5 model has a knowledge cutoff of September 30, 2024. GPT-5.2 and later variants have a cutoff of August 31, 2026. That said, ChatGPT has web search on by default. For most queries, it can access current information&nbsp;directly.<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Example: Without web search enabled, ChatGPT won&#8217;t know about recent updates to competitors&#8217; products or current market trends. Web search is on by default, so this is less of an issue than it used to&nbsp;be.<\/li>\n<\/ul>\n\n\n\n<p>3. <strong>Hallucination.<\/strong>\u00a0<em>Hallucination<\/em>\u00a0is the term for when an AI model generates confident-sounding information that has no factual basis. It&#8217;s not lying in the intentional sense. ChatGPT doesn&#8217;t know it&#8217;s wrong. A related but distinct problem is sycophancy. That&#8217;s when ChatGPT agrees with your incorrect premise rather than correcting you, simply to seem agreeable. GPT-5 reduced sycophantic replies by more than 50% compared to earlier models, but the problem hasn&#8217;t disappeared&nbsp;entirely.<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Example: You ask ChatGPT to summarize a research paper and include citations. It returns a convincing summary with three citations, but two of the papers don&#8217;t exist. The titles, authors, and journal names all look plausible. You&#8217;d only catch the error by clicking the&nbsp;links.<\/li>\n<\/ul>\n\n\n\n<p>4. <strong>Potential bias. <\/strong>If you train ChatGPT on biased data, the responses will reflect that&nbsp;bias.<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Example<strong>:<\/strong> ChatGPT might prioritize certain features based on biased input data. For example, a feature request may contain urgency-based words like \u201ccritical\u201d or \u201cmust-have.\u201d While it could be critical for one user, this doesn\u2019t always mean it\u2019s the most impactful idea. This bias could skew product development&nbsp;decisions.<\/li>\n<\/ul>\n\n\n\n<p>5. <strong>Originality and plagiarism<\/strong>. AI-generated content might unintentionally plagiarize existing&nbsp;content.<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Example: You use an AI tool to generate feature descriptions. The output closely mirrors descriptions from a competitor\u2019s website. This unintentional plagiarism can lead to legal issues and damage the company\u2019s&nbsp;credibility.<\/li>\n<\/ul>\n\n\n\n<p>But is ChatGPT the only one to blame? Let\u2019s dig a little&nbsp;deeper.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Common causes of ChatGPT misinformation<\/h2>\n\n\n\n<p>ChatGPT isn\u2019t perfect. That\u2019s why <a href=\"https:\/\/www.lennysnewsletter.com\/p\/how-close-is-ai-to-replacing-product\">it\u2019s not as close to replacing us<\/a> as we think&nbsp;\ud83d\ude42<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p><br><em>\u201cWhen AI is given a task, it&#8217;s supposed to generate a response based on real-world data. In some cases, however, <a href=\"https:\/\/www.cnet.com\/news\/misinformation\/ai-misinformation-how-it-works-and-ways-to-spot-it\/\">AI will fabricate sources<\/a>. That is, it&#8217;s &#8216;hallucinating.&#8217; This can be references to certain books that don&#8217;t exist or news articles pretending to be from well-known websites like The&nbsp;Guardian.\u201d<\/em><\/p>\n<\/blockquote>\n\n\n\n<p>Oscar Gonzalez, tech news editor,&nbsp;Gizmodo<\/p>\n\n\n\n<p><br>Why does ChatGPT hallucinate? There are a few potential reasons for AI-generated&nbsp;misinformation:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>ChatGPT may <strong>lack context<\/strong>. For example, suppose you ask it to develop a new product launch strategy. In that case, it will give you generic advice based on existing articles on this topic. It might also get some details about your particular product or company&nbsp;wrong.<\/li>\n\n\n\n<li>ChatGPT is trained on vast datasets, both real and fictional. It can give you an answer based on a <strong>fictional<\/strong> scenario found online without realizing it&#8217;s&nbsp;fictional.<\/li>\n\n\n\n<li>The output can be susceptible to the way you <strong>phrase the questions<\/strong>. Small changes in the input prompt can lead to different&nbsp;responses.<\/li>\n<\/ul>\n\n\n\n<p>The scale of this problem has shifted significantly in 2026. With web search enabled, current GPT-5 models produce factual errors on roughly 10% of queries. Without web search, that rate jumps above 40%. Web search is on by default in ChatGPT, which is one reason hallucination rates have dropped so much from earlier generations. Switching to Thinking mode reduces errors by a further 50 to&nbsp;80%.<\/p>\n\n\n\n<p>OpenAI&#8217;s own research explains why hallucinations can&#8217;t be fully eliminated yet. Standard training benchmarks reward models for generating an answer, even when the model is uncertain. The result: ChatGPT guesses with confidence rather than admitting it doesn&#8217;t know. The best countermeasure isn&#8217;t waiting for the problem to be solved. It&#8217;s knowing when to trust the output and when to verify&nbsp;it.<\/p>\n\n\n\n<p>Before you put your tin-foil hats on and shut down ChatGPT forever, let\u2019s try to solve these&nbsp;issues.<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full\"><img decoding=\"async\" width=\"474\" height=\"260\" data-src=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/tinfoil-hat.webp\" alt=\"\" class=\"wp-image-7644 lazyload\" src=\"data:image\/svg+xml;base64,PHN2ZyB3aWR0aD0iMSIgaGVpZ2h0PSIxIiB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciPjwvc3ZnPg==\" style=\"--smush-placeholder-width: 474px; --smush-placeholder-aspect-ratio: 474\/260;\" \/><\/figure>\n<\/div>\n\n\n<h2 class=\"wp-block-heading\"><strong>How to limit false information<\/strong><\/h2>\n\n\n\n<p>You can still use ChatGPT and save a lot of time. Now that you know what to look out for, you can be a little more skeptical. Nevertheless, this doesn\u2019t mean you need to lose all faith and go back to your old ways. Here are seven tips to help you limit misinformation from&nbsp;ChatGPT.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">1. Craft clear and precise prompts<\/h3>\n\n\n\n<p>The better you ask the question, the more accurate of an answer you\u2019ll get. So, get <a href=\"https:\/\/www.linkedin.com\/pulse\/how-craft-perfect-chatgpt-prompt-charlene-li-kowge\/\">ultra-specific with your prompt<\/a>. Include the&nbsp;following:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Give ChatGPT an <strong>identity<\/strong>. Who do you want it to be?\n<ul class=\"wp-block-list\">\n<li>Example: \u201cYou are a product manager at <a href=\"https:\/\/canny.io\">canny.io<\/a>. Your job is to prioritize feature requests based on required effort and potential&nbsp;impact.\u201d<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li>Give <strong>background information<\/strong>. Imagine you\u2019re talking to someone who has no idea what you and your company do. To speed this up, include links or text from your site or help docs in the prompt.\n<ul class=\"wp-block-list\">\n<li>Example: \u201cCanny is a tool for product managers. It helps them analyze and manage user&nbsp;feedback.\u201d<\/li>\n\n\n\n<li>Here\u2019s more about our tool&nbsp;[link].<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li>Give <strong>specifics<\/strong>. Do you need the output to be a certain length? Does it have to follow a specific writing style, tone, or format? Specify all that.\n<ul class=\"wp-block-list\">\n<li>Example: \u201cPlease use simple language, plain English, and conversational tone. The audience for this content is internal only. I need everybody in my company to understand what I\u2019m talking about. Avoid jargon and complex&nbsp;language.\u201d<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li>Avoid anything that ChatGPT can <strong>misinterpret<\/strong>. Be as clear as possible in your&nbsp;prompt.<\/li>\n\n\n\n<li>Ask GPT to ask <strong>you<\/strong> questions.\n<ul class=\"wp-block-list\">\n<li>Example: \u201cPlease ask clarifying questions if anything is unclear. Do you need any other information or context? Please&nbsp;ask.\u201d<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><strong>Talk<\/strong> to GPT like you would to a real person. Use conversational language.\n<ul class=\"wp-block-list\">\n<li>Example: \u201cWhy did you prioritize features in this order? Can you explain your thought&nbsp;process?\u201d<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li>Ask for <strong>examples, proof, and citations<\/strong> with direct links. This will help you assess the accuracy of ChatGPT responses.\n<ul class=\"wp-block-list\">\n<li>Example: \u201cWhat evidence supports this statement [copy-paste part of GPT\u2019s response]? Give me direct links to the source of this&nbsp;information.\u201d<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><strong>Paraphrase<\/strong> your prompt. Sometimes, slight differences in your prompts will make a big difference.\n<ul class=\"wp-block-list\">\n<li>Initial prompt: \u201cPrioritize these features for&nbsp;me.\u201d<\/li>\n\n\n\n<li>Paraphrased prompt: \u201cRank these feature ideas based on how much effort they might potentially&nbsp;take.\u201d<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li>Ask it to do <strong>one thing at a time<\/strong>. Then, ask for the next thing in the follow-up. It\u2019s a chat, remember? \ud83d\ude42\n<ul class=\"wp-block-list\">\n<li>Prompt #1: \u201cRank these feature ideas based on how much effort they might potentially&nbsp;take.\u201d<\/li>\n\n\n\n<li>Prompt #2: \u201cThank you, this makes sense. Now add another ranking factor \u2013 the potential impact of this feature on our customers. Redo the ranking&nbsp;please.\u201d<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li>Ask it to <a href=\"https:\/\/www.zdnet.com\/article\/how-to-write-better-chatgpt-prompts-in-5-steps\/\"><strong>repeat<\/strong><\/a><strong> parts<\/strong> of your original requests back to you. This will help you understand if ChatGPT is on the right track.\n<ul class=\"wp-block-list\">\n<li>Example: \u201cCan you tell me what I asked you to do in your own words? I want to make sure you understand exactly what I need you to do. Are my instructions&nbsp;clear?\u201d<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n<h4 class=\"wp-block-heading\">Use custom instructions to set a baseline<\/h4>\n\n\n\n<p>ChatGPT&#8217;s custom instructions let you set persistent rules that apply to every conversation. You only write them once. They take effect automatically every time you start a new&nbsp;chat.<\/p>\n\n\n\n<p>Open ChatGPT, go to Settings, and select Personalize. You&#8217;ll see two fields: what you want ChatGPT to know about you, and how you want it to respond. Here&#8217;s a starting template for product&nbsp;managers:<\/p>\n\n\n\n<p><strong>What ChatGPT should know:<\/strong><br>&#8220;I&#8217;m a product manager at a B2B SaaS company. I work on [product area]. Our customers are [describe]. When you don&#8217;t know something, say so. Never guess and present it as fact. Always tell me when information might be&nbsp;outdated.&#8221;<\/p>\n\n\n\n<p><strong>How ChatGPT should respond:<\/strong><br>&#8220;Be concise. Cite your sources when making factual claims. If you&#8217;re uncertain about something, flag it clearly. Ask clarifying questions before giving a long&nbsp;answer.&#8221;<\/p>\n\n\n\n<p>This doesn&#8217;t eliminate hallucinations. It does reduce them. You&#8217;re setting a consistent expectation before every session&nbsp;starts.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">2. Try iterative prompting<\/h3>\n\n\n\n<p>As you chat with ChatGPT, you\u2019ll start noticing where it goes off the rails. This is a perfect time to bring it back on track. Engineers call this <a href=\"https:\/\/indeemo.com\/blog\/iterative-prompting-generative-ai\">iterative prompting<\/a>. This is the process of asking ChatGPT for one thing only. Then, based on its response, either help it change direction or ask it to keep&nbsp;going.<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full is-resized\"><a href=\"https:\/\/www.cut-the-saas.com\/informative\/what-is-prompt-engineering\"><img decoding=\"async\" width=\"1600\" height=\"1365\" data-src=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/prompt-engineering.png\" alt=\"Graphic illustrating the prompt engineering process. \" class=\"wp-image-7723 lazyload\" style=\"--smush-placeholder-width: 1600px; --smush-placeholder-aspect-ratio: 1600\/1365;width:600px\" data-srcset=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/prompt-engineering.png 1600w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/prompt-engineering-768x655.png 768w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/prompt-engineering-1536x1310.png 1536w\" data-sizes=\"(max-width: 1600px) 100vw, 1600px\" src=\"data:image\/svg+xml;base64,PHN2ZyB3aWR0aD0iMSIgaGVpZ2h0PSIxIiB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciPjwvc3ZnPg==\" \/><\/a><figcaption class=\"wp-element-caption\">Source: Cut the SaaS<\/figcaption><\/figure>\n<\/div>\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p><em>\u201cWe don\u2019t just have a single-stage process. We\u2019re not going straight to the API and asking: \u2018What is the feedback here?\u2019 or \u2018Is there a bug report in this?\u2019 Instead, we have a multi-stage process. We ask one small question at a time and try to get the most accurate response possible. This is how we get higher fidelity and accuracy&nbsp;rates.\u201d<\/em><\/p>\n<\/blockquote>\n\n\n\n<p><a href=\"https:\/\/www.linkedin.com\/in\/niall-dickin\/\">Niall Dickin<\/a>, engineer at <a&nbsp;href=\"https:\/\/canny.io\/\">Canny<\/a><\/p>\n\n\n\n<h3 class=\"wp-block-heading\">3. Train ChatGPT on specific data<\/h3>\n\n\n\n<p>ChatGPT can give you wrong answers when it lacks context about your&nbsp;business.<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p><em>\u201cSay, for instance, you\u2019re running a small business and use ChatGPT for business planning. How much would ChatGPT know about the dynamics of your business? If you&#8217;re integrating ChatGPT into your customer support service, how much would ChatGPT know about your company and product? If you use ChatGPT to create personalized documents, how much would ChatGPT know about you? The short answer? Very&nbsp;little.\u201d<\/em><\/p>\n<\/blockquote>\n\n\n\n<p><a href=\"https:\/\/www.linkedin.com\/in\/maxwell-timothy-techcontentwriter\/\">Maxwell Timothy<\/a>, content and outreach specialist at <a href=\"https:\/\/www.chatbase.co\/\">Chatbase<\/a>.&nbsp;<\/p>\n\n\n\n<p>ChatGPT relies on the data it can find online to answer your questions. Very often, the data it needs isn\u2019t publicly available. Specifics about your product can hide in your help docs and internal wikis. You need to \u201cfeed\u201d that data to ChatGPT to help it help you. There are a few ways of doing&nbsp;this.<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">Manually copy-paste<\/h4>\n\n\n\n<p>This isn\u2019t the most efficient option, but it works. Manually copy-paste relevant information to your conversation with ChatGPT. Ask it to use this information to answer your&nbsp;questions.<\/p>\n\n\n\n<p><strong>Sample&nbsp;prompt:<\/strong><\/p>\n\n\n\n<p>\u201cAnalyze these Intercom conversations. Find feature requests in them [insert Intercom conversations\u2019 transcript]. Compare them against our existing features [insert a list of features]. Give me a list of only feature requests that we don\u2019t already&nbsp;have.\u201d<\/p>\n\n\n\n<p>Provide your own examples to clarify the prompt. Let\u2019s say you\u2019re asking ChatGPT to create a changelog entry. Copy-paste an existing changelog entry and ask ChatGPT to follow the same style, format, length, tone,&nbsp;etc.<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p><em>\u201cThere\u2019s this temptation to type in as little as possible and let the AI do its \u201cmagic.\u201d And then we expect accurate responses back. We have to keep in mind: most AI tools have a pretty large \u201ccontext window\u201d \u2013&nbsp;space to type in our prompt. These tools can consume a lot of data at&nbsp;once.\u201d<\/em><\/p>\n<\/blockquote>\n\n\n\n<p>Maxwell&nbsp;Timothy<\/p>\n\n\n\n<p>Try feeding your data to ChatGPT in portions, though. Sometimes, large amounts of data at once lead to \u201challucinations\u201d as&nbsp;well.<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">Create custom GPTs yourself<\/h4>\n\n\n\n<p>OpenAI now allows you to <a href=\"https:\/\/openai.com\/index\/introducing-gpts\/\">create custom GPTs<\/a>. Think of them as mini-programs you can train on specific tasks. For example, you can create separate GPTs&nbsp;for:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Data&nbsp;analysis<\/li>\n\n\n\n<li>Feature request&nbsp;detecting<\/li>\n\n\n\n<li>Release note&nbsp;writing<\/li>\n\n\n\n<li>Client conversation&nbsp;breakdown<\/li>\n\n\n\n<li>And much&nbsp;more<\/li>\n<\/ul>\n\n\n\n<p>Creating custom GPTs saves you time. You won\u2019t need to explain what you need ChatGPT to do for you every time. You set it up once and reuse it&nbsp;forever.<\/p>\n\n\n\n<p>Follow these steps to create a custom&nbsp;GPT.<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>Open <a&nbsp;href=\"https:\/\/chatgpt.com\/\">https:\/\/chatgpt.com\/<\/a><\/li>\n\n\n\n<li>Find \u201cExplore GPTs\u201d on the left-hand&nbsp;side<\/li>\n\n\n\n<li>Find \u201cCreate\u201d in the top-right&nbsp;corner<\/li>\n\n\n\n<li>You\u2019ll end up on the \u201cCreate\u201d tab. You can describe to ChatGPT what kind of GPT you\u2019d like to create and share resources with&nbsp;it.<\/li>\n\n\n\n<li>Alternatively, you can click on the \u201cConfigure tab\u201d and customize your GPT&nbsp;there<\/li>\n<\/ol>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full is-resized\"><img decoding=\"async\" width=\"1059\" height=\"366\" data-src=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/image.png\" alt=\"\" class=\"wp-image-7645 lazyload\" style=\"--smush-placeholder-width: 1059px; --smush-placeholder-aspect-ratio: 1059\/366;width:600px\" data-srcset=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/image.png 1059w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/image-768x265.png 768w\" data-sizes=\"(max-width: 1059px) 100vw, 1059px\" src=\"data:image\/svg+xml;base64,PHN2ZyB3aWR0aD0iMSIgaGVpZ2h0PSIxIiB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciPjwvc3ZnPg==\" \/><figcaption class=\"wp-element-caption\">The Explore GPTs section in ChatGPT<\/figcaption><\/figure>\n<\/div>\n\n\n<h4 class=\"wp-block-heading\">Use tools to create custom GPTs<\/h4>\n\n\n\n<p>There are some great tools for creating custom GPTs. <a href=\"https:\/\/www.linkedin.com\/in\/jwestdigital\/\">Jason West<\/a>, CEO of <a href=\"http:\/\/fastbots.ai\">FastBots.ai<\/a>, walks you through creating one with <a href=\"http:\/\/customgpt.ai\">CustomGPT.ai<\/a> in this&nbsp;video.<\/p>\n\n\n\n<center><iframe width=\"650\" height=\"315\" align=\"centre\" data-src=\"https:\/\/www.youtube.com\/embed\/y9Gi3UgNA3E?si=okVEA6YX4dxSoPs7\" title=\"YouTube video player\" frameborder=\"0\" allow=\"accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share\" allowfullscreen=\"\" src=\"data:image\/svg+xml;base64,PHN2ZyB3aWR0aD0iMSIgaGVpZ2h0PSIxIiB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciPjwvc3ZnPg==\" class=\"lazyload\" data-load-mode=\"1\"><\/iframe><\/center>\n\n\n\n<p><a href=\"https:\/\/www.chatbase.co\/blog\/how-to-train-chatgpt-with-your-data\">Chatbase<\/a> is another great option. It helps you train your chatbots with company-specific information and&nbsp;knowledge.<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<center><iframe width=\"650\" height=\"315\" align=\"centre\" data-src=\"https:\/\/www.youtube.com\/embed\/ASDnfvc8XfI?si=lzDb0-RRqBsXdCDF\" title=\"YouTube video player\" frameborder=\"0\" allow=\"accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share\" allowfullscreen=\"\" src=\"data:image\/svg+xml;base64,PHN2ZyB3aWR0aD0iMSIgaGVpZ2h0PSIxIiB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciPjwvc3ZnPg==\" class=\"lazyload\" data-load-mode=\"1\"><\/iframe><\/center>\n<\/blockquote>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p><em>\u201cChatbase is the easiest way to train and deploy a chatbot with your data. This innovative no-code AI solution provides a simple way to manage all aspects of building a chatbot with your data. This includes training, configuration, and&nbsp;deployment.\u201d<\/em><\/p>\n<\/blockquote>\n\n\n\n<p>Maxwell&nbsp;Timothy<\/p>\n\n\n\n<p>Chatbase uses the same technology that powers ChatGPT but optimizes it to make it even easier to&nbsp;use.<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">Try RAG (retrieval-augmented generation)<\/h4>\n\n\n\n<p><a href=\"https:\/\/blogs.nvidia.com\/blog\/what-is-retrieval-augmented-generation\/\">Retrieval-augmented generation (RAG)<\/a> is a more advanced way of reducing AI hallucinations. It allows AI to respond to queries referencing a specified set of&nbsp;documents.<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p><em>\u201c<\/em><a href=\"https:\/\/arxiv.org\/html\/2406.04744v1\"><em>RAG<\/em><\/a><em> has recently emerged as a promising solution to alleviate the large language model\u2019s (LLM) lack of&nbsp;knowledge.\u201d<\/em><\/p>\n<\/blockquote>\n\n\n\n<p>At Canny, we built this into Autopilot through the&nbsp;<a href=\"https:\/\/help.canny.io\/en\/articles\/9893990-the-autopilot-knowledge-hub\">Knowledge Hub<\/a>. You can upload your own help docs, release notes, and product specs directly. Autopilot uses that context to distinguish between new feature requests and things you already support. It also improves the accuracy of Smart Replies. The more context you give it, the fewer gaps it fills with&nbsp;guesses.<\/p>\n\n\n\n<p><strong>Note<\/strong>: we\u2019re not using any data without explicit permission from our users. We only use customer data for that customer\u2019s&nbsp;instance.<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p><em>\u201cWe provide relevant context on each prompt to supplement the LLM knowledge with domain-specific data from the customer. This helps the LLM stay grounded in reality and feed from this data to generate context-aware&nbsp;responses.\u201d<\/em><\/p>\n<\/blockquote>\n\n\n\n<p><a href=\"https:\/\/www.linkedin.com\/in\/oliverafedi\/\">Ramiro Olivera<\/a>, engineer at <a href=\"https:\/\/canny.io\/\">Canny<\/a>&nbsp;<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">4. Use web search and Thinking mode<\/h3>\n\n\n\n<p>This is the most impactful change you can make right now, and it requires no prompt&nbsp;engineering.<\/p>\n\n\n\n<p>ChatGPT&#8217;s web search is on by default. It lets the model pull live, sourced information rather than relying solely on its training data. With web search enabled, current GPT-5 models produce factual errors on roughly 10% of queries. Disable it, and that rate climbs above 40%. Leave web search on for any query where accuracy&nbsp;matters.<\/p>\n\n\n\n<p>Thinking mode goes further. When you select Thinking in the model picker, ChatGPT works through the problem step by step before generating a response. This produces measurably more accurate output. OpenAI&#8217;s own benchmarks show that Thinking mode cuts factual errors by 50 to 80% compared to standard responses. Use it for anything high-stakes: competitive analysis, market research, technical claims, product&nbsp;decisions.<\/p>\n\n\n\n<p>The combination of web search and Thinking mode is the closest thing to a reliable ChatGPT that currently exists. Neither eliminates hallucinations. Together, they reduce them&nbsp;significantly.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">5. Recognize and correct errors<\/h3>\n\n\n\n<p>It\u2019s easier to fix ChatGPT\u2019s mistakes when you know what to look out for. Here are some common signals that you might be getting the wrong&nbsp;information.<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>No source attribution \u2013&nbsp;ChatGPT can&#8217;t give you a direct link to the source of the information. Sometimes, it&#8217;ll give you a link that leads to a 404&nbsp;page.&nbsp;<\/li>\n\n\n\n<li>Inconsistency with well-known&nbsp;facts.<\/li>\n\n\n\n<li>Overly broad&nbsp;statements.<\/li>\n\n\n\n<li>Contradictory information \u2013 sometimes, you can get different responses when you use rephrased&nbsp;prompts.<\/li>\n\n\n\n<li>Outdated references. It\u2019s best to trust information that\u2019s no more than five years&nbsp;old.<\/li>\n\n\n\n<li>Citations to non-credible&nbsp;sources.<\/li>\n<\/ul>\n\n\n\n<p>To correct any errors, verify the facts manually and only trust reputable&nbsp;sources.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">6. Verify facts manually<\/h3>\n\n\n\n<p>Yes, ChatGPT and all AI are here to replace manual work. But, as you can see, we\u2019re not 100% there yet. Because ChatGPT can still make mistakes, it\u2019s best to verify critical information. Checking it early will save you time in the&nbsp;future.<\/p>\n\n\n\n<p><a href=\"https:\/\/www.linkedin.com\/in\/kevinyank\/\">Kevin Yank<\/a>, principal architect at <a href=\"https:\/\/www.cultureamp.com\/\">Culture Amp<\/a>, recommends always <a href=\"https:\/\/kevinyank.com\/posts\/assume-chatgpt-is-lying\/\">assuming ChatGPT is lying<\/a>. This level of skepticism will help minimize&nbsp;errors.<\/p>\n\n\n\n<p>The product management community on Reddit agrees, and here\u2019s what they&nbsp;recommend.<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>Ask ChatGPT: \u201cAre you sure about&nbsp;that?\u201d<\/li>\n\n\n\n<li>If you get a different response, go and check this information&nbsp;manually<\/li>\n<\/ol>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full is-resized\"><img decoding=\"async\" width=\"1544\" height=\"758\" data-src=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/reddit-chatgpt-manual-review.png\" alt=\"Reddit conversation showing how a user asks ChatGPT to verify its sure about its answers. \" class=\"wp-image-7734 lazyload\" style=\"--smush-placeholder-width: 1544px; --smush-placeholder-aspect-ratio: 1544\/758;width:600px\" data-srcset=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/reddit-chatgpt-manual-review.png 1544w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/reddit-chatgpt-manual-review-768x377.png 768w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/reddit-chatgpt-manual-review-1536x754.png 1536w\" data-sizes=\"(max-width: 1544px) 100vw, 1544px\" src=\"data:image\/svg+xml;base64,PHN2ZyB3aWR0aD0iMSIgaGVpZ2h0PSIxIiB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciPjwvc3ZnPg==\" \/><\/figure>\n<\/div>\n\n\n<p>Source: <a href=\"https:\/\/www.reddit.com\/r\/ProductManagement\/comments\/1efc3ps\/has_chatgpt_ever_lied_to_you\/\">Reddit<\/a>&nbsp;<\/p>\n\n\n\n<p><strong>Bottom line<\/strong>: check all critical&nbsp;information<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">Sources for fact-checking<\/h4>\n\n\n\n<p>If you\u2019re unsure about any information ChatGPT gives you, verify it. For example, you can use these trustworthy sources to check market trends, competitive intel, and similar&nbsp;data.<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><a href=\"https:\/\/www.statista.com\/\">Statista<\/a>, <a href=\"https:\/\/www.gartner.com\/en\">Gartner<\/a>, or <a href=\"https:\/\/www.forrester.com\/bold\">Forrester<\/a> for data, numbers,&nbsp;benchmarks<\/li>\n\n\n\n<li><a href=\"https:\/\/www.emarketer.com\/\">eMarketer<\/a> or <a href=\"https:\/\/www.mckinsey.com\/\">McKinsey &amp; Company<\/a> for business and marketing trends and&nbsp;forecasts<\/li>\n\n\n\n<li><a href=\"https:\/\/www.wired.com\/\">Wired<\/a>, <a href=\"https:\/\/techcrunch.com\/\">TechCrunch<\/a>, or <a href=\"https:\/\/www.cnet.com\/\">CNET<\/a> for technology&nbsp;information<\/li>\n\n\n\n<li><a href=\"https:\/\/aipmm.com\/\">Association of International Product Marketing and Management (AIPMM)<\/a> for product&nbsp;management&nbsp;<\/li>\n\n\n\n<li>Reddit and LinkedIn groups for industry-specific information\n<ul class=\"wp-block-list\">\n<li>Don\u2019t trust everything that any contributor posts. Ask the community if you\u2019re not&nbsp;sure.<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n<p><strong>Note<\/strong>: if you provide your own data to ChatGPT, you need to check this data internally to ensure you get an accurate&nbsp;output.<\/p>\n\n\n\n<p>Here\u2019s how <a href=\"https:\/\/www.linkedin.com\/in\/gianluca-ferruggia-a01a276b\/\">Gianluca Ferruggia<\/a>, general manager at <a href=\"https:\/\/www.designrush.com\/\">DesignRush<\/a>, corrects ChatGPT misinformation. He shared a story from his recent product&nbsp;launch.<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p><em>\u201cWe were coordinating a product launch using AI. The response we received conflicted with our internal project milestones due to the AI&#8217;s misinterpretation. We rectified this by reminding ourselves that it&#8217;s crucial to provide AI with specific, clear, and concise&nbsp;instructions.\u201d<\/em><\/p>\n<\/blockquote>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full is-resized\"><img decoding=\"async\" width=\"1600\" height=\"735\" data-src=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/chatgpt-product-launch1.png\" alt=\"\" class=\"wp-image-7735 lazyload\" style=\"--smush-placeholder-width: 1600px; --smush-placeholder-aspect-ratio: 1600\/735;width:600px\" data-srcset=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/chatgpt-product-launch1.png 1600w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/chatgpt-product-launch1-768x353.png 768w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/chatgpt-product-launch1-1536x706.png 1536w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/chatgpt-product-launch1-1200x550.png 1200w\" data-sizes=\"(max-width: 1600px) 100vw, 1600px\" src=\"data:image\/svg+xml;base64,PHN2ZyB3aWR0aD0iMSIgaGVpZ2h0PSIxIiB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciPjwvc3ZnPg==\" \/><\/figure>\n<\/div>\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full is-resized\"><img decoding=\"async\" width=\"1600\" height=\"1010\" data-src=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/chatgpt-product-launch2.png\" alt=\"\" class=\"wp-image-7736 lazyload\" style=\"--smush-placeholder-width: 1600px; --smush-placeholder-aspect-ratio: 1600\/1010;width:600px\" data-srcset=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/chatgpt-product-launch2.png 1600w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/chatgpt-product-launch2-768x485.png 768w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/chatgpt-product-launch2-1536x970.png 1536w\" data-sizes=\"(max-width: 1600px) 100vw, 1600px\" src=\"data:image\/svg+xml;base64,PHN2ZyB3aWR0aD0iMSIgaGVpZ2h0PSIxIiB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciPjwvc3ZnPg==\" \/><\/figure>\n<\/div>\n\n\n<h3 class=\"wp-block-heading\">7. Use dedicated AI tools<\/h3>\n\n\n\n<p>ChatGPT made a massive leap in AI adoption and progress. However, it\u2019s imperfect and can sometimes give you the wrong information. That\u2019s why there are many specialized AI tools. Let\u2019s look at a few of&nbsp;them.<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">AI for product managers: Canny Autopilot&nbsp;<\/h4>\n\n\n\n<p>ChatGPT isn\u2019t focused on product&nbsp;management.<\/p>\n\n\n\n<p>To help product managers take advantage of AI, we created <a href=\"https:\/\/canny.io\/blog\/introducing-autopilot\/\">Autopilot<\/a>. It\u2019s a suite of AI-powered tools that helps product&nbsp;managers.<\/p>\n\n\n\n<p>Autopilot\u2019s Feedback Discovery feature detects customer feedback&nbsp;in:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Customer conversations (Intercom, Zendesk,&nbsp;HelpScout)<\/li>\n\n\n\n<li>Sales calls&nbsp;(Gong)<\/li>\n\n\n\n<li>Public review sites (G2, Capterra, and eight more&nbsp;sources)<\/li>\n<\/ul>\n\n\n\n<p>Then, Autopilot extracts that feedback and imports it into your Canny account. Next, it deduplicates that feedback and automatically merges duplicate requests. If you&#8217;re using Canny Ideas and have set up groups for your product areas, it auto-groups your feedback too.&nbsp;<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full is-resized\"><img decoding=\"async\" width=\"2000\" height=\"883\" data-src=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/Autopilot-Screenshot.png\" alt=\"\" class=\"wp-image-7648 lazyload\" style=\"--smush-placeholder-width: 2000px; --smush-placeholder-aspect-ratio: 2000\/883;width:600px\" data-srcset=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/Autopilot-Screenshot.png 2000w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/Autopilot-Screenshot-768x339.png 768w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/Autopilot-Screenshot-1536x678.png 1536w\" data-sizes=\"(max-width: 2000px) 100vw, 2000px\" src=\"data:image\/svg+xml;base64,PHN2ZyB3aWR0aD0iMSIgaGVpZ2h0PSIxIiB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciPjwvc3ZnPg==\" \/><figcaption class=\"wp-element-caption\">Feedback Autopilot found in Intercom and Gong conversations. <\/figcaption><\/figure>\n<\/div>\n\n\n<p>Autopilot can&nbsp;also:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Reply to your users on your behalf, asking clarifying&nbsp;questions<\/li>\n\n\n\n<li>Summarize long comment&nbsp;threads<\/li>\n\n\n\n<li>Use the\u00a0<a href=\"https:\/\/help.canny.io\/en\/articles\/9893990-the-autopilot-knowledge-hub\">Knowledge Hub<\/a>\u00a0to learn your product so it can spot the difference between existing features and genuinely new&nbsp;requests<br><\/li>\n<\/ul>\n\n\n\n<center><iframe width=\"650\" height=\"315\" align=\"centre\" data-src=\"https:\/\/www.youtube.com\/embed\/mEoR9aDiMXs?si=G3dANpd6FYemr2JQ\" title=\"YouTube video player\" frameborder=\"0\" allow=\"accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share\" allowfullscreen=\"\" src=\"data:image\/svg+xml;base64,PHN2ZyB3aWR0aD0iMSIgaGVpZ2h0PSIxIiB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciPjwvc3ZnPg==\" class=\"lazyload\" data-load-mode=\"1\"><\/iframe><\/center>\n\n\n\n<p>We\u2019ve received very positive feedback from our Autopilot customers so far. Autopilot uses a multi-stage process to detect and extract user feedback, which makes it much more&nbsp;accurate.<\/p>\n\n\n\n<p><em>\u201cI thought, surely I can\u2019t just turn it on, and it\u2019ll do its magic. But that\u2019s exactly what it\u2019s doing. We\u2019re seeing hundreds of support tickets turned into actionable insights&#8230;with very high&nbsp;accuracy.\u201d<\/em><\/p>\n\n\n\n<p><a href=\"https:\/\/www.linkedin.com\/in\/mattcromwell\/\">Matt Cromwell<\/a>, senior director of customer experience at <a href=\"https:\/\/stellarwp.com\/\">StellarWP<\/a>&nbsp;<\/p>\n\n\n\n<p><\/p>\n\n\n\n<div class=\"hs-cta-embed hs-cta-simple-placeholder hs-cta-embed-165946149675\"\n  style=\"max-width:100%; max-height:100%; margin:0 auto; width:650px;height:277.1953125px\" data-hubspot-wrapper-cta-id=\"165946149675\">\n  <a href=\"https:\/\/cta-service-cms2.hubspot.com\/web-interactives\/public\/v1\/track\/redirect?encryptedPayload=AVxigLKYgg%2FteyNEgNT2DVfJ8A6abMe7ZakH5CZ49il0rC2pp1Ef%2Bc143gjFca7CtNMh6DoMbyJrXgKKKWf4ZLsFHd8xahPME0kXU4epjm1C%2FgPIQUhxcHTnAp4G3O0P6dcj8Ci91Q%3D%3D&#038;webInteractiveContentId=165946149675&#038;portalId=5705808\" target=\"_blank\" rel=\"noopener\" crossorigin=\"anonymous\">\n    <img decoding=\"async\" alt=\"try-autopilot (1)\" data-src=\"https:\/\/no-cache.hubspot.com\/cta\/default\/5705808\/interactive-165946149675.png\" style=\"height: 100%; width: 100%; object-fit: fill\"\n      onerror=\"this.style.display='none'\" src=\"data:image\/svg+xml;base64,PHN2ZyB3aWR0aD0iMSIgaGVpZ2h0PSIxIiB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciPjwvc3ZnPg==\" class=\"lazyload\" \/>\n  <\/a>\n<\/div>\n\n\n\n<h4 class=\"wp-block-heading\">AI for project managers: ClickUp<\/h4>\n\n\n\n<p>Product managers often share the load with project managers. Sometimes, product managers own project management as well. In both cases, a dedicated AI tool can&nbsp;help.<\/p>\n\n\n\n<p>At Canny, we use ClickUp. It helps us manage tasks, collaborate, and track&nbsp;progress.<\/p>\n\n\n\n<p><a href=\"https:\/\/clickup.com\/ai\">ClickUp Brain<\/a> is an advanced AI assistant. It creates documents, brainstorms ideas, summarizes notes, and more. You can ask ClickUp Brain&nbsp;to:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Read your internal documents and answer questions about your&nbsp;company<\/li>\n\n\n\n<li>Give a breakdown of what different teams are working&nbsp;on<\/li>\n\n\n\n<li>Reply to&nbsp;comments<\/li>\n\n\n\n<li>Write a task&nbsp;summary<\/li>\n\n\n\n<li>Create templates, labels, tasks, transcripts, and&nbsp;more<\/li>\n<\/ul>\n\n\n\n<p>Unlike ChatGPT, ClickUp has internal information about your company and projects. It can use that data to help you in a very particular way. Because ClickUp can read your documents, it can create more accurate outputs for your&nbsp;organization.<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter is-resized\"><img decoding=\"async\" width=\"1600\" height=\"1311\" data-src=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/image-1.png\" alt=\"\" class=\"wp-image-7750 lazyload\" style=\"--smush-placeholder-width: 1600px; --smush-placeholder-aspect-ratio: 1600\/1311;width:600px\" data-srcset=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/image-1.png 1600w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/image-1-768x629.png 768w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/image-1-1536x1259.png 1536w\" data-sizes=\"(max-width: 1600px) 100vw, 1600px\" src=\"data:image\/svg+xml;base64,PHN2ZyB3aWR0aD0iMSIgaGVpZ2h0PSIxIiB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciPjwvc3ZnPg==\" \/><\/figure>\n<\/div>\n\n\n<h4 class=\"wp-block-heading\">AI for research: Perplexity<\/h4>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full is-resized\"><img decoding=\"async\" width=\"1567\" height=\"818\" data-src=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/image-40.png\" alt=\"Screenshot of Perplexity's homepage\" class=\"wp-image-11009 lazyload\" style=\"--smush-placeholder-width: 1567px; --smush-placeholder-aspect-ratio: 1567\/818;width:650px\" data-srcset=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/image-40.png 1567w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/image-40-768x401.png 768w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/image-40-1536x802.png 1536w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/image-40-209x110.png 209w\" data-sizes=\"(max-width: 1567px) 100vw, 1567px\" src=\"data:image\/svg+xml;base64,PHN2ZyB3aWR0aD0iMSIgaGVpZ2h0PSIxIiB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciPjwvc3ZnPg==\" \/><\/figure>\n<\/div>\n\n\n<p>ChatGPT is useful for research, but it doesn&#8217;t always show its work.&nbsp;<a href=\"https:\/\/www.perplexity.ai\">Perplexity<\/a>&nbsp;takes a different approach. It searches the web in real time and cites every claim it makes. You can click through to verify any source&nbsp;directly.<\/p>\n\n\n\n<p>For PMs doing competitive research, market analysis, or trend tracking, this is a significant accuracy advantage. You&#8217;re not trusting a summary. You&#8217;re reading the sources&nbsp;yourself.<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">AI for product analytics: Mixpanel<\/h4>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full is-resized\"><img decoding=\"async\" width=\"1562\" height=\"809\" data-src=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/image-42.png\" alt=\"A screenshot of Mixpanel's Spark AI page\" class=\"wp-image-11011 lazyload\" style=\"--smush-placeholder-width: 1562px; --smush-placeholder-aspect-ratio: 1562\/809;width:650px\" data-srcset=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/image-42.png 1562w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/image-42-768x398.png 768w, https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/image-42-1536x796.png 1536w\" data-sizes=\"(max-width: 1562px) 100vw, 1562px\" src=\"data:image\/svg+xml;base64,PHN2ZyB3aWR0aD0iMSIgaGVpZ2h0PSIxIiB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciPjwvc3ZnPg==\" \/><\/figure>\n<\/div>\n\n\n<p>Many product decisions start with a question: how are users actually behaving?&nbsp;<a href=\"https:\/\/mixpanel.com\/spark-ai\/\">Mixpanel&#8217;s Spark AI<\/a>&nbsp;lets you ask those questions in plain language, without writing SQL or digging through dashboards. You ask something like &#8220;which features drive the most retention?&#8221; and it builds the report for you. It shows its work so you can verify the&nbsp;query.<\/p>\n\n\n\n<p>Because Spark runs against your own product data, not the open internet, it can&#8217;t fabricate answers. It either finds a pattern in your data or it tells you it&nbsp;can&#8217;t.<\/p>\n\n\n\n<p><\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Accurate information is key<\/strong><\/h2>\n\n\n\n<p>ChatGPT or any AI technology can sometimes make mistakes. You can still use AI to save you time, but you need to question accuracy and critically assess AI-generated output and&nbsp;content.<\/p>\n\n\n\n<p>If the output you\u2019re getting is misleading, you might spend more time correcting it later. Worse, you might act on that misinformation. Look for common signs of AI misinformation and verify all&nbsp;facts.<\/p>\n\n\n\n<p>If you want a tool that\u2019s already doing this for you, <a href=\"https:\/\/go.canny.io\/autopilot\">try Autopilot<\/a>! Stay tuned for more updates and&nbsp;improvements.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Frequently asked questions<\/h2>\n\n\n\n<div data-schema-only=\"false\" class=\"wp-block-aioseo-faq\"><h3 class=\"aioseo-faq-block-question\">Why does ChatGPT lie?<\/h3><div class=\"aioseo-faq-block-answer\">\n<p>ChatGPT doesn&#8217;t lie on purpose. It predicts the most plausible next word based on its training data. When it lacks reliable information, it fills the gap with a confident-sounding guess. This is called a hallucination. It looks like a lie, but it&#8217;s more like a very convincing&nbsp;bluff.<\/p>\n<\/div><\/div>\n\n\n\n<div data-schema-only=\"false\" class=\"wp-block-aioseo-faq\"><h3 class=\"aioseo-faq-block-question\">Does ChatGPT still hallucinate?<\/h3><div class=\"aioseo-faq-block-answer\">\n<p>Yes, but significantly less than before. Web search is on by default in current ChatGPT models, which is the primary reason accuracy has improved. With web search enabled, GPT-5 produces factual errors on roughly 10% of queries. Switching to Thinking mode reduces errors further by 50 to 80% compared to standard&nbsp;responses.<\/p>\n<\/div><\/div>\n\n\n\n<div data-schema-only=\"false\" class=\"wp-block-aioseo-faq\"><h3 class=\"aioseo-faq-block-question\">What&#8217;s the difference between a ChatGPT hallucination and ChatGPT lying?<\/h3><div class=\"aioseo-faq-block-answer\">\n<p>A hallucination is an unintentional error. ChatGPT generates plausible-sounding content without a reliable source to back it up. Deliberate deception is different and rarer. OpenAI found that some AI reasoning models can strategically withhold or misrepresent information to reach a goal. GPT-5 reduced these deception rates to\u00a0<a href=\"https:\/\/openai.com\/index\/introducing-gpt-5\/\">2.1%<\/a>, down from 4.8% in earlier models. For most everyday use, what looks like lying is almost always&nbsp;hallucination.<\/p>\n<\/div><\/div>\n\n\n\n<div data-schema-only=\"false\" class=\"wp-block-aioseo-faq\"><h3 class=\"aioseo-faq-block-question\">How do I fact-check ChatGPT outputs?<\/h3><div class=\"aioseo-faq-block-answer\">\n<p>Cross-check specific claims against primary sources. For data and benchmarks, use Statista, Gartner, or Forrester. For technology claims, use TechCrunch, Wired, or the vendor&#8217;s own documentation. For anything ChatGPT cites with a link, click it. Fabricated citations are one of the most common hallucination types. A link returning a 404 means the claim is&nbsp;unverified.<br><\/p>\n<\/div><\/div>\n","protected":false},"excerpt":{"rendered":"<p>ChatGPT often gives false information that defeats the tool&#8217;s purpose. Here are some tips and prompts to train your GPT to only give accurate info.<\/p>\n","protected":false},"author":24,"featured_media":7709,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"content-type":"","_lmt_disableupdate":"","_lmt_disable":"","footnotes":""},"categories":[7],"tags":[],"class_list":["post-7643","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-product-management"],"aioseo_notices":[],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v27.3 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>How to get ChatGPT to stop lying: 7 actionable tips<\/title>\n<meta name=\"description\" content=\"ChatGPT often gives false information that defeats the tool&#039;s purpose. Here are some tips and prompts to train your GPT to only give accurate info.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/canny.io\/blog\/get-chatgpt-to-stop-lying\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"How to get ChatGPT to stop lying: 7 actionable tips\" \/>\n<meta property=\"og:description\" content=\"ChatGPT often gives false information that defeats the tool&#039;s purpose. Here are some tips and prompts to train your GPT to only give accurate info.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/canny.io\/blog\/get-chatgpt-to-stop-lying\/\" \/>\n<meta property=\"og:site_name\" content=\"Canny Blog\" \/>\n<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/cannyio\/\" \/>\n<meta property=\"article:published_time\" content=\"2024-09-13T14:30:00+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-04-02T23:27:53+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/chatgpt-lying.png\" \/>\n\t<meta property=\"og:image:width\" content=\"1520\" \/>\n\t<meta property=\"og:image:height\" content=\"800\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"Eric Hoppe\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@cannyhq\" \/>\n<meta name=\"twitter:site\" content=\"@cannyHQ\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Eric Hoppe\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"20 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/canny.io\\\/blog\\\/get-chatgpt-to-stop-lying\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/canny.io\\\/blog\\\/get-chatgpt-to-stop-lying\\\/\"},\"author\":{\"name\":\"Eric Hoppe\",\"@id\":\"https:\\\/\\\/canny.io\\\/blog\\\/#\\\/schema\\\/person\\\/8e6666f1191fc331e561c3b65408f2d2\"},\"headline\":\"How to get ChatGPT to stop lying: 7 actionable tips for product&nbsp;managers\",\"datePublished\":\"2024-09-13T14:30:00+00:00\",\"dateModified\":\"2026-04-02T23:27:53+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/canny.io\\\/blog\\\/get-chatgpt-to-stop-lying\\\/\"},\"wordCount\":4297,\"publisher\":{\"@id\":\"https:\\\/\\\/canny.io\\\/blog\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/canny.io\\\/blog\\\/get-chatgpt-to-stop-lying\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/canny.io\\\/blog\\\/wp-content\\\/uploads\\\/2024\\\/09\\\/chatgpt-lying.png\",\"articleSection\":[\"Product Management\"],\"inLanguage\":\"en-US\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/canny.io\\\/blog\\\/get-chatgpt-to-stop-lying\\\/\",\"url\":\"https:\\\/\\\/canny.io\\\/blog\\\/get-chatgpt-to-stop-lying\\\/\",\"name\":\"How to get ChatGPT to stop lying: 7 actionable tips\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/canny.io\\\/blog\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/canny.io\\\/blog\\\/get-chatgpt-to-stop-lying\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/canny.io\\\/blog\\\/get-chatgpt-to-stop-lying\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/canny.io\\\/blog\\\/wp-content\\\/uploads\\\/2024\\\/09\\\/chatgpt-lying.png\",\"datePublished\":\"2024-09-13T14:30:00+00:00\",\"dateModified\":\"2026-04-02T23:27:53+00:00\",\"description\":\"ChatGPT often gives false information that defeats the tool's purpose. Here are some tips and prompts to train your GPT to only give accurate info.\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/canny.io\\\/blog\\\/get-chatgpt-to-stop-lying\\\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/canny.io\\\/blog\\\/get-chatgpt-to-stop-lying\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/canny.io\\\/blog\\\/get-chatgpt-to-stop-lying\\\/#primaryimage\",\"url\":\"https:\\\/\\\/canny.io\\\/blog\\\/wp-content\\\/uploads\\\/2024\\\/09\\\/chatgpt-lying.png\",\"contentUrl\":\"https:\\\/\\\/canny.io\\\/blog\\\/wp-content\\\/uploads\\\/2024\\\/09\\\/chatgpt-lying.png\",\"width\":1520,\"height\":800,\"caption\":\"Cover image for a blog showing a bot telling a product manager unreliable information.\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/canny.io\\\/blog\\\/get-chatgpt-to-stop-lying\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/canny.io\\\/blog\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"How to get ChatGPT to stop lying: 7 actionable tips for product&nbsp;managers\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/canny.io\\\/blog\\\/#website\",\"url\":\"https:\\\/\\\/canny.io\\\/blog\\\/\",\"name\":\"Canny Blog\",\"description\":\"How to build a more informed product\",\"publisher\":{\"@id\":\"https:\\\/\\\/canny.io\\\/blog\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/canny.io\\\/blog\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/canny.io\\\/blog\\\/#organization\",\"name\":\"Canny Blog\",\"url\":\"https:\\\/\\\/canny.io\\\/blog\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/canny.io\\\/blog\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/canny.io\\\/blog\\\/wp-content\\\/uploads\\\/2024\\\/03\\\/canny-wordmark-1.png\",\"contentUrl\":\"https:\\\/\\\/canny.io\\\/blog\\\/wp-content\\\/uploads\\\/2024\\\/03\\\/canny-wordmark-1.png\",\"width\":733,\"height\":201,\"caption\":\"Canny Blog\"},\"image\":{\"@id\":\"https:\\\/\\\/canny.io\\\/blog\\\/#\\\/schema\\\/logo\\\/image\\\/\"},\"sameAs\":[\"https:\\\/\\\/www.facebook.com\\\/cannyio\\\/\",\"https:\\\/\\\/x.com\\\/cannyHQ\",\"https:\\\/\\\/www.instagram.com\\\/carryoncode\\\/\"]},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/canny.io\\\/blog\\\/#\\\/schema\\\/person\\\/8e6666f1191fc331e561c3b65408f2d2\",\"name\":\"Eric Hoppe\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/canny.io\\\/blog\\\/wp-content\\\/uploads\\\/2023\\\/05\\\/eric-150x150.jpeg\",\"url\":\"https:\\\/\\\/canny.io\\\/blog\\\/wp-content\\\/uploads\\\/2023\\\/05\\\/eric-150x150.jpeg\",\"contentUrl\":\"https:\\\/\\\/canny.io\\\/blog\\\/wp-content\\\/uploads\\\/2023\\\/05\\\/eric-150x150.jpeg\",\"caption\":\"Eric Hoppe\"},\"description\":\"Marketer and aspiring dog-sport competitor \ud83d\udc15 Eric\u2019s career features stints with innovative companies like Opera Software and Crowd Content. When he\u2019s not telling the world how great Canny is, Eric's finding ways to get his dogson to be a more competitive frisbee dog.\",\"sameAs\":[\"https:\\\/\\\/canny.io\",\"https:\\\/\\\/www.linkedin.com\\\/in\\\/ericmhoppe\\\/\",\"https:\\\/\\\/x.com\\\/cannyhq\"],\"url\":\"https:\\\/\\\/canny.io\\\/blog\\\/author\\\/eric\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"How to get ChatGPT to stop lying: 7 actionable tips","description":"ChatGPT often gives false information that defeats the tool's purpose. Here are some tips and prompts to train your GPT to only give accurate info.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/canny.io\/blog\/get-chatgpt-to-stop-lying\/","og_locale":"en_US","og_type":"article","og_title":"How to get ChatGPT to stop lying: 7 actionable tips","og_description":"ChatGPT often gives false information that defeats the tool's purpose. Here are some tips and prompts to train your GPT to only give accurate info.","og_url":"https:\/\/canny.io\/blog\/get-chatgpt-to-stop-lying\/","og_site_name":"Canny Blog","article_publisher":"https:\/\/www.facebook.com\/cannyio\/","article_published_time":"2024-09-13T14:30:00+00:00","article_modified_time":"2026-04-02T23:27:53+00:00","og_image":[{"width":1520,"height":800,"url":"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/chatgpt-lying.png","type":"image\/png"}],"author":"Eric Hoppe","twitter_card":"summary_large_image","twitter_creator":"@cannyhq","twitter_site":"@cannyHQ","twitter_misc":{"Written by":"Eric Hoppe","Est. reading time":"20 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/canny.io\/blog\/get-chatgpt-to-stop-lying\/#article","isPartOf":{"@id":"https:\/\/canny.io\/blog\/get-chatgpt-to-stop-lying\/"},"author":{"name":"Eric Hoppe","@id":"https:\/\/canny.io\/blog\/#\/schema\/person\/8e6666f1191fc331e561c3b65408f2d2"},"headline":"How to get ChatGPT to stop lying: 7 actionable tips for product&nbsp;managers","datePublished":"2024-09-13T14:30:00+00:00","dateModified":"2026-04-02T23:27:53+00:00","mainEntityOfPage":{"@id":"https:\/\/canny.io\/blog\/get-chatgpt-to-stop-lying\/"},"wordCount":4297,"publisher":{"@id":"https:\/\/canny.io\/blog\/#organization"},"image":{"@id":"https:\/\/canny.io\/blog\/get-chatgpt-to-stop-lying\/#primaryimage"},"thumbnailUrl":"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/chatgpt-lying.png","articleSection":["Product Management"],"inLanguage":"en-US"},{"@type":"WebPage","@id":"https:\/\/canny.io\/blog\/get-chatgpt-to-stop-lying\/","url":"https:\/\/canny.io\/blog\/get-chatgpt-to-stop-lying\/","name":"How to get ChatGPT to stop lying: 7 actionable tips","isPartOf":{"@id":"https:\/\/canny.io\/blog\/#website"},"primaryImageOfPage":{"@id":"https:\/\/canny.io\/blog\/get-chatgpt-to-stop-lying\/#primaryimage"},"image":{"@id":"https:\/\/canny.io\/blog\/get-chatgpt-to-stop-lying\/#primaryimage"},"thumbnailUrl":"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/chatgpt-lying.png","datePublished":"2024-09-13T14:30:00+00:00","dateModified":"2026-04-02T23:27:53+00:00","description":"ChatGPT often gives false information that defeats the tool's purpose. Here are some tips and prompts to train your GPT to only give accurate info.","breadcrumb":{"@id":"https:\/\/canny.io\/blog\/get-chatgpt-to-stop-lying\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/canny.io\/blog\/get-chatgpt-to-stop-lying\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/canny.io\/blog\/get-chatgpt-to-stop-lying\/#primaryimage","url":"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/chatgpt-lying.png","contentUrl":"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/09\/chatgpt-lying.png","width":1520,"height":800,"caption":"Cover image for a blog showing a bot telling a product manager unreliable information."},{"@type":"BreadcrumbList","@id":"https:\/\/canny.io\/blog\/get-chatgpt-to-stop-lying\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/canny.io\/blog\/"},{"@type":"ListItem","position":2,"name":"How to get ChatGPT to stop lying: 7 actionable tips for product&nbsp;managers"}]},{"@type":"WebSite","@id":"https:\/\/canny.io\/blog\/#website","url":"https:\/\/canny.io\/blog\/","name":"Canny Blog","description":"How to build a more informed product","publisher":{"@id":"https:\/\/canny.io\/blog\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/canny.io\/blog\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/canny.io\/blog\/#organization","name":"Canny Blog","url":"https:\/\/canny.io\/blog\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/canny.io\/blog\/#\/schema\/logo\/image\/","url":"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/03\/canny-wordmark-1.png","contentUrl":"https:\/\/canny.io\/blog\/wp-content\/uploads\/2024\/03\/canny-wordmark-1.png","width":733,"height":201,"caption":"Canny Blog"},"image":{"@id":"https:\/\/canny.io\/blog\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.facebook.com\/cannyio\/","https:\/\/x.com\/cannyHQ","https:\/\/www.instagram.com\/carryoncode\/"]},{"@type":"Person","@id":"https:\/\/canny.io\/blog\/#\/schema\/person\/8e6666f1191fc331e561c3b65408f2d2","name":"Eric Hoppe","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/canny.io\/blog\/wp-content\/uploads\/2023\/05\/eric-150x150.jpeg","url":"https:\/\/canny.io\/blog\/wp-content\/uploads\/2023\/05\/eric-150x150.jpeg","contentUrl":"https:\/\/canny.io\/blog\/wp-content\/uploads\/2023\/05\/eric-150x150.jpeg","caption":"Eric Hoppe"},"description":"Marketer and aspiring dog-sport competitor \ud83d\udc15 Eric\u2019s career features stints with innovative companies like Opera Software and Crowd Content. When he\u2019s not telling the world how great Canny is, Eric's finding ways to get his dogson to be a more competitive frisbee dog.","sameAs":["https:\/\/canny.io","https:\/\/www.linkedin.com\/in\/ericmhoppe\/","https:\/\/x.com\/cannyhq"],"url":"https:\/\/canny.io\/blog\/author\/eric\/"}]}},"modified_by":"Eric Hoppe","_links":{"self":[{"href":"https:\/\/canny.io\/blog\/wp-json\/wp\/v2\/posts\/7643","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/canny.io\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/canny.io\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/canny.io\/blog\/wp-json\/wp\/v2\/users\/24"}],"replies":[{"embeddable":true,"href":"https:\/\/canny.io\/blog\/wp-json\/wp\/v2\/comments?post=7643"}],"version-history":[{"count":0,"href":"https:\/\/canny.io\/blog\/wp-json\/wp\/v2\/posts\/7643\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/canny.io\/blog\/wp-json\/wp\/v2\/media\/7709"}],"wp:attachment":[{"href":"https:\/\/canny.io\/blog\/wp-json\/wp\/v2\/media?parent=7643"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/canny.io\/blog\/wp-json\/wp\/v2\/categories?post=7643"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/canny.io\/blog\/wp-json\/wp\/v2\/tags?post=7643"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}