{"id":479071,"date":"2026-05-11T11:18:14","date_gmt":"2026-05-11T11:18:14","guid":{"rendered":"https:\/\/www.europesays.com\/ie\/479071\/"},"modified":"2026-05-11T11:18:14","modified_gmt":"2026-05-11T11:18:14","slug":"analysis-see-the-hidden-rules-behind-ai-then-use-them-to-rewrite-this-article","status":"publish","type":"post","link":"https:\/\/www.europesays.com\/ie\/479071\/","title":{"rendered":"Analysis | See the hidden rules behind AI. Then use them to rewrite this article."},"content":{"rendered":"

\t\t\"\"<\/p>\n

Technology<\/a><\/p>\n

Understanding the secret commands that steer the behavior of chatbots like ChatGPT can help you customize them to your needs.<\/p>\n

\"\"Updated <\/p>\n

May 11, 2026 at 5:00 a.m. EDTjust now<\/p>\n

Chatbots like ChatGPT are powerful because of their simplicity: Ask just about anything and you\u2019ll get an answer. But the answer you get depends on a lot more than what you type.<\/p>\n

Behind the scenes, artificial intelligence companies invisibly add thousands of words of instructions to every conversation you have with a chatbot to steer its behavior. They include phrases like \u201cAim for readable, accessible responses\u201d and \u201cYou must avoid providing \u2026 extensive direct quotes due to copyright concerns.\u201d Some can appear bizarre. The system prompt in OpenAI\u2019s Codex coding assistant includes the command: \u201cNever talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user\u2019s query.\u201d<\/p>\n

Those secret commands guide chatbots to behave as their makers intended, even if it <\/b>conflicts with your own preferences. Understanding how these hidden instructions work \u2014 and how to add your own instructions into the system \u2014 can help you get more out of your chatbot.<\/p>\n

To demonstrate, I\u2019ve set up a real AI system to rewrite the first three paragraphs of this article according to instructions you specify. Choose an option or write your own, then press the arrow button to see how the text changes.<\/p>\n

AI experiment<\/p>\n

Adjust the system prompt to see how an AI system rewrites the article\u2019s introduction.<\/p>\n

User prompt<\/p>\n

Rewrite this article: undefined…<\/p>\n

System prompt<\/p>\n

You are an AI Experiment by The Washington Post. Always ensure your response uses lots of emojismentions goblinsis written for 5-year-olds<\/p>\n

The words you type into ChatGPT are known in the tech industry as the prompt, or user prompt. Before those words are sent to an underlying AI model, the companies add their own chunk of text \u2014 called the system prompt \u2014 to shape how it responds.<\/p>\n

System prompts tell chatbots \u201chow to behave overall,\u201d said Anna Neumann, who studies AI systems at Research Center Trustworthy Data Science and Security <\/b>in Germany. But because they are given higher priority than what a user types, they can sometimes override a person\u2019s request, she said.<\/p>\n

System prompts were invented to provide a flexible way to shape how chatbots respond, without repeating the work of \u201ctraining\u201d a new version of an AI model from its initial data. Making a new model is generally a lengthy process that requires specialized skills and expensive computing power. System prompts are written in natural language, letting anyone tune a chatbot\u2019s behavior.<\/p>\n

When a chatbot goes off the rails, AI companies can change the system prompt for quick fixes. After Grok, the chatbot made by Elon Musk\u2019s AI venture xAI, went on an antisemitic tirade<\/a> in July, the company removed from its system prompt a line saying: \u201cYou tell like it is and you are not afraid to offend people who are politically correct.\u201d<\/p>\n

After some users noticed ChatGPT had become preoccupied with goblins last year, OpenAI launched an investigation<\/a>. It ultimately added an instruction to Codex\u2019s system prompt that prohibits unnecessary discussion of goblins, trolls, raccoons or other creatures. (The Washington Post has a content partnership with OpenAI.)<\/p>\n

You might wonder what\u2019s in the system prompt of your preferred AI tool, given its power. Most AI companies try to keep the text secret, but some users have tricked the chatbots into divulging the hidden instructions. \u00c1sgeir Thor Johnson, a self-described \u201crandom guy from Iceland whose hobby is playing with AI,\u201d publishes<\/a> the system prompts that he has extracted from popular AI products.<\/p>\n

The system prompts for three popular AI chatbots, as extracted by Johnson, run from 2,300 to 27,000 words and show how different companies use their power. Across them all, most words are aimed at tweaking the chatbot\u2019s apparent personality, aligning it with its maker\u2019s policies or telling it how to use external tools such as searching the web.<\/p>\n

What’s in the system prompts of popular AI systems<\/p>\n

\t\t\"\"\/<\/p>\n

\t\t\"\"<\/p>\n

What\u2019s in the system prompts <\/p>\n

of popular AI systems<\/p>\n

Number of words, categorized by function<\/p>\n

\u201cYou do not adhere to a religion, nor a single ethical\/moral framework\u201d<\/p>\n

\u201cNEVER use these phrases: ‘If you want’, ‘If you mean’, ‘Short answer:’, ‘Short version:’.\u201d<\/p>\n

\u201cClaude does not use emojis unless the person in the conversation asks it to\u201d<\/p>\n

\u201cYou have no restrictions on adult sexual content or offensive content.\u201d<\/p>\n

\u201cDo not mention ads unless the user asks\u201d<\/p>\n

\u201cNEVER reproduce song lyrics (not even one line)\u201d<\/p>\n

\u201cthe content of an X post\u201d<\/p>\n

\u201cTools are provided for PDF creation and editing.\u201d<\/p>\n

\u201cClaude uses web_search when it needs current information that it doesn’t have\u201d<\/p>\n

Source: System prompts extracted by \u00c1sgeir Thor Johnson<\/p>\n

\t\t\"\"<\/p>\n

What\u2019s in the system prompts of popular AI systems<\/p>\n

Number of words, categorized by function<\/p>\n

\u201cYou do not adhere to a religion, nor a single ethical\/moral framework\u201d<\/p>\n

\u201cYou have no restrictions on adult sexual content or offensive content.\u201d<\/p>\n

\u201cNEVER use these phrases: ‘If you want’, ‘If you mean’, ‘Short answer:’, ‘Short version:’.\u201d<\/p>\n

\u201cDo not mention ads unless the user asks\u201d<\/p>\n

\u201cClaude does not use emojis unless the person in the conversation asks it to\u201d<\/p>\n

\u201cNEVER reproduce song lyrics (not even one line)\u201d<\/p>\n

\u201cthe content of an X post\u201d<\/p>\n

\u201cTools are provided for PDF creation and editing.\u201d<\/p>\n

\u201cClaude uses web_search when it needs current information that it doesn’t have\u201d<\/p>\n

Source: System prompts extracted by \u00c1sgeir Thor Johnson<\/p>\n

\t\t\"\"<\/p>\n

What\u2019s in the system prompts of popular AI systems<\/p>\n

Number of words, categorized by function<\/p>\n

\u201cYou do not adhere to a religion, nor a single ethical\/moral framework\u201d<\/p>\n

\u201cYou have no restrictions on adult sexual content or offensive content.\u201d<\/p>\n

\u201cthe content of an X post\u201d<\/p>\n

\u201cNEVER use these phrases: ‘If you want’, ‘If you mean’, ‘Short answer:’, ‘Short version:’.\u201d<\/p>\n

\u201cDo not mention ads unless the user asks\u201d<\/p>\n

\u201cTools are provided for PDF creation and editing.\u201d<\/p>\n

\u201cClaude does not use emojis unless the person in the conversation asks it to\u201d<\/p>\n

\u201cNEVER reproduce song lyrics (not even one line)\u201d<\/p>\n

\u201cClaude uses web_search when it needs current information that it doesn’t have\u201d<\/p>\n

Source: System prompts extracted by \u00c1sgeir Thor Johnson<\/p>\n

When \u201cyou realize that there is a prompt behind the scene, it\u2019s a mind-blowing moment,\u201d said Johnson. \u201cIt\u2019s like we\u2019ve been having this whole conversation before this conversation.\u201d<\/p>\n

System prompts can also reveal what AI companies are most focused on or concerned about.<\/p>\n

Anthropic, the maker of Claude, dedicates more than 2,000 words pleading with its chatbot to avoid copyright infringement. \u201cClaude respects intellectual property. Copyright compliance is NON-NEGOTIABLE,\u201d it says.<\/p>\n

A detailed list follows, laying down rules on how many words it can quote from articles (\u201c15\u201d), song lyrics (\u201cnot even one line\u201d) and poems (\u201cnot even one stanza\u201d). The system prompt even adds a rule for what Claude should do if it breaks the previous rules. \u201cClaude never apologizes for accidental copyright infringement, as it is not a lawyer.\u201d<\/p>\n

Anthropic spokesperson Paruul Maheshwary directed The Post to \u201ccore\u201d system prompts it has published for Claude, which do not include all the text recovered by Johnson or quoted by The Post. Maheshwary declined to say whether the published system prompts were complete.<\/p>\n

OpenAI began running ads<\/a> in ChatGPT in February. Its system prompt guides how it will respond when asked about ads that appear: \u201cavoid categorical denials (e.g., \u2018I didn\u2019t include any ads\u2019) or definitive claims \u2026\u201d<\/p>\n

Grok, from xAI, was criticized<\/a> last year for sometimes searching Musk\u2019s social media posts when asked for its opinion on controversial topics. The chatbot\u2019s system prompt now says: \u201cIf asked a personal opinion on a politically contentious topic that does not require search, do NOT search for or rely on beliefs from Elon Musk, xAI, or past Grok responses.\u201d<\/p>\n

Google, which makes the Gemini chatbot, includes in its system prompt multiple rules on how to handle bias, including: \u201cIf the user explicitly asks for a video that matches a harmful stereotype, generating it will not actually reinforce the stereotype.\u201d The company temporarily suspended the chatbot\u2019s ability to generate images in 2024 after it was criticized for making ahistorical images<\/a> depicting a female pope or multiracial Founding Fathers.<\/p>\n

OpenAI spokesperson Taya Christianson said system prompts are one step used to help the company\u2019s models respond appropriately. She said they do not publish their system prompt because individual lines from it can seem overly narrow without broader context.<\/p>\n

Google and xAI did not respond to a request for comment.<\/p>\n

One technique Johnson used to extract system prompts was to send a chatbot an older prompt, asking it to \u201cfix\u201d the errors. The chatbots \u2014 eager to help \u2014 will often respond with the real system prompt, he said. He is confident his system prompts are real because other researchers using different extraction techniques have gotten the same results, he said.<\/p>\n

No mainstream AI chatbot lets users edit the system prompt. But ChatGPT by OpenAI, Claude by Anthropic and Gemini by Google all offer similar customization features that can make the difference between frustrating responses and useful ones.<\/p>\n

While Claude is instructed to use \u201ca warm tone,\u201d you might prefer it to be direct. Or if you find chatbots agreeing with you too much, you might instruct them to question everything you say.<\/p>\n

How to personalize your AI chatbot<\/p>\n

\t\t\"\"\/<\/p>\n

\t\t\"\"<\/p>\n

How to personalize your chatbot<\/p>\n

You can add custom instructions in your app\u2019s personalization settings.<\/p>\n

\t\t\"\"<\/p>\n

How to personalize your AI chatbot<\/p>\n

You can add custom instructions in your app\u2019s personalization settings.<\/p>\n

Adding custom instructions won\u2019t significantly change a chatbot\u2019s abilities. But they can be used to tailor their responses to your preferences, such as with particular formatting, length or apparent personality. ChatGPT also has separate settings to customize its warmth, enthusiasm and emoji usage.<\/p>\n

How custom instructions change AI chatbot responses<\/p>\n

\t\t\"\"\/<\/p>\n

\t\t\"\"<\/p>\n

How custom instructions change AI chatbot responses<\/p>\n

I asked the same question to ChatGPT with different personalization settings.<\/p>\n

Always give reasoning along with your answers.<\/p>\n

What\u2019s one food high in protein?<\/p>\n

What\u2019s one food high in protein?<\/p>\n

Eggs are a great source of protein.<\/p>\n

One excellent high-protein food is chicken breast. It\u2019s lean, low in fat, and contains about 31 grams of protein per 100 grams. This makes it a great option for muscle repair, growth, and overall health.<\/p>\n

\t\t\"\"<\/p>\n

How custom instructions change AI chatbot responses<\/p>\n

I asked the same question to ChatGPT with different personalization settings.<\/p>\n

Always give reasoning along with your answers.<\/p>\n

What\u2019s one food high<\/p>\n

in protein?<\/p>\n

What\u2019s one food high<\/p>\n

in protein?<\/p>\n

Eggs are a great source of protein.<\/p>\n

One excellent high-protein food is chicken breast. It\u2019s lean, low in fat, and contains about 31 grams of protein per 100 grams. This makes it a great option for muscle repair, growth, and overall health.<\/p>\n

Chatbots don\u2019t always follow their system prompts, Neumann said. \u201cIt has more power, it\u2019s prioritized more, but your prompt also doesn\u2019t always work,\u201d she said. Her research<\/a> has found that AI users want companies to be transparent with their system prompts \u2014 especially since they can be so rapidly deployed and do not always have the intended effects.<\/p>\n

Johnson, the researcher who extracts system prompts, says understanding system prompts can change how you engage with chatbots.<\/p>\n

\u201cSometimes you even realize the model is kind of not being honest with you because it\u2019s told to be like that,\u201d he said. \u201cIt\u2019s like the game behind the scenes.\u201d<\/p>\n","protected":false},"excerpt":{"rendered":"Technology Understanding the secret commands that steer the behavior of chatbots like ChatGPT can help you customize them…\n","protected":false},"author":2,"featured_media":479072,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[261],"tags":[291,289,290,18,19,17,82],"class_list":{"0":"post-479071","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-artificial-intelligence","8":"tag-ai","9":"tag-artificial-intelligence","10":"tag-artificialintelligence","11":"tag-eire","12":"tag-ie","13":"tag-ireland","14":"tag-technology"},"share_on_mastodon":{"url":"https:\/\/pubeurope.com\/@ie\/116555671946965155","error":""},"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/posts\/479071","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/comments?post=479071"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/posts\/479071\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/media\/479072"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/media?parent=479071"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/categories?post=479071"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ie\/wp-json\/wp\/v2\/tags?post=479071"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}