2026-06-07

chatbot usage examples

tools such as chatgpt expose a natural-language interface to a language model and an extensive knowledge corpus. they can replace web searches by extracting and synthesizing information that would otherwise require manual effort. openai identifies its main audiences as software developers, content creators, businesses and enterprises, and students and educators.

prompts

  • what is x always
  • create a short question whose answer is the preceding response
  • create a list of all questions that i or you explicitly or implictly asked in this chat
  • create an image that represents me as a person
  • given a doodle, colorize it
  • rewrite the following text for clarity
  • write in the prevalent writing style of the 1960s

use cases

learning and research

  • scientific calculation
  • converse to clarify unfamiliar concepts

language

  • multilingual translation
  • grammar correction
  • rewrite sentences for clarity
  • serve as a conversation partner in a different language
  • translate application localization files to additional languages
  • explore language nuances

writing

  • chatbots in 2026 still suck at writing. i do not recommend it. it will turn every text into standard language that will hardly express the original emotions, rhythm, or style and that can not be fixed. even worse, it will express different emotions, which makes it really unusable.

  • prime with questions about a topic, then generate a story based on this knowledge

  • detect potential faux pas

  • identify content worth responding to

  • detect content to drop to keep things shorter

  • create an outline of thematic directions (limited use)

  • collect text of your sent emails and have it adopt your writing style

interpersonal communication

  • dont rely on it for personal relationships. use it only to explore topics or nuance, but use your own ideas, words, rhythm, and the full emotional context as you have it
  • when analyzing text, do so from the other persons perspective. say "i am , ...". chatgpt is biased towards the user
  • take its words as opinionated hints, not authority

images

  • images of clothing from a shop, a photo of a person - show the person wearing the clothes
  • show the person 20 years older

coding and technical

  • create domain-specific calculators
  • write a mathematical function, then convert it to code
  • migrate code from an old api to a new api after being provided with two examples
  • translate source code between languages
  • improve variable names
  • add comments
  • generate command lines
  • convert mathematical equations to code
  • generate a list of commands for renaming multiple files

data and text handling

  • extract information
  • text reformatting
  • convert text in images to text

daily life and planning

  • given a detailed timeline of life, propose future goals
  • fitness plans
  • recipes
  • personal finances
  • scheduling
  • estimate food calories
  • upload product manuals and ask questions about the product

entertainment

  • interpret song lyrics
  • identify media from partial details
  • comedy and interactive stories
  • role playing
  • adopt the speaking style of a specific persona
  • recommend music, video channels, fashion, and gifts (note: bias)

more

  • summarize all assumptions made in this conversation
  • list all unresolved questions from this chat
  • identify implicit constraints or preferences in my prompts
  • convert this exchange into a faq format
  • list all named entities in the text
  • identify my cognitive biases based on this conversation
  • guess what i might be trying to accomplish, even if not explicitly stated
  • invert this idea and explore the opposite

projects

projects with instructions and project files are useful because they allow iterative development towards a goal by regularly updating project files and instructions with the current development state.

project file examples

  • collect documentation of a software library or a product you are researching as a single markdown text file

llm agent automation

especially when programming, applying suggested changes can be inefficient. exchanging updated results or error messages is also not automated.

some ideas and potential solutions:

  • applying patches

    • ast-based search and replace; replace entire functions
    • fuzzy string matching (problem: llm diffs are unstable and imprecise)

  • filesystem modifications

    • generate necessary shell commands

  • state analysis

    • connect to a debugger that automatically sends state data
    • use a controllable sandbox via text instructions