AI Tools Hub
110 prompts · no signup

110 AI prompts that actually work

Prompts for the work people actually do: reviewing a diff, cutting a draft, interrogating a source, choosing a chart, threat-modelling a feature. Every one specifies an output format, because prompts that do not produce different results each time.

Square brackets mark the parts you fill in. Nothing here requires an account.

Coding10

Implementation, refactoring, review, test generation and debugging — prompts for each stage of development.

ChatGPT, Claude, Cursor, GitHub Copilot

UI and UX design10

Screen structure, empty states, error handling and onboarding — prompts for working through interface decisions.

ChatGPT, Claude, v0, Figma AI

Writing and copy10

Landing page copy, email, article structure, rewriting and tone — prompts for work that ships as text.

ChatGPT, Claude, Gemini

Research10

Competitor analysis, market framing, extracting the real question and checking primary sources.

ChatGPT, Claude, Perplexity

Marketing10

Personas, ad copy, social posts, landing page diagnosis and metric design — prompts for demand work.

ChatGPT, Claude, Gemini

Working with AI10

Role setting, stepwise instruction, fixing the output format and self-checking — prompts that make other prompts work.

ChatGPT, Claude, Gemini

Summarising and structuring10

Meeting notes, long documents, papers and multi-source material — prompts for compressing without losing the conditions.

ChatGPT, Claude, NotebookLM

Image generation10

Composition, lighting, material and subject direction — prompts written for image models rather than chat models.

Midjourney, DALL-E, Stable Diffusion, Ideogram

Branding and strategy10

Articulating a brand, positioning and testing business assumptions — prompts for decisions that are hard to reverse.

ChatGPT, Claude

Data analysis10

Designing the aggregation, generating SQL, choosing the chart and interpreting the result.

ChatGPT, Claude, Gemini

Security10

Vulnerability review, threat modelling, configuration audit and incident response.

ChatGPT, Claude

How these are written

  • Every prompt fixes the output format. Without one, the same prompt gives a different shape each time.
  • Several ask the model to refuse, to ask first, or to say what it could not determine. That is usually where the value is.
  • They are written for English use rather than translated, so the phrasing is the phrasing that works.

The Japanese library is separate and larger →