Prompt chaining is one of the most underused techniques in AI prompting. Instead of asking an AI to do everything in one prompt, you break the task into a sequence of focused steps — each using the output of the previous one as input. The results are consistently better than single-prompt approaches for anything complex.
Why prompt chaining works
AI models perform worse when asked to do many things at once. A model asked to 'research, analyse, and write a 2,000-word article' is balancing three competing demands simultaneously. Splitting these into three prompts — research first, analysis second, writing third — lets the model focus fully on each step, producing better output at every stage.
The 4-step content creation chain
Step 1: Research prompt
Research the topic: [topic]. Identify: 1. The 5 most important subtopics to cover. 2. The most common questions people have about this topic. 3. The angle that is least covered but most interesting. 4. 3 specific statistics or data points that would add credibility. Format as a research brief.
Step 2: Outline prompt
Using this research brief [paste Step 1 output], create a detailed article outline for a 1,500-word piece targeting [keyword]. Include: H2 headings, the main point under each section, and which statistic or example goes where. Format as a structured outline.
Step 3: Draft prompt
Write the full article using this outline [paste Step 2 output]. Tone: [tone]. Target audience: [audience]. Do not add anything not in the outline. Match the word count target. Write the H1 as: [title].
Step 4: Edit prompt
Edit this draft [paste Step 3 output] to: 1. Cut any filler phrases or padding. 2. Ensure the first sentence of every paragraph earns the reader's attention. 3. Flag any factual claims that should be verified. 4. Shorten to [word count] without losing substance.
The 3-step analysis chain
- Step 1: 'List all the key data points in this document [paste doc]'
- Step 2: 'Using these data points [paste], identify the 3 most significant insights and explain why they matter'
- Step 3: 'Based on these insights [paste], write a 5-bullet executive summary for a non-technical audience'
Tips for effective prompt chaining
- Each step should have one clear job — don't let steps bleed into each other
- Explicitly paste the previous step's output — don't assume the model remembers
- Add a quality check step at the end: 'Review this output against these criteria: [list]'
- For research chains, put the web search step first and all synthesis steps after
- Save your best chains as templates — they're reusable across similar tasks