Glossary
Prompt engineering
Prompt engineering is the practice of writing and refining the instructions, context and examples given to an AI model so it produces accurate, consistent output for a task, and testing those prompts on real cases before relying on them.
What Is Prompt Engineering?
In short A strong prompt states the task, the audience, the inputs the model can use, the output format and what to do when information is missing.
The prompt guides from OpenAI, Anthropic and Google recommend the same core techniques: clear instructions, a few worked examples (few-shot prompting), structure that separates instructions from reference material, and splitting complex tasks into a chain of smaller prompts.
It matters because the same model can give a usable answer or a useless one depending on the prompt. In outbound, a vague prompt such as write a personalized email produces generic flattery. A specific one gives the prospect's role, one fact from their website, your offer, a 90-word limit and two good examples, and produces something a rep would send.
Prompt Engineering Example
For example, a team personalizes the first line of 2,000 emails. With version 1 of the prompt (write a compliment about the company), a reviewer rejects 30% of lines in a 50-record test. Version 2 requires one specific fact from the company's website, a 20-word limit and two good examples, and rejections on the same 50 records drop to 6%.
Prompt Engineering in Practice
Treat prompts like code. Define what a good output looks like first, keep a test set of 20 to 50 real records, and compare outputs before rolling a change out to a full list. Anthropic's guide makes the same point: start with success criteria and a way to test against them, and accept that some problems are better fixed by a different model than by a longer prompt.
FAQ
What are the main prompt engineering techniques?
Clear and specific instructions, few-shot examples, relevant context, asking the model to think before answering, structured formatting such as XML tags or Markdown sections, and prompt chaining, where one prompt's output feeds the next.
What is few-shot prompting?
Including a few examples of inputs and the outputs you want, so the model follows the pattern without fine-tuning. Google recommends always including few-shot examples, and warns that too many can make the model overfit to them.
Is prompt engineering the same as fine-tuning?
No. Prompt engineering changes the instructions and examples you send. Fine-tuning changes the model itself by training it further on your data. Try prompting first, since it is faster and cheaper to change.
When is prompt engineering the wrong fix?
When the problem is latency, cost or a capability the model lacks. Anthropic notes that not every failing test is best solved by prompt engineering, and that switching models can sometimes fix speed or cost more easily.
Do This in Ken
Do This in Ken
Try research-based personalization on one real prospect with the free AI personalization demo, then run your outreach as email and LinkedIn sequences in Campaigns.
- Module
- Campaigns
- Open in Ken
- /tools/ai-personalization
The new Ken opens to the waiting list in batches.
Related
- GTM Glossary
Every GTM term, grouped by topic.
- LLM (large language model)
- AI agent
- RAG (retrieval-augmented generation)
- Cold email
- A/B testing
- Ken
Sources
| Source | Checked |
|---|---|
| OpenAI API docs: Prompt engineering | |
| Claude docs: Prompt engineering overview | |
| Google AI for Developers: Prompt design strategies |
Changelog
- Page published.