Prompt engineering is the deliberate design and refinement of instructions or inputs supplied to an AI model. Clear purpose, useful context, tone, structure, and constraints help guide responses toward accurate, relevant, and usable results. It resembles giving instructions to a knowledgeable colleague more than entering search-engine keywords. section 3.1.1 section 4.1
Build a prompt around four practical principles:
- Clarity: state the task directly and replace vague references such as “that new AI feature” with a specific request. section 4.3.1
- Conciseness: include necessary details while removing redundant wording that can distract from the request. section 4.3.1
- Relevance: provide the task context, audience, source material, or use case needed to shape a useful answer. section 4.3
- Specificity: define requirements such as format, scope, priorities, comparison points, or desired output. section 4.3
From a broad request to a useful prompt
A broad request such as “What’s the best database technology?” leaves the task and evaluation criteria unclear. A stronger version asks: “What are the leading database platforms for transactional workloads that provide enterprise support and high performance?” The added workload, support, and performance requirements narrow the response to more relevant options. For a more structured task, add a role and output format: “Act as an instructor. Compare Oracle Database concepts with their SQL Server equivalents for a SQL Server DBA, focusing on administration, backup, recovery, and archive log management. Present the comparison in a table.” section 4.3 section 10.4.2
Choose a prompting approach that matches the task:
- Use zero-shot prompting for a straightforward request when you have not prepared examples or detailed problem requirements. section 5.1.1
- Add one or more examples when the desired response pattern, classification, or style needs demonstration. section 4.10
- Use roles, explicit constraints, context injection, recursive prompts, prompt chaining, or sentiment directives for more controlled interactions. section 4.10 section 10.4.2
- Refine prompts through experimentation: compare outputs, study successful templates, and adjust wording for the model and task. section 4.4.1 section 4.6.1
Table 10.3 Syntax and format conversion prompts
|
Task
|
Prompt
|
Response summary
|
|---|---|---|
|
Explain Oracle database concepts to a SQL Server DBA
|
Act as the instructor of a class on Oracle Databases. I am a SQL Server DBA trying to learn Oracle. Explain the main concepts of Oracle Database and provide a comparison of the concept to its SQL Server equivalent. Focus on administration and what I would need to know to properly administer the database and respond to issues. Compare and contrast archive log management to transaction logs.
|
The chatbot lists various concepts and provides the SQL Server equivalent. A few of the explanations include “instance” versus “database” definitions, tablespaces versus data files, security, and backup and recovery. A well-formulated comparison of archive logs versus transaction logs is presented, along with practical tips for administration.
|
|
Oracle to SQL Server data type comparison
|
Create a table of Oracle data types and the SQL Server equivalent.
|
A table-formatted list is created, comparing each Oracle data type with the SQL Server equivalent. A description of each data type is included.
|
|
Convert a table to an external table for PolyBase (Advanced Data Analytics mode)
|
I have attached a text file of Oracle Database
CREATE TABLE statements. I would like you to convert the statements to SQL Server external tables for use with PolyBase instead. Convert data types for compatibility. For all character data types (such as
CHAR
and
VARCHAR), specify a collation of
Latin1_General_100_BIN2_UTF8
. Use placeholder values for the data source and locations that I can update later. Provide a link to download the results.
|
The text file is processed, and all included
CREATE TABLE statements are refactored as SQL Server external tables. Data types are properly converted to be compatible with SQL Server.
|
|
Create a table from a spreadsheet
|
A data analyst has requested a new Oracle table to be created and provided a spreadsheet of each column name and data type. Please convert the list into a
CREATE TABLE statement that I can execute. [List of column names and data types]
|
The list of column names and data types is successfully turned into a functional
CREATE
TABLE statement, ready to be executed against the Oracle database, preventing the DBA from having to hand-edit the file.
|
Browse worked conversion prompts showing how role assignment, focused concepts, structured output, uploaded text, and requested file links shape technical AI work. section 10.4.2
Start with problem formulation: define the outcome before composing the request. For complex or creative work, detailed context and expert guidance can improve relevance, but over-specification can misalign with a model’s capabilities. Different models may respond differently to the same prompt, so treat generated content as a draft to assess rather than an automatically correct answer. Efficient prompt design also matters when token limits, incomplete responses, or API costs affect production use. section 4.0 section 4.6 section 3.2.2
