3.1.2 AI & Automation — Content Assistance — Writing and Completion
Content Assistance provides AI-supported tools for drafting, refining, and completing text across the platform. These tools are embedded directly into editors and input fields, enabling users to accelerate writing tasks while maintaining full editorial control and accountability.
Assisted Drafting
Assisted drafting helps users generate initial content based on explicit instructions and selected context. Drafts are created as editable text and never overwrite existing content.
Supported drafting actions:
Generate first drafts from prompts
Expand short notes into structured text
Rewrite content for clarity or tone
Example invocation:
$response = PraismaGPT::prompt([ 'instruction' => 'draft_intro', 'context' => $fields ]);Writing Completion
Completion features assist users while typing by suggesting continuations that align with the current sentence, paragraph, or section. Suggestions are contextual and limited to the active field.
Completion characteristics:
Inline, optional suggestions
No background insertion
User-accepted only
Completion does not alter surrounding content unless explicitly applied.
Refinement and Editing
Content Assistance supports refinement operations that improve readability and consistency without changing meaning.
Common refinement actions:
Simplify language
Improve structure and flow
Adjust tone or formality
Remove redundancy
Refinements are applied to selected text only.
Consistency and Style
AI assistance can help enforce stylistic consistency across content by aligning wording, terminology, and structure with existing material.
Style assistance may consider:
Previously approved content
Module-specific conventions
User-selected guidelines
No global style rules are enforced implicitly.
Context Boundaries
AI input context is explicitly defined and bounded. Only user-selected text and metadata are included in requests.
Context safeguards:
No implicit document-wide access
No cross-entry context sharing
No inclusion of unpublished data unless selected
Error Handling and Fallback
If AI assistance is unavailable, editors continue to function normally. Content Assistance features fail gracefully without blocking input or saving operations.
Usage Tracking and Limits
Content Assistance usage is tracked per tenant and user. Limits are enforced transparently to prevent unexpected consumption.
Tracked indicators:
Requests per period
Tokens consumed
Feature-level usage
Security and Privacy
All AI-assisted writing operations are tenant-scoped. Generated content is treated as user data and stored only when explicitly saved. Requests and responses are logged for audit without retaining model internals or training data.