July 2025 Meetups

Coffee Hour: Preparing Content for AI, Coffee Hour: Prompt Engineering and Personas

July 2025

Early Observations and Use Cases

💬 Kalyn Howard (Westerra Credit Union)

Shared that GenSearch has been live internally since April.
Contributors now revise inaccurate articles spotted by AI completions.
AI struggles with interpreting complex tables in financial content.
Kalyn uses a monthly Q&A session to refresh contributor knowledge and best practices.
Emphasized time constraints as the greatest challenge for AI-related content upkeep.

💬 Rod West (Aria Systems)

Highlighted hallucination issues in API documentation (e.g., AI misnaming functions).
Resolved by rewriting descriptions to include exact API names.
Plans to use GenSearch as a quality check tool post-documentation updates.
Suggested testing prompt-writing skills in future hiring assessments.

💬 Marc Noble (Waters Corporation)

Observed GenAI prefers long-form narrative over traditional KCS bullets.
Reported difficulty measuring content usefulness without KCS metrics.
Discovered certain articles are disproportionately cited by GenAI, even when irrelevant.
Currently in the fact-finding stage to identify formatting standards for AI success.

Templates, Formatting, and Chunking

💬 Gray Shekkola (Thermo Fisher Scientific)

💬 Adam Allen (Netsmart)

Shared excitement about launching GenSearch sitewide.
Asked about chunking mechanics — how headers and summaries influence what GenAI selects.
Later clarified that reused sections from “No LLM Index” articles can still be indexed if embedded elsewhere.

Demonstration: Using the AI Editor

Daniel Dalessio showcased how the AI Editor Tool:

Converts complex tables into readable articles using structured prompts.
Summarizes articles for faster indexing and improved comprehension.
Helps content creators test and refine GenAI-ready formatting with minimal effort.

Final Tips & Wrap-Up

Kalyn Howard

Reiterated the value of "No LLM Index" tags for excluding non-relevant content (e.g., release notes).
Warned against vague hyperlinks like "click here" — AI won't understand or surface these properly.

Adam Allen

Backed up Kalyn’s point with a general accessibility article discouraging ambiguous hyperlinking.

Daniel Dalessio

Closed by emphasizing iterative testing, time investment, and the evolving nature of best practices.

Prompt Engineering in GenSearch

Cody Sackett (Expert) led the session, guiding participants through:
- What prompt engineering is and why it matters
- Components of a strong GenSearch prompt: role, context, instructions, input data (kernels), and output indicators
- The importance of clarity, precision, and negative prompting to improve AI output

Customer Insights and Use Cases

Chris Blad (ETC) shared that their team:
- Makes one change at a time and waits 1–2 weeks to evaluate outcomes
- Relies on tweaking kernel counts and threshold settings to improve answer accuracy from highly technical documents
- Noted that some answers only live in a single article, requiring precision in retrieval

Interactive Q&A and Feature Requests

Jessica Betterly (Sylogist) asked about a prompt library for newer users—Sharon encouraged the community to post examples in the online forum.
Kalyn Howard (Westerra Credit Union) asked about revision history for persona settings. This sparked a round of feedback from:
- Jessica Betterly, Frank Tagader (Sylogist), Adam Allen, and Kalyn Howard
  They requested:
  - Version control for prompts
  - Ability to comment on persona edits
  - A more seamless way to tie changes to completion reports
  - Easier export and user mapping of GenSearch queries