Use spacebar
/ arrows to advance; Esc for overview; s for notes.
A survey of the AI landscape in user assistance content in 2025-ish
Presented by
Lief Erickson
lief.erickson@intuitivestack.io
www.intuitivestack.io
2 Feb. 2026
Can a submarine swim?
Can a submarine swim?
Worse
When did it last snow in Minneapolis?
Sometimes it snows in April
Egg on face
Alt text for images
Agent
Microsoft Style Guide
WCAG guidelines
Company-specific rules
Benefits
Improved accessibility for all users
Consistent, literal descriptions across images
Faster publishing without adding review burden
Scales with content growth and reuse
Photo of a dog wearing sunglasses
Why AI Projects Fail
Goal is too broad.
“Let’s build an AI that answers all customer questions”
Assuming AI can fix bad content and bad data
“The AI will clean it up as it learns”
Full-text search + embeddings = “semantic”
“We don’t need taxonomies or ontologies. The LLM will find the connections automatically”
Vector databases = understanding
“Embedding the content captures its meaning, so we don’t need metadata or structure.”
Ignoring governance (until something goes wrong)
“We don’t want to slow things down with policies or review boards.”
Trying to corral chaos
“You're making me deal with questions I'd rather avoid.”
Better
How Alkemi Technology uses GPT Agents
AI doesn’t replace the work of a technical writer. It extends it. The real value comes
from how we guide it
and how rigorously we validate outcomes at every step.
-Tara English-Sweeney
User Story Architect GPT: Reverse-engineered product documentation into user stories we didn’t have.
Document Breakdown Assistant: Drafted overview topics in a specified format.
Short Description Creator: Created concise overviews.
Style Guide Assistant: Checked tone and consistency.
Transcript > SME review > refine into article
And many more automate-the-boring-stuff agents
How Mintlify Uses Claude
Looked at personal workflow and decided what to offload to Claude.
LMMs help non-writers to write better.
Motto to team: You are smarter than Claude.
Detailed prompt instructions in Claude.md.
Breaking any rule is failure.
Document new features: Claude can examine a Git diff and describe what is being changed.
Claude summarizes changes >> Claude Verify >> Propose changes to docs. Tip: Keep context window small.
Not wholesale changes, but focused updates.
Claude is good at the mechanics but not at seeing the strategy behind the content.
CT Smith at Payabli: 10x Efficient
SEO audit to find existing/missing frontmatter and metadata (3 hours, $15, Claude
Code to update)
For making voice notes outlines, proofreading, and brainstorming
Vibe code custom React components for internal processes. “A livesaver.”
Generate monthly reports for management
AI chatbots aren't a magic spell that compensates for crappy content.
-CT Smith
Best Available Human
Study of 5,172 support agents at a Fortune 500 firm
+14% productivity overall (issues resolved/hour)
Biggest gains for newer or lower-performing agents: +34%
Strongest impact in moderately rare situations where agents lack experience but data exists
If top performers stop contributing new approaches
