Case Study 02 Content Design AI Interface

AI Assistant Interface

A content design critique

Conversational interfaces fail in predictable ways. They over-explain when they should be concise. They apologize when they should redirect. They pivot to what they can do without acknowledging what the user actually needed.

This case study examines three response patterns from the HPE Networking AI assistant — one structural, one functional, one a failure — and shows what changes when the writing is working for the user instead of around them.

Applied to the HPE Networking AI assistant — an enterprise tool for network engineers.

Pattern 01 of 03

The opening state — scope is invisible

The tool greets every user the same way regardless of what they're there to do.

Screenshot 1 — Opening state

Annotated screenshot to be placed here.
Export from Keynote or Preview with callout labels applied.

Annotations

  • Tells the user nothing about what the tool does or covers
  • Four prompts, all SD Cloud and Mist — a Junos engineer sees nothing relevant and may assume the tool doesn't cover their work

The suggested questions are all drawn from one product family. An engineer working on Junos, MX, or PTX sees no signal that the tool covers their domain and has no obvious entry point.

A conversational interface that doesn't communicate its scope is one that users abandon before they get an answer.

Pattern 02 of 03

A successful response — with friction that could be removed

The response answers the question and is largely correct. The formatting choices work against the user.

Screenshot 2 — Successful response

Annotated screenshot to be placed here.
Export from Keynote or Preview with callout labels applied.

Annotations

  • Bold on the description, not the command — the reverse of what helps scanning
  • Decodes column headers the user just saw in the table above
  • Ten sources for one question

Bold is applied to the description of a command, not to the command itself. Engineers scanning for what to type read right past it. Code formatting puts the command first, in the format that matches how it appears in the CLI.

The bullet list below the table decodes column headers the user just read in the table. It's redundant, and it pushes the content the user needs further down the page.

Ten inline citations fragment a sentence that doesn't need attribution. Two curated source links at the bottom would serve the user better without the visual noise.

Pattern 03 of 03

A failure case — refusing a reasonable question

User question: How do Aruba and Juniper access points compare?

Screenshot 3 — Failure case

Annotated screenshot to be placed here.
Export from Keynote or Preview with callout labels applied.

Annotations

  • Throat-clearing that signals a dodge before the refusal arrives
  • A policy explanation nobody asked for
  • Asks the user to give up on what they wanted and accept something else

Before

I appreciate your question, but I'm unable to provide comparisons between Aruba and Juniper access points. My role is to focus on Juniper Networks solutions rather than comparative analysis with other vendors.

However, I'd be happy to help you with information about Juniper's wireless access point solutions. I can explain:

  • Juniper Mist wireless access points and their AI-native capabilities
  • How Juniper APs integrate with Mist for cloud-based management
  • Juniper's wireless security and performance features
  • Integration options with Juniper switching and routing platforms

Would you like details on any of these Juniper wireless topics?

After

This tool covers Juniper's product documentation and doesn't include cross-portfolio comparisons. For a side-by-side look at Juniper and Aruba access points, the HPE Networking product pages are the better starting point.

If you're evaluating Juniper APs specifically — capabilities, platform requirements, or Mist integration — I can help with that.

"I appreciate your question, but" is a softening phrase that signals evasion before the refusal even arrives. Engineers read through it immediately. The rewrite states the limitation directly without apologizing for it.

The original frames the refusal as a role constraint — which is beside the point. The user doesn't need a policy explanation. They need to know where to go instead. The rewrite redirects to a better resource and reopens the conversation on narrower, answerable ground.

The original pivot to a four-bullet promotional list ignores what the user was actually trying to do. Asking the user to accept a different question than the one they asked is not a redirect — it's a dead end.

The Through-Line

The fixes are mostly subtractive. Remove the throat-clearing. Cut the redundant explanation. Surface one good source instead of ten. What remains is a tool that respects the user's time and treats them as capable of handling a direct answer.