Revvity - RevvityAI usability heuristics evaluation

Now let's get into some research & recommendations

RESEARCH & RECOMMENDATIONS

1. Auto Scrolling s. Non Auto Scrolling

Current Experience

RevvityAI has currently implemented auto-scrolling where the screen follows the latest text generated. Many of our users have complained that the auto-scrolling disturbs their reading.

Recommendation

In general: we want to give our users control on when to read and when to scroll. Auto-scroll behaviour should be adjustable or pause when users manually scroll up to review past messages (Nielson Normon Group)

Our recommendation is to remove auto-scrolling. To give our users more control, once the AI agent generates content that overflows the viewport, a button to “jump to bottom” should appear.Nielson Normon Gr

Once a new user message is sent the screen should auto scroll so that the new query appears at the top of the screen, shifting user focus to their current query.Nielson Normon Gr

Research

Check out these pros and cons for auto scrolling vs. non auto-scrolling:Nielson Normon Group

Auto scrolling to bottom

Keeps the latest content in view, matching user expectations in chat interfaces where new messages appear at the bottom

Supports natural conversational flow, especially when responses are streamed incrementally

Reduces user effort to manually scroll down, improving perceived responsiveness and engagement

Users generally don’t read as fast as the scrolling, so it might be disruptive

Interrupts user-initiated scroll positions (e.g., if user scrolls up to review history, auto-scroll should pause)

Keeping scroll position at the top

Allows users to read or interact with earlier messages without interruption (since users generally want to read from the top of the message)

Users may miss new content appearing below, causing confusion or requiring manual scrolling

Examples from other AI products

Products that DON’T auto-scroll: Claude, Gemini, Perplexity, ChatGPT

RESEARCH & RECOMMENDATIONS

2. Streaming vs. Non Streaming

Current Experience

RevvityAI has currently implemented streaming where the chat will pause while thinking, and then stream in the response. Bug: some users cannot see streaming animations (some browsers do not support it).

Recommendation

Our recommendation is to not only keep streaming but include also a “Thinking” state to keep users in the loop on the work going on in the background.Nielson Normon Group).

Research

Check out these pros and cons for streaming vs. non streaming chats:Nielson Normon Group

Streaming

LLMs can sometimes take awhile to work, streaming enables the user to understand exactly what is happening under the hood → helps users build trust

LMs can mess up often, streaming provides a great interface to naturally correct and guide it → you can know exactly where and what they messed up on, stop generating, and correct it

Faster perceived response time

It can be a bit awkward for longer-running tasks — am I just going to sit there and watch the agent work?

Non Streaming

Non streaming would only work better if trust is assumed → say users are simply delegating tasks to an agent and interacting in a more async nature rather than micro managing

More native to use – similar experience as just texting a human

Users can’t see what’s going on under the hood, leaving them in the dark

If a task takes longer, they may be wondering what the chat is doing in the time

Issue of “double texting” → what happens if user messages once, agent starts doing something, and then user messages again with a different (and sometimes unrelated) thought before the agent finishes it’s task? With streaming this isn’t an issue because users can see when the AI is still thinking

AI “thinking”

Dynamically delayed responses improve user experience with chatbots by making them feel more human-like. The paper shows that in conversational interfaces, slower can be better, as long as the delay mimics human behavior.

Faster is Not Always Better: Understanding the Effect of Dynamic Response Delays in Human-Chatbot Interaction

Examples from other AI products

AI Chats that use thinking, streaming:

ChatGPT

AI Chats that use progress indicator:

Perplexity
Claude
Gemini

RESEARCH & RECOMMENDATIONS

3. Differentiating User vs. System Messages

Current Experience

The current RevvityAI chat interface makes it difficult to tell at a glance which messages were sent by user and which were sent by AI.

Research

Check out what other AI products are doing!Nielson Normon Group

Option 1: Current prompt the user is reading is sticky at the top as user scrolls.

Perplexity

Option 2: Right aligned (expected experience of messaging apps)

ChatGPT

Gemini

Option 3: Left aligned with a text message bubble styling

Claude

Recommendation

Since our users often complain about the lack of vertical real estate, we recommend to not use the sticky message version like Perplexity. The latter is also a more unique / unexpected experience. Our recommendation is to copy the common text messaging software design where the user’s message is in a bubble and right aligned.Nielson Normon Group.

Designs

See mockup below.Nielson Normon Group.

RESEARCH & RECOMMENDATIONS

4. Indicating Continued Response

Current Experience

If users scroll up, there is no visual indication of continued response. User often miss overflow content.

Research

Many other existing AI platforms either have the text cut off or blurred on the last line so users know there is more text to scroll down to. There is also often a button that appears to jump to the bottom of the response as additional affordance.Nielson Normon Group

Claude

ChatGPT

Gemini

Recommendation

Mimic Gemini's experience: blurred last line. Add "jump to bottom" button as soon as users scroll up.Nielson Normon Group.

Designs

See mockup below.Nielson Normon Group.

RESEARCH & RECOMMENDATIONS

5. Stop Generating

Current Experience

Currently RevvityAI does not have any "stop generating" functionality. Once users send a message, they must wait until RevvityAI has finished responding before proceeding.

Research

UX Best Practices:

Give users an “out” – users should always have the agency to end a conversation or loop
Useful for aborting overly long threads or misdirected content

Virtually every major AI product has a stop generation feature, typically indicated by a "stop" icon in the chat box where the enter button is.

ChatGPT

Recommendation

We should provide users with the option to stop the chat and correct it at any point when the AI agent is generating its response – that way our users always have an “out”. Another recommendation would be to include confirmation that the user has stopped the response.Nielson Normon Group.

Designs

See mockup below.Nielson Normon Group.

RESEARCH & RECOMMENDATIONS

6. New Features

Current Experience

Currently new features are hidden in a change log page. Navigation to this page is hidden in the profile section, which is unintuitive and not discoverable

Research

What other products are doing:

ChatGPT

Gemini

Perplexity

Recommendation

The way most AI products call out new features lack discoverability. So, the recommendations we came up with were taken from a combination of our current design patterns, the Revvity design system, and feedback from the wider UX team.Nielson Normon Group.

The proposed solution here was to create a notifications system to house all recent / unread updates, and a button to link out to the change log.Nielson Normon Group.

Designs

See mockup below.Nielson Normon Group.