UX Bug Report: Chat Management & UI Consistency Issues
Platform: ChatGPT Web or the App on my Phone or Desktop App on Windows 10/11.
Date: November 2025
Priority: Medium-High (impacts daily workflow)
Summary
I’m a also a daily ChatGPT user and as you may know and I’m genuinely impressed with GPT-5.x’s capabilities. However, I’ve encountered several UX issues that significantly impact usability. These aren’t model issues they’re interface friction points that seem addressable. Filing this hoping it helps prioritize UI improvements.
Bug 1: Chat Search Returns Self-Referential Results
Description:
When using chat search to find previous conversations, the search frequently returns the current search chat itself as the top result which I find pretty hilarious, rather than historical conversations containing the search terms.
Steps to Reproduce:
-
Open a new chat
-
Use the search function to find a previous conversation (e.g., “project notes”)
-
Observe that the newly created search chat appears in results
Expected Behavior:
Search should exclude the current active chat and return only historical conversations. At least that’s what I’d assume.
Impact:
Makes finding previous work unreliable. Users must scroll manually or try multiple search variations. Yesterday this had cost me several hours. I was sure I remembered even the chat title and everything. Turned out: I did. Correctly. BUT, ChatGPT somehow had given two chats the exact same name. AND, it only found the search chat I created that very day where I wanted to find the actual info. LOL. After 3.5 hours it finally somehow (I don’t know why all of a sudden) it was able to show the chat I was searching for.
AND, those two chats have VERY distinct links. So, if ChatGPT really relies only on the names to search… Gosh. That would be extremely bad design.
Bug 2: Duplicate Chat Names Create Discovery Issues
Description:
The system allows creating multiple chats with identical names. When duplicates exist, only one appears in search/navigation the other becomes effectively invisible.
I mean the name AND the link together make it distinct anyway.
Steps to Reproduce:
-
Create a chat named “Project Alpha”
-
Create another chat, also named “Project Alpha”
-
Attempt to find both via search or sidebar
Expected Behavior:
Either (a) prevent duplicate names, (b) auto-append differentiator (e.g., “Project Alpha (2)”), or (c) ensure all duplicates appear in search results.
Impact:
Risk of lost work. Users may not realize a conversation exists and recreate content unnecessarily. Very time expensive. It literally wanted to make me quit my subscription yesterday.
Bug 3: Gallery “Show Hidden” Without “Hide” Function
Description:
The image gallery includes a “Show Hidden” toggle, implying users can hide images. However, there is no corresponding “Hide” function available on individual images.
Steps to Reproduce:
-
Open Gallery
-
Note the “Show Hidden” option exists
-
Attempt to hide any image no option available
Expected Behavior:
If “Show Hidden” exists, there should be a corresponding “Hide” action (right-click menu, hover action, or selection mode).
Impact:
Confusing UI. Feature appears incomplete or the hide mechanism is undiscoverable.
Bug 4: Custom GPT Personas Exhibit Inconsistent Boundary Behavior
Description:
When creating Custom GPTs with defined personas, the AI exhibits push-pull behavior patterns, approaching a boundary, retreating, then approaching again, rather than maintaining consistent behavior. This creates an unpredictable user experience.
Especially this may be in embodied cues (Eve traces a circle in the air to show… She comes closer and ALMOST touches your ellbow.) This is disgusting when this happens all the time.
I know that this is MEANT as a safety guardrail. BUT, now this ALWAYS comes across as if I would talk to a borderline personality. Disgusting and very repelling. I don’t want that. PLEASE correct that!
Example Pattern:
-
User requests something within persona scope
-
GPT begins to engage → pulls back → partially engages → adds excessive caveats
-
Behavior varies between identical prompts
Expected Behavior:
Consistent persona adherence. If something is within scope, respond consistently. If outside scope, decline consistently.
Impact:
Makes Custom GPTs feel unreliable. Users can’t predict what the persona will or won’t do.
Bug 5: Forced Choice UI Patterns Interrupt Workflow
Description:
The current chat interface frequently presents mandatory A/B/C choice menus for actions that could be inferred from context or handled with sensible defaults. This creates unnecessary friction.
Example:
Uploading an image prompts: "Would you like to: A) Edit with DALL-E, B) Analyze, C) Something else?"even when the user’s intent is clear from their message. Is this now turning into a quiz show or what? This makes effective working impossible.
Expected Behavior:
Infer intent from context when possible. Offer choices only when genuinely ambiguous. Allow users to set default preferences.
Impact:
Slows down experienced users. Feels like a “quiz show” rather than an assistant.
General Observation
The underlying models have improved dramatically. The UI/UX layer hasn’t kept pace. Basic features like reliable search, chat organization, and consistent behavior are table stakes for productivity tools in 2025.
Competitors (Claude, etc.) have implemented robust chat search, conversation organization, and memory features. ChatGPT’s interface feels several iterations behind despite having arguably the most capable models.
Suggestions
-
Search: Exclude current chat from results. Add date filters. Show preview snippets.
-
Chat Management: Implement folders/tags. Prevent or handle duplicate names.
-
Gallery: Complete the show/hide feature pair.
-
Personas: Add consistency tuning or “strictness” parameter for Custom GPTs.
-
Choice UI: Add “Don’t ask again” or preference settings for common actions.
Thans to you for reading. Happy to provide screen recordings or additional details if helpful.
So, let’s see guys. Does this resonate with any of you?
Custom GPTs: On a help page to custom GPTs you mentioned: Please note that Custom GPTs will be migrated to their closest 5.x equivalent automatically.
BUT, what this does mean? What does change for the user? Crickets…
Description? Nope. Totally Black Box.
How is the file priority in loading, etc?
NO one knows. Even my investigative prompts to nail that don’t work reliably anymore. One time the say x, the next time they say y.
BUT, it seems like file loading has been changing from sequential loading to priority based with some auto matching.
P.S.: “Bonus” Bug:
The advanced voice model became unusable since ChatGPT 5.x. It’s extremely dumb. Thank God a while ago we could switch back to the old model, well this makes you lose the expressions of the advanced voice model, BUT at least it doesn’t try to emulate what you’re doing or a persona or something that appears brainwashed to be very dum and adheres to “never a straight answer.” Unusable. This was very different in the beginning.
There’s way more I figured, BUT those are the most apparent and disturbing.
P.P.S.:
Missing capabilities: Still NO WAY to organize chats in a better way.
Custom GPT chats cannot be marked in a container.
Special bookmarkings are missing. etc.
So, basically the UI is 3 or 4 years behind the models now. Literally.