AI provides inaccurate information about Cursor itself - destroys user trust

[Jordi_G_V]

Where does the bug appear (feature/product)?

Somewhere else…

Describe the Bug

Problem Summary

Cursor’s AI assistant provides incorrect information about Cursor’s own features, versions, and UI paths. Users must verify every statement, destroying trust and wasting time.

Specific Examples

1. Version Information (Dec 27, 2025)

• AI says: “Cursor v2.3 released Dec 22, 2025”
• Reality: User has v2.2.44 compiled Dec 24, 2025 (newer build)
• Impact: User must verify every version/changelog statement

2. Model Count

• AI says: “6 models available”
• Reality: 50+ models (Settings > Models)
• Impact: Wrong competitor comparisons

3. UI Paths

• AI says: “Go to Settings → Models”
• Reality: Path doesn’t exist as described
• Impact: 8 iterations, 60 min wasted

4. Changelog Information

• AI provides outdated changelog data
• User must check official changelog manually
• Impact: Every update question requires verification

Impact

• Trust level: 0% - must verify every Cursor-related statement
• User quote: “If I have to verify everything you say about your own product, I lose confidence”
• Time wasted: Hours verifying AI’s own product information
• Subscription risk: Considering cancellation due to lack of trust

Request

Update AI’s knowledge base with accurate, current Cursor documentation. Users shouldn’t document your product better than you do.

User Info

• Subscription: Pro ($20/mo)
• Version: 2.2.44 (Dec 24, 2025 build)
• OS: Windows 10
• Date: Dec 27, 2025

Steps to Reproduce

1. Ask Cursor AI about Cursor’s current version or features
2. AI provides information (e.g., “v2.3 released Dec 22”)
3. User checks actual Cursor version (e.g., v2.2.44 compiled Dec 24)
4. Information doesn’t match - user must verify manually
5. Repeat for any Cursor-related question (models, UI paths, features)

Example conversation:

• User: “What changed in the latest Cursor update?”
• AI: “Cursor v2.3 released Dec 22, 2025 includes…”
• Reality: User has v2.2.44 compiled Dec 24 (newer build)
• Result: User loses trust, must verify every statement

Operating System

MacOS

Current Cursor Version (Menu → About Cursor → Copy)

Version: 2.2.44 (user setup)
VSCode Version: 1.105.1
Commit: 20adc1003928b0f1b99305dbaf845656ff81f5d0

For AI issues: which model did you use?

Multiple models - issue occurs across all models when asking about Cursor itself

Does this stop you from using Cursor

No - Cursor works, but with this issue

[jokerfool]

Settings Models works fine, you’re probably not using the right version

2.3 is the latest but it not released to everyone as of yet. Coming Soon to everyone else later.

[Mirshod_W]

good suggestions

@Jordi_G_V

Users shouldn’t document your product better than you do.

[Naufaldi_Rafif]

we need to know that each AI dont have updated realtime information, since certainly AI have knowledge cutoff. You can check in HaoooWang/llm-knowledge-cutoff-dates: This repository contains a summary of knowledge cut-off dates for various large language models (LLMs), such as GPT, Claude, Gemini, Llama, and more.

so its not about AI provide accurate information, but AI cant provide information without certaintly tools, thats why MCP exits and web Search exist.

[deanrie]

Hey, thanks for the report.

This is actually expected behavior. All LLMs have a “knowledge cutoff” date, meaning they don’t have real-time information about Cursor versions, features, or UI changes. This limitation applies across all AI models (GPT, Claude, Gemini, etc.), not just within Cursor.

As others mentioned in the thread:

• Settings > Models does work correctly in current versions
• v2.3 is rolling out gradually. You may be on v2.2.x, which is normal
• For real-time Cursor info, you can use @Web to enable web search, or set up MCP tools for documentation access

For accurate Cursor information, it’s best to check:

• Official docs: Cursor Docs
• Changelog: Changelog · Cursor

This isn’t a bug we can fix since it’s inherent to how LLMs work, but I appreciate the feedback.