Why Spell Check Correction Stopped Working After Perplexity Desktop Update — Causes and Solutions

Event Overview

Recently, a long-time user of the Perplexity Windows desktop app reported a frustrating issue on Reddit: after accepting a pushed update for the "all-new Windows desktop app," the right-click spell correction feature that previously worked fine suddenly stopped functioning.

According to the user, they had been using the Perplexity desktop app on multiple Windows 11 machines for over a year. In text input fields, misspelled words would always show a red wavy underline, and right-clicking would provide spelling correction suggestions — a standard feature in many desktop applications. However, this week's update migrated their previously manually installed standalone version to a Microsoft Store-hosted version, and the spell correction feature disappeared along with it.

The Peculiar Nature of the Bug: Red Underlines Appear but Correction Suggestions Are Missing

The most noteworthy aspect of this bug is its "half-broken" state. The user pointed out that red wavy underlines still appear normally beneath misspelled words, indicating that the underlying spell-check engine is still working and correctly identifying errors. However, when the user right-clicks these flagged words, the context menu no longer provides any spelling correction options.

In other words, the issue isn't with spell checking itself, but rather a breakdown in the "correction suggestions" interaction layer. This split-failure state is closely related to how the underlying spell-check engine works. Taking Chromium's built-in spell-check system as an example, it uses the Hunspell open-source spell-checking library as its backend engine, and its workflow is divided into two independent stages: the first stage is "marking," which compares user input against a dictionary in real-time and adds red wavy underlines to unrecognized words; the second stage is "suggestion," which invokes Hunspell's suggestion algorithm to calculate possible correct spellings and populate them into menu items when the user triggers the context menu. Since these two stages are relatively independent in their code paths, when issues arise in the menu rendering layer, it's entirely possible to see "marking works but suggestions are missing" as a split failure.

The user also checked the app's settings interface and confirmed there were no toggles related to spell correction, ruling out the possibility of accidentally disabling the feature.

The problematic version was clearly identified as Perplexity for Windows Version 26.8.0 (build 24557), the new version number after migrating to MS Store hosting, showing a massive version number gap compared to the previous standalone version 1.7.0.

Technical Differences Between the Standalone Version and MS Store Version

Why would the same Perplexity desktop app lose spell correction functionality after migrating from standalone installation to the MS Store version? This involves technical differences in app packaging and runtime environments.

Impact of MSIX Packaging Sandbox Mechanism

Apps distributed through the Microsoft Store typically use the MSIX packaging format and run in a sandboxed environment to some degree. MSIX is Microsoft's next-generation app packaging format introduced in 2018, integrating the advantages of previous installation technologies like MSI, AppX, and ClickOnce. The core design philosophy of MSIX is to achieve "clean install, clean uninstall" through containerized runtime environments — all of an app's files and registry modifications are confined to a virtualized file system and registry, leaving no residue upon uninstallation. However, this sandbox mechanism also means that app access to system resources is constrained by declarative permissions (Capabilities), with undeclared capabilities being denied by default.

While this sandboxing brings better security and a cleaner install/uninstall experience, it may also restrict the app's access to certain system-level APIs or WebView2 component features.

For desktop apps built on Electron or WebView2, the right-click spell correction menu often depends on proper integration between the underlying rendering engine and the operating system's context menu. It's worth noting that Electron and WebView2 have fundamental architectural differences: Electron packages a complete Chromium browser and Node.js runtime into the app, giving developers full control over the rendering engine and allowing complete customization of right-click menu content; WebView2 is Microsoft's lightweight solution that reuses the Edge browser engine already installed on the user's system, requiring context menu customization through intercepting and modifying the CoreWebView2ContextMenuRequested event. If developers switched the underlying framework during the migration from standalone to MS Store version, or failed to properly handle menu events in the new framework, spelling suggestion items could be accidentally filtered out.

If the new version has defects in its packaging or menu construction logic, the "red underline shows but correction suggestions are missing" phenomenon can occur — meaning the spell-check data was computed but wasn't correctly passed to the right-click menu's rendering pipeline.

Architecture Refactoring Implied by the Version Number Jump

The massive version number jump from 1.7.0 to 26.8.0 likely indicates that Perplexity underwent a significant architectural refactoring of its desktop client or adopted a new version naming convention. In software engineering, Semantic Versioning typically follows the "major.minor.patch" format, where major version changes signify incompatible API changes. A non-sequential version number change from 1.x to 26.x typically occurs in industry under several circumstances: adopting date-based version naming (e.g., YY.M.patch, where 26 might represent 2026 or week 26), switching to a version numbering system unified with backend services, or a fundamental replacement of the underlying framework. The build number 24557 also suggests this might be an auto-incrementing number from a continuous integration system, indicating the product has undergone extensive build iterations.

In such major version iterations, unintentionally omitting existing features or introducing regression bugs is quite common in software development.

User's Solution: Rolling Back to the Standalone Installation Version

Facing this issue, the user offered a pragmatic solution: uninstall the new MS Store version and roll back to the previously stable standalone version 1.7.0.

They mentioned that two other Windows 11 machines still had version 1.7.0 installed, with spell correction working perfectly. This provides ample verification for the rollback approach — the issue is indeed a regression bug introduced by the new version, not an environment or configuration problem.

For users encountering similar situations, consider the following approaches:

• Be cautious about "migrate to MS Store version" update prompts: These updates are often not just version upgrades but a wholesale switch of distribution channels and runtime environments, potentially bringing compatibility changes. The permission model under MSIX sandbox environments is fundamentally different from traditional Win32 applications, and some features that rely on system-level calls may behave abnormally after migration.
• Keep old installation packages: If you depend on specific features, backing up working standalone installers before upgrading is a wise move.
• Provide timely feedback to the developer: Missing spell correction menus represent a typical feature regression, and reporting through official channels helps the development team fix it in subsequent versions.

Implications for AI Product Desktop Adoption

This seemingly minor bug actually reflects a common tension in the rapid iteration of AI products: the balance between the pace of feature expansion and the stability of fundamental user experience.

As an AI search and Q&A product, Perplexity's core competitive advantage lies in its powerful retrieval and generation capabilities. But when it enters users' daily workflows as a desktop application, those seemingly insignificant "basic features" — like spell correction, copy-paste, and keyboard shortcuts — form the very foundation of user experience. Once this foundation develops cracks, no matter how strong the core AI capabilities are, it erodes users' trust in the product's reliability.

For the many AI companies currently bringing their web products to the desktop, this case offers a warning: while pursuing feature innovation and distribution channel optimization, thorough regression testing of fundamental interaction features is essential. Regression Testing is a key component of software quality assurance that ensures new code changes don't break existing functionality. In modern CI/CD pipelines, automated regression tests can typically cover core business logic, but scenarios like "does the right-click menu correctly display spelling suggestions" — involving UI interaction and native system component integration — often fall under end-to-end (E2E) testing territory, which is costly and easily overlooked in testing matrices. Especially when distribution channels switch from standalone installation to app stores, subtle differences in runtime environments (such as permission models and file system virtualization) can produce issues that only manifest in production.

Users' tolerance for "a feature that worked for a year suddenly disappearing" is typically much lower than for "lacking some new feature." This psychological phenomenon is known as "loss aversion" in behavioral economics — the pain of losing something one already has is approximately twice the pleasure of gaining something of equivalent value. For product teams, this means that each destructive update's damage to user trust requires several times the positive experience to recover from.

Key Takeaways