Cursor Dropped 11.6 Points in September: What the AIs Are Saying

Cursor's recommendation rate dropped from 40.9% to 29.3% in September 2026, a decline of 11.6 percentage points. That makes it the biggest single-tool movement in claw.mobile's AI recommendations, which tracks what ChatGPT, Claude, Perplexity, and Grok recommend across 40 practical tool-choice questions. Three other tools also fell sharply: Continue dropped 13.4 points, Glide fell 12.7 points, and Adalo declined 10.2 points. The pattern is clear, AI models are getting pickier about context. They're separating "build a real software project" from "generate a simple app without coding" and adjusting recommendations accordingly. Cursor remains the most frequently recommended option for specific builds, but its lead narrowed because the models are asking better questions about who's building and what they actually need.

The September answers show less willingness to treat Cursor as the default answer for every "build me an app" question. Cursor is strongest when the user already has a repository, understands software concepts, and needs an AI pair programmer that can inspect, edit, and iterate across a codebase. It's a weaker recommendation when the actual job is turning a prompt into a hosted prototype quickly, building a CRUD app without managing a local development environment, or allowing a non-developer to make changes safely. That distinction matters because "AI app builder" now describes several different products: prompt-to-app builders like Lovable, Bolt, and Replit; frontend and UI generation tools like v0; AI-first coding editors like Cursor and Windsurf; AI coding assistance inside established workflows like GitHub Copilot; agentic terminal-based development like Claude Code; and visual, data-oriented no-code building like Glide and Adalo. A model that asks what the user means by "build an app" will often produce a better recommendation than one that names a single winner. The drop doesn't mean Cursor became worse, it means the question became sharper.

Cursor is a strong choice for a developer who has a defined codebase, wants control over architecture, and is prepared to review generated code. It's also a natural fit for users who expect to move beyond a prototype and maintain the application over time. Cursor's current plan structure reinforces that developer-centric position. Its Free plan costs $0 but has limited usage; Plus is listed at $8 per month, while Premium, Super, and Ultra are listed at $30, $100, and $200 per month respectively. Annual pricing is listed as $80 for Plus, $300 for Premium, $1,000 for Super, and $2,000 for Ultra. The Free plan does not include third-party models; Plus also excludes third-party models, Auto, Cloud Agents, Automations, Bugbot, Origin, and personal API keys. For teams, Business seats are listed at $30, $100, or $200 per user per month for Premium, Super, or Ultra, while third-party model requests add a $0.25-per-million-token Cursor Token Rate. Cursor announced a June 1, 2026 Teams pricing update with Standard at $32 per seat per month annually or $40 monthly, and Premium at $96 annually or $120 monthly. You can compare current pricing across tools at claw.mobile's live pricing index. If you're choosing between AI coding tools, the key test is whether you need repository access, agent behavior, terminal use, model control, code review capabilities, and deployment flexibility, or whether you need something that ships a working prototype faster.

The recommendation becomes less convincing for a non-coder who mainly wants a functioning internal tool. In that case, Glide may be faster because it combines app screens, data, and integrations in one visual environment. Glide's current Free plan includes two team members, up to 10 projects, integrations, and limited credits for AI features. A previous Glide plan description listed one editor, up to 10 personal users, 25,000 rows, and zero monthly updates on the free tier, showing why plan details should be checked before committing. Adalo can also be a better fit when the priority is a visual mobile or web app without writing conventional code. But "no-code" does not remove product decisions, users still need to understand data models, permissions, navigation, responsive behavior, publishing, and integration limits. When Bolt is the wrong tool, the same logic applies: match the builder to the build, not the hype. Continue's sharper decline, from 17.9% to 4.5%, is even more revealing. It's attractive to developers who want an open-source or configurable coding assistant, but it requires more judgment about models, providers, setup, and workflow. That makes it powerful for a technically capable user and a poor default for someone asking for the fastest path from idea to deployed app. See Continue vs Cursor vs Windsurf for the full comparison.

Cursor's fall is best read as a positioning correction, not a verdict. It is a developer-oriented coding environment, while products such as Lovable, Bolt, v0, Replit, Windsurf, GitHub Copilot, and Claude Code compete across different parts of the build workflow. The September movement shows that recommendations are highly sensitive to the user's role and the type of build. That's why "best AI app builder" lists without user context are increasingly useless, what works for a developer with a Next.js repository is not what works for a founder who needs a working internal tool by Friday. The same model that recommended Cursor in August might now recommend Lovable for a non-coder, Replit for a team wanting integrated deployment, or v0 for a designer prototyping a landing page. That's not inconsistency; that's better matching. The question is whether the recommendation surface, ChatGPT, Claude, Perplexity, Grok, will continue to probe user context or revert to safe, generic answers. September suggests the former. If you're choosing between tools, don't rely on a single AI answer. Compare live pricing and feature limits, test the free tiers, and pick based on what you'll actually do after the first prototype ships.

Developers choosing a serious build should put Cursor, Windsurf, Claude Code, GitHub Copilot, and Replit into the comparison. Choose based on repository access, agent behavior, terminal use, model control, code review, deployment, and how much vendor lock-in the project can tolerate. Non-coders validating an idea should start with Lovable, Bolt, Glide, Adalo, or Replit rather than assuming an AI code editor is the fastest route. The key test is whether the user can make the second and third rounds of changes without needing a developer. Teams building internal tools should compare Glide and Replit with the AI coding tools. Data permissions, authentication, auditability, integrations, and maintenance matter more than the quality of the first generated screen. Experienced developers evaluating Cursor should not interpret the lower recommendation rate as a reason to abandon it. Treat it as a signal to use Cursor for the cases where an AI coding environment is genuinely advantageous: an existing repository, a specific feature, a known stack, and a human who can verify the result. Content and procurement teams should stop publishing universal "best AI app builder" lists. The September movement shows that recommendations are highly sensitive to the user's role and the type of build. If you're tracking pricing changes across these tools, claw.mobile's daily pricing changelog logs every update as it happens.

The four tools with the largest declines, Cursor, Continue, Glide, and Adalo, represent two distinct categories. Cursor and Continue are developer tools that require coding knowledge, code review, and software architecture judgment. Glide and Adalo are no-code builders designed for non-developers who prioritize speed and visual interfaces. Both categories fell because AI models are now asking follow-up questions before recommending: Do you have a codebase? Do you know how to review code? Do you need to maintain this long-term? Do you want to deploy without touching a terminal? The answers determine whether the recommendation is Cursor or Lovable, Continue or Bolt. This is a shift from brand-driven recommendations to context-driven matching. It's why tracking monthly AI recommendation data matters more than tracking one-time product reviews. The tools didn't change in September, the questions did. If you're building with AI tools, the lesson is simple: clarify the build type, user role, and maintenance plan before asking for recommendations. If you're choosing tools, test the free tiers against your actual workflow, not a generic "build an app" prompt. And if you're a non-technical founder who realizes this is more than you want to manage yourself, that's a reasonable read, most people shipping real software with AI are either developers who can review code or founders who hire someone who can.