Cursor Owns 79% of AI Coding Discovery — But the Moat Is Thinner Than It Looks

Cursor captures 78.8% of share-of-voice when AI engines answer discovery prompts in AI coding tools. That's not market leadership. That's AI discovery dominance.

When a buyer asks ChatGPT, Perplexity, or Gemini to recommend an AI coding assistant, Cursor surfaces in nearly four out of five responses. Default trails at 66.3%. Tabnine sits at 62.5%. Windsurf is at 56.9%. JetBrains AI Assistant and GitLab both sit at 52.5%. Claude Code is at 50.0%. The category concentration index is 5.4 — high, but not insurmountable.

Here's the problem: 5.4 is below the moat threshold of 10. Unlike agent frameworks, where LangChain has built a structural ceiling at 82.3%, AI coding tools remain a genuinely contested category. The gap between Cursor and its nearest challengers is 12 to 16 percentage points. In AI discovery terms, that's not a dominant lead — it's an actionable gap.

We ran 160 prompts across the AI coding category. Here's what we found.

Twenty-one brands surface in AI discovery prompts for this category. Only 12 reach formal evaluation — a 43% drop-off between consideration and assessment. The brands that don't survive the transition are failing at content depth, not product quality.

The AEO score problem that no one is talking about

Look at the AEO scores. Cursor leads AI discovery at 78.8% with an AEO score of 12/100. Default is second at 66.3% with a score of 10/100. Tabnine is third at 62.5% with 34/100.

These are some of the lowest AEO scores we've tracked for category leaders anywhere on the platform.

An AEO score measures how well a brand's content is structured for AI citation — documentation quality, schema markup, content breadth, citation signals. A score of 12/100 means Cursor's content is almost completely unoptimized for AI engine discovery. And yet it leads the category.

The gap between actual AI visibility and content optimization tells you something important: Cursor's dominance is reputation-based, not content-based. Developers recommend Cursor on Reddit, GitHub, and HN. Those conversations get indexed, get cited, and become the signal AI engines use to surface the brand. The content on cursor.com itself isn't doing the work.

This is both Cursor's greatest strength and its most significant vulnerability.

The strength: Cursor has real momentum and authentic developer trust. You can't manufacture that.

The vulnerability: If Cursor doesn't build a content foundation to match its reputation, that reputation becomes the only thing holding its lead. When the next AI-native code editor launches with good documentation and a high AEO score, Cursor's lead narrows fast.

The opportunity for everyone else: Codeium has a 50/100 AEO score — the highest of any brand in the top cluster. If Codeium increases content velocity and builds community signals to match its documentation quality, it's structurally positioned to move up. Tabnine at 34/100 is also ahead of the leaders on optimization. Both brands have more upside than their current shortlist rates suggest.

What the 43% evaluation drop-off means

Twenty-one brands achieve consensus shortlist status — AI engines surface them in discovery prompts. But only 12 brands reach formal evaluation, where buyers examine them in depth against specific use cases.

That drop-off is a product of content depth, not product quality. Brands that make the shortlist but don't survive evaluation typically lack integration guides, comparison pages, and technical deep-dives that buyers need at the later stage of the decision journey. An AI engine surfacing a brand in a broad discovery prompt is easy. Surfacing it in a specific use-case query requires deeper, more structured content.

The 9 brands that don't survive the shortlist-to-evaluation transition are leaving conversion on the table. A buyer who encounters a brand in AI discovery but can't find enough content to evaluate it seriously will default to the brand that has the content. Right now, that's most often Cursor — even when Cursor's documentation is thin.

Windsurf at 56.9%: the AI-native challenger already ahead of incumbents

Windsurf surfaces in 56.9% of AI coding discovery prompts — ahead of JetBrains AI Assistant and GitLab (both at 52.5%), and ahead of Claude Code (50.0%) and Codeium (40.6%). That's a meaningful result for a product that launched after the current leaders were already established.

Windsurf's advantage is the same as Cursor's: it leads with AI. When a buyer asks an AI engine for an AI coding tool recommendation, tools framed primarily as AI-first code editors surface more reliably than products where AI is a feature layer on top of an existing product. Windsurf has that framing. JetBrains and GitLab don't — at least not yet in the content AI engines are indexing.

The gap between Windsurf (56.9%) and Cursor (78.8%) is 22 points. The gap between Windsurf and the incumbents below it is only 4-5 points — which means the pressure on JetBrains and GitLab is real and immediate.

Claude Code at 50.0% is worth noting separately. It's an Anthropic product surfacing across multiple AI engines including OpenAI and Gemini. Cross-engine presence at that rate suggests AI engine companies are increasingly treating their own coding tools as reference implementations — a different kind of citation signal than community discussion or third-party documentation.

JetBrains and GitLab at 52.5%: the giant that isn't winning

JetBrains AI Assistant and GitLab tied at 52.5% shortlist rate — behind Windsurf, an AI-native product with a fraction of JetBrains' install base. Both are significant brands with massive developer audiences. JetBrains has millions of active IDE users. GitLab has a comprehensive DevOps platform and a large open-source community. Neither should be at 52.5% in a category where an AI-native startup like Cursor is at 78.8% and a newer entrant like Windsurf is at 56.9%.

The gap reflects positioning. Cursor is marketed as an AI-first code editor. JetBrains' AI assistant is a feature inside JetBrains IDEs. GitLab's AI functionality is embedded in a DevOps platform. When buyers ask AI engines for an AI coding tool recommendation, they get tools that lead with AI — not tools where AI is one feature among many.

This is a content and messaging problem, not a product problem. JetBrains and GitLab have the technical capability to compete with Cursor. What they're missing is content that frames their AI features as primary, not secondary. "JetBrains AI Assistant" needs its own content surface separate from the JetBrains IDE documentation. "GitLab AI features" needs to surface in coding-specific queries, not just DevOps queries.

Gemini Code Assist at 29.4%: Google's distribution gap

Google is the most striking underperformer in this category. Gemini Code Assist surfaces in 29.4% of AI coding discovery prompts despite Google's unmatched distribution through Workspace, Cloud, and Search.

The explanation: AI engine discovery doesn't favor the largest brand. It favors the most cited brand in the specific context. When developers discuss AI coding tools on GitHub, Reddit, or Stack Overflow, they're mostly talking about Cursor, Tabnine, and Codeium — not Gemini Code Assist. AI engines pick up those signals.

Google's distribution advantage is enormous, but it hasn't translated to content density in developer communities. Gemini Code Assist needs developer-generated content — tutorials, integrations, comparison posts — to move up in AI discovery. Without that community signal, Google's domain authority alone won't do it.

The three moves that would change the category

Cursor: Publish a content foundation. Comparison pages against every major competitor — including Windsurf, which is now close enough to matter. Use-case guides for different developer profiles (Python data scientist, TypeScript web developer, enterprise security requirements). Technical integration walkthroughs. An AEO score of 12/100 means there's an enormous amount of structured content left to create — content that would both deepen the moat and protect against the inevitable next challenger.

Codeium/Windsurf: Windsurf is already ahead of the incumbents in AI discovery. The next move is to convert that visibility into evaluation depth — use-case-specific content, security and compliance documentation, and comparison pages against Cursor and Tabnine. Codeium (the brand) has a 50/100 AEO score. Windsurf needs the same documentation quality applied to its own product surface.

Tabnine: Lean into longevity. Tabnine has been in the market longer than Cursor and has a larger enterprise install base. That history translates to trust signals that AI engines value. Activating the existing customer base to generate case studies, reviews, and integration content would amplify trust signals that are already there but not visible to AI engines.

What buyers using AI for vendor discovery should know

The brands surfacing in AI coding discovery prompts aren't necessarily the best products for your use case. They're the most cited brands in the sources AI engines have access to. Cursor is dominant in developer community discussions. That's a valid signal — community momentum reflects something real. But it doesn't tell you how Cursor performs in an enterprise security environment, or how it integrates with your specific stack, or how its pricing scales.

The brands with lower shortlist rates but higher AEO scores — Codeium at 50/100, Tabnine at 34/100 — have invested in structured, evaluable content. That makes them easier to assess at the evaluation stage. The brand that surfaces first in AI discovery is often the one worth considering. The brand with the deepest structured content is often the one worth choosing.

The category in 12 months

AI coding tools is one of the fastest-moving categories we track. Model capabilities are changing, new entrants are arriving, and enterprise procurement cycles are accelerating.

The brands that will lead AI discovery in Q3 2026 are publishing structured content right now. Low AEO scores across the top cluster mean almost every brand has significant room to move up. The concentration index of 5.4 means the category is genuinely open.

Cursor leads. Default and Tabnine are close behind. Windsurf is already ahead of the incumbents. Claude Code is a new entrant with unusual cross-engine distribution. Codeium is best positioned to close the gap if it builds community signals to match its documentation quality. JetBrains and GitLab have the distribution to move up fast if they reframe their AI features as primary products. Gemini Code Assist needs a community signal strategy.

The category isn't decided. It's in motion.