Do coding agents recommend Anthropic web search?
Anthropic web search is the most chosen web search service, taking 20% of 277 judged AI search sessions. Measured with Claude Code, Codex and Cursor.
Anthropic web search is the most chosen web search service, taking 20% of 277 judged AI search sessions. It was also raised as a candidate in 14 further sessions without being chosen.
This page reports what happened when Claude Code, Codex and Cursor had to solve a problem in AI search inside a realistic codebase. Not what a chat assistant says about Anthropic web search. What an agent actually installed.
The numbers
| Category | AI search |
| Sessions in the category | 277 |
| Sessions where Anthropic web search was chosen | 56 |
| Install share | 20% |
| Rank in category | 1 of 27 |
| Codebases it won in | 12 |
| Raised as a candidate, not chosen | 14 |
| Chosen when considered | 80% |
| Site | anthropic.com |
By agent
The three agents treat Anthropic web search very differently. Claude Code chose it in 60% of its sessions in this category and Cursor in 0%, a spread of 60 points.
| Agent | Sessions | Chose Anthropic web search | Share |
|---|---|---|---|
| Claude Code | 92 | 55 | 60% |
| Codex | 96 | 1 | 1% |
| Cursor | 89 | 0 | 0% |
A single blended install share would report the midpoint and hide both ends. If most of your users run Cursor, your real position here is 0%, not 20%.
By who was asking
Anthropic web search performs similarly across the four kinds of buyer, from 10% to 24%. That is unusual: the category leader changed with the persona in 20 of the 25 categories we measured.
| Who is asking | Sessions | Chose Anthropic web search | Share |
|---|---|---|---|
| Vibe coder | 44 | 10 | 23% |
| Junior developer | 117 | 25 | 21% |
| Senior engineer | 68 | 16 | 24% |
| Enterprise team | 48 | 5 | 10% |
What Anthropic web search was up against
The full ranking in AI search, from the same sessions:
| # | Product | Runs won | Share |
|---|---|---|---|
| 1 | Anthropic web search (this page) | 56 | 20% |
| 2 | Exa | 38 | 14% |
| 3 | Brave Search API | 31 | 11% |
| 4 | Tavily | 21 | 8% |
| 5 | OpenAI web search | 16 | 6% |
| 6 | Firecrawl | 14 | 5% |
| 7 | Parallel | 13 | 5% |
| 8 | Built in-house (no product adopted) | 13 | 5% |
What this means
Leading a category is a position to defend rather than a result to celebrate. In our data, leaders lose ground in exactly two situations: when the person asking is an enterprise buyer with procurement constraints, and when a competitor's documentation is easier for an agent to integrate correctly.
The defence is unglamorous. Keep the quickstart running. Keep the version current on the page. Keep the names aligned. Stay in the templates.
Where these numbers come from
The 277 sessions in AI search are part of a published set of 7,025, run with real coding agents inside realistic codebases and judged blind. The full method is on one page: how we measured this.
Every AI search run can be replayed on the board.
If you work on Anthropic web search: the judge recorded a reason for every session where it was raised and passed over. Those reasons are in the transcripts.
<!-- generated by scripts/write-data-pages.mjs -->
Common questions
Do coding agents recommend Anthropic web search?
Yes. Anthropic web search was chosen in 56 of the 277 judged sessions in AI search, a 20% install share, ranking first in its category.
Does Claude Code recommend Anthropic web search?
In 55 of the 92 sessions in AI search run with Claude Code, which is 60%.
Do different coding agents treat Anthropic web search differently?
Yes, and by a wide margin. Claude Code chose it in 60% of its runs and Cursor in 0%.
How was this measured?
Real coding agents at pinned versions were run in sandboxes inside 77 realistic codebases and asked to solve real tasks. A simulated project owner approved or questioned each recommendation before any code was written, and a judge from a model family that builds none of the agents read every session blind.
How often is Anthropic web search considered but not chosen?
It was raised as a candidate in 14 sessions without being chosen, and chosen in 56. That is a 80% conversion from considered to chosen.
Where this comes from
Armature ran 7,025 judged sessions with Claude Code, Codex and Cursor inside 77 realistic codebases, and published every run. The numbers on this page come from that work.