- The Evidence Chain — Five Independent Confirmations in Sequence
- The Story Nobody Is Reporting — Claude Has a Second Search Provider
- When Claude Actually Searches — The 36.6% Rule
- How Claude's Search Architecture Works — Under the Hood
- How Claude Compares to Every Other Major AI Platform
- Your Brave-First Claude Visibility Strategy — Six Actions in Priority Order
- Frequently Asked Questions
- The Bottom Line
The question "what search engine does Claude use?" sounds deceptively simple. Type it into any AI visibility forum today and you'll get two flavours of the same incomplete answer: "Brave Search, confirmed." Both the confidence and the incompleteness matter — because as of May 2026, Anthropic quietly added a second web search provider to its official trust documentation, and the 86.7% Brave overlap figure that everyone cites from March 2025 has been revised downward in a larger June 2026 re-study. Getting the full picture right changes how you build a Claude visibility strategy — and separates practitioners using current evidence from those recycling a fifteen-data-point finding from launch day.
I've managed AI search visibility across healthcare, legal services, hospitality, and e-commerce client accounts for five years. Understanding exactly where Claude retrieves its web data — not just directionally but precisely — directly affects which actions I prioritise when a client asks why their competitors appear in Claude's answers and they don't. This article gives you the most complete, most current picture available as of July 2026: every confirmed data point, the honest caveats around what's still unconfirmed, the newly discovered second provider, and the practical SEO strategy that follows from all of it.
The Evidence Chain — Five Independent Confirmations in Sequence
The case for Brave Search rests on five distinct evidence types, not one. Understanding each one — and what it does and doesn't prove — gives you a calibrated picture of how strong the Brave confirmation actually is.
The Story Nobody Is Reporting — Claude Has a Second Search Provider
While the Brave relationship is well-documented, Anthropic quietly added a second web search subprocessor on May 6, 2026 — nearly fourteen months after Brave. The provider is TurboPuffer, and almost every AI visibility publication as of July 2026 has not reported it.
The addition of TurboPuffer — a vector database optimised for semantic similarity search — alongside Brave's keyword index is architecturally significant. Brave retrieves results based on keyword and query-term matching. TurboPuffer enables semantic retrieval: finding content that means the same thing as the query even when the exact words don't match. The two systems together suggest Claude's search infrastructure is evolving toward a hybrid retrieval model — keyword precision from Brave, semantic coverage from TurboPuffer — rather than a single-source keyword lookup.
For SEO practitioners, the honest strategic implication of TurboPuffer's addition is this: optimising purely for Brave keyword ranking remains your most direct, measurable lever for Claude citation. But the existence of a semantic retrieval layer means content that clearly expresses topical expertise — even without exact keyword matching — may increasingly surface in Claude's responses through TurboPuffer's contribution to the retrieval stack. The "write for humans with genuine expertise" advice that has always underpinned good content strategy gains an additional technical justification.
"When I audit a client site for Claude visibility, my first three checks are now: BraveBot access in robots.txt, Bing Webmaster Tools sitemap submission, and Brave-specific rank tracking on their ten most important commercial queries. The TurboPuffer finding adds a fourth check that I'm still evolving — it points toward semantic content quality signals that keyword tracking doesn't capture. For now, I treat it as additional evidence that the 'expert layer' principle I already apply to every content audit is the right direction: content that only this organisation could write, drawn from direct experience rather than summarised from public sources, is exactly the kind of semantically distinctive content a vector search system rewards. The TurboPuffer addition makes the expert layer not just a Google quality signal but a Claude citation signal too."
When Claude Actually Searches — The 36.6% Rule
Understanding which search engine Claude uses matters most when Claude actually uses it. And here's what separates Claude from every other major AI platform: Claude only activates web search for approximately 36.6% of prompts — compared to roughly 90% for ChatGPT. The query type you ask about dramatically affects whether Brave (and TurboPuffer) ever get called.
- 🟢 "Best [product/service] 2026" — recency signals trigger search 81% of the time
- 🟢 "Top 10 / leading / ranked" — ranking-focused queries trigger at ~67%
- 🟢 "Near me / in [city]" — location queries trigger at ~55%
- 🟢 "X vs Y / alternatives to X" — comparison queries trigger ~50%
- 🟢 Queries containing specific recent events or named current developments
- 🟢 Queries about current prices, availability, and specifications
- 🔴 "How does X work" — conceptual / explanatory queries
- 🔴 "What is X" — definition queries covered in training data
- 🔴 "Steps to X / how to X" — procedural how-to queries
- 🔴 Topics thoroughly covered in Claude's training knowledge
- 🔴 Writing, creative, and code generation tasks
- 🔴 Mathematical and logical reasoning queries
The practical consequence: Brave ranking only produces Claude citation on the ~36.6% of queries where Claude's web search activates. Your highest-return investment targets the trigger categories — comparison pages, best-of lists with current-year signals, and location-specific content — because these are the query types that reliably send Claude to Brave's index rather than drawing from training data alone.
How Claude's Search Architecture Works — Under the Hood
How Claude Compares to Every Other Major AI Platform
Positioning Claude's search infrastructure against every competitor reveals where your optimisation effort reaches multiple platforms at once and where Claude requires separate investment.
| AI Platform | Search Backend | Searches How Often | Alignment with Backend | SEO Priority |
|---|---|---|---|---|
| Claude (Anthropic) | Brave Search + TurboPuffer (vector) | 36.6% of prompts | 79.2% — very high | Brave ranking, BraveBot access, Bing WMT, semantic content quality |
| ChatGPT (OpenAI) | Bing (official) + Google supplementary | ~90% of prompts | 26.7% Bing alignment | Strong Google + Bing SEO; Google carries more influence than Bing despite official Bing partnership |
| Perplexity | PerplexityBot own crawler + Bing | Nearly all prompts | N/A — own primary crawl | Allow PerplexityBot; strong Google indexation correlates; freshness rewarded strongly |
| Google AI Overviews / AI Mode | Google Search Index — direct | Query-triggered | 100% — same index | Standard Google SEO; schema, EEAT, structured content, inline-link-eligible content structure |
| Microsoft Copilot | Bing Search — direct | Most prompts | High — built on Bing | Bing Webmaster Tools sitemap, Bing-specific crawl access |
| Meta AI | Bing Search | When web data needed | Direct Bing integration | Bing SEO; Facebook/Instagram brand presence for cross-signal benefit |
Claude is the only major AI platform using Brave Search and TurboPuffer. Every practitioner who optimises for Google and Bing covers ChatGPT, Copilot, Meta AI, and Google AI Overviews — but none of that work directly targets Claude's primary search backend. Claude requires deliberate Brave-specific optimisation as a separate, additive investment to capture the full AI search visibility opportunity.
Your Brave-First Claude Visibility Strategy — Six Actions in Priority Order
Audit robots.txt for BraveBot Access Right Now — This Is Action One
Open your robots.txt and search for BraveBot. Many sites catch it in blanket disallow rules designed to block scrapers. A blocked BraveBot removes every page from Brave's index — and from Claude's search results — regardless of any other optimisation you do. Fix it immediately: add an explicit User-agent: BraveBot Allow: / rule or remove the blanket rule catching it. Verify the fix using Brave's URL submission tool at search.brave.com/webmaster.
Submit Your Sitemap to Bing Webmaster Tools Today
Brave's independent crawler benefits from the broader multi-engine crawl ecosystem. Pages well-indexed across multiple major engines consistently outperform Google-only properties in Brave's rankings. Bing Webmaster Tools sitemap submission is free, takes five minutes, and reaches Brave's supplementary indexation pipeline alongside Bing's own index — which also powers ChatGPT, Copilot, and Meta AI. One action, four platforms. Do it before any content work.
Add Brave Search to Your Rank-Tracking Stack
You cannot manage what you don't measure. Add Brave as a separate rank-tracking channel in your preferred rank-tracking tool for the queries most likely to trigger Claude's web search: comparison terms, best-of queries with current-year signals, location queries, and ranking-focused terms. Tracking Brave-specific positions gives you the only observable proxy for Claude citation probability that correlates directly with what Claude actually serves.
Build Content That Specifically Targets Claude's Search Trigger Categories
Given Claude's 36.6% search activation rate, the highest-return content for Claude visibility targets the categories that reliably trigger search: comparison pages ("X vs Y"), current-year best-of lists ("best healthcare CRM 2026"), location-specific guides, and ranking-focused content. Include explicit year signals in title tags and H1s for these page types — Claude's internally-constructed search queries frequently include current year terms, and explicit year signals improve the probability of appearing in freshness-driven search retrievals.
Implement Schema Markup and Clean HTML for Dynamic Filtering Survival
Since Claude's February 2026 dynamic filtering update strips boilerplate before content reaches its context window, pages with clean semantic HTML and properly structured schema markup extract more accurately than JavaScript-heavy or navigation-cluttered pages. Implement Article, FAQPage, and Person schema on your content. Ensure your key content appears in crawlable HTML — not loaded dynamically after the initial HTML response — so it survives the filtering step and reaches Claude's reasoning process intact.
Write the Expert Layer That TurboPuffer's Semantic Retrieval Rewards
TurboPuffer's addition as a semantic retrieval provider suggests Claude's search architecture increasingly rewards content expressing genuine topical expertise — not just content matching keyword terms. For every key content cluster, add the layer of firsthand expertise only your organisation can provide: case observations, proprietary data, practitioner perspective, named author credentials. This "expert layer" content performs better in both Brave's keyword ranking and TurboPuffer's semantic similarity retrieval — making it the highest-compounding single investment in your Claude visibility strategy.
A substantial portion of Claude-originated traffic reaches GA4 as "Direct" or generic "Referral" because Claude doesn't consistently pass source referral data. Build a custom GA4 channel group with a regex filter for claude.ai referrals to reclassify this traffic into a named, measurable AI channel. Do this before investing in Brave optimisation so you have a clean baseline to measure improvement against. Without this, every gain from Brave-specific optimisation becomes invisible in your standard reporting.
Frequently Asked Questions
The Bottom Line
Claude uses Brave Search as its primary web search engine — confirmed by five independent evidence sources including Anthropic's own subprocessor list and a 79.2% citation overlap study across 400 queries in June 2026. As of May 6, 2026, a second provider named TurboPuffer now also appears on that same subprocessor list under "Web Search," suggesting Claude's retrieval architecture is evolving toward a hybrid keyword-plus-semantic model. Audit BraveBot access in your robots.txt today. Submit your sitemap to Bing Webmaster Tools. Add Brave to your rank-tracking stack. Build content targeting Claude's search trigger categories. Implement clean schema markup that survives dynamic HTML filtering. And add the expert layer that TurboPuffer's semantic retrieval rewards. Your Google rankings give Claude partial coverage. Your Brave rankings, clean crawlability, and genuine content expertise give Claude the complete picture it needs to cite you.
.webp)
Driven by advanced SEO expertise, deep marketing analytics, high-impact content strategy
With 5+ years of hands-on experience, I specialize in holistic search strategies that don’t just rank—they drive real, measurable business growth. I’ve worked across industries including healthcare, hospitality, legal, e-commerce, and professional services, helping brands dominate their target markets. My approach bridges the gap between raw data and creative execution. Every strategy I build is rooted in rigorous market analysis, structured SEO frameworks, and tailored content ecosystems—no templates, no shortcuts. Whether you’re a single-location brand or scaling across multiple cities, I create data-driven marketing systems designed to compound results and grow with you.
Want Higher Visibility in AI Search?
Boost your presence across Google AI Overview, ChatGPT, Claude, Gemini, and Perplexity with a free SEO audit from DigitalArka. We'll uncover content gaps, schema improvements, topical authority opportunities, and AI search optimization strategies to help your website attract more qualified organic traffic.
Claim Your Free SEO Audit →