Ask Brave (search.brave.com) has no manual Deep Research toggle. Mode is auto-classified from query complexity. To get deep, multi-source synthesis instead of a one-line answer, write queries that signal complexity:

Bad (stays shallow):  "What is post-quantum cryptography?"
Good (triggers Deep Research): "What is the current state of post-quantum
cryptography standardization and what are the open problems?"

Access: type query + ?? in address bar (if Brave is default search), click Ask on any SERP, or go directly to search.brave.com/ask. Free, no account, encrypted, deleted after 24h inactivity.

Progress:
- [ ] Identify if the user's question is single-fact or genuinely complex
- [ ] Rewrite as a complexity-signaling query (multi-entity, criteria, "state of the art", "limitations")
- [ ] Submit via Ask Brave; expect seconds (standard) or minutes (Deep Research)
- [ ] Review the synthesized answer and progress/citation trail
- [ ] Use follow-ups to drill down, constrain, reformat, or challenge
- [ ] For broad topics, split into an overview call + targeted sub-topic calls
- [ ] Consolidate results into final deliverable for the user

Step 1 — Diagnose query type. Single fact / named-entity lookup / simple how-to → standard mode is fine, don't over-engineer. Multi-entity comparisons, surveys, edge-case processes, gap-finding, cross-domain synthesis → rewrite for Deep Research.

Step 2 — Rewrite using trigger patterns:

• Multi-entity comparison: name 2+ entities + explicit criteria ("Compare X, Y, Z on A, B, C")
• Survey/state-of-the-art: "current state of...", "landscape of...", "open problems"
• Process with edge cases: "walk through the full flow including edge cases and pitfalls"
• Gap-finding: "known limitations", "unresolved issues with..."
• Cross-domain: "how do A, B, and C interact in..."

Step 3 — Submit and wait. Deep Research shows a visible progress display and takes minutes (dozens of sub-queries against the index). Don't resubmit thinking it's stuck.

Step 4 — Exploit follow-up chat. This is the main lever once in a session:

• Drill down: "Go deeper on [specific subsection]"
• Constrain: "Now focus only on what changed in 2026"
• Reformat: "Redo that as a table"
• Challenge: "What sources contradict that claim?"
• Prune: "Drop the X section, I only care about Y"
• Enrich: "Show me videos on this" / "what are the product options?"

Step 5 — Split and conquer for broad topics. One call rarely covers everything:

1. First call: broad overview ("give me an overview of [topic] with major sub-areas")
2. Identify 2–4 sub-areas from the answer
3. New session per sub-area with specific criteria
4. Synthesize manually or ask for a consolidated summary

Example 1: Input: User asks "I want to understand Rust vs Go vs Zig for systems programming." Output: Rewrite as: "Compare Rust, Go, and Zig for systems programming — concurrency, FFI, deployment, ecosystem." This hits multi-entity + explicit criteria, triggering Deep Research. Follow up with: "Go deeper on the FFI section — specifically how Rust's #[no_mangle] interacts with Go's cgo" to drill into one axis.

Example 2: Input: User asks a one-shot question: "What is a B-tree?" Output: Leave as-is — this is a single-concept definition and will correctly stay in fast standard mode. Don't artificially inflate it; that wastes the user's time waiting minutes for something answerable in seconds.

Example 3: Input: User wants comprehensive research on "the autonomous vehicle industry." Output: Too broad for one call. Step 1: "Give me an overview of the autonomous vehicle industry with the major sub-areas (sensors, ML stack, regulation, safety, economics)." Step 2: Based on returned sub-areas, run separate follow-up sessions, e.g. "How do real-time streaming, ML inference, and edge computing interact in autonomous vehicle architectures?" and "What are the known limitations and unresolved issues with current AV safety validation approaches?"