The research engine plans its searches from your question and the context you give it. A vague question gets a broad survey. A specific question tied to a decision gets a report that leads with the answer you need. This guide covers what to include and shows a few before-and-after examples.
The five things worth stating
1. The decision
Say what choice the research feeds. "Pick a primary database for a new SaaS backend" tells the engine what a useful finding looks like. Put it in What decision will this support? on the new research form (up to 4,000 characters). Before the first round, the engine turns your question and this context into a research brief (objective, scope, success criteria, exclusions), and that brief steers every round of searching.
The final write-up works from your question and the sources the run kept. So if the decision is the point, say it in the question as well: "Should we pick SQLite or PostgreSQL for..." beats "Compare SQLite and PostgreSQL".
2. The audience
Who needs to trust the answer? is a short field (160 characters). "Engineering leads", "an investment committee" or "a client's marketing team" goes into the same research brief and shapes which sources the engine looks for.
3. Constraints
Budget, team size, existing stack, regulatory limits. Write them into the question or the decision field. "We have two engineers and no dedicated DBA" rules out answers that need one.
4. Region
Markets differ. "In the EU", "US mid-market" or "Southeast Asia" narrows the search and keeps the report from mixing figures that don't apply to you.
5. Time frame
Say which period matters: "between 2023 and 2026", "announced in the last 12 months", "current pricing as of this quarter". Put the time frame in the question itself. The clarify step also asks about it if you leave it out.
The clarify step
After the first screen, the form asks up to three short questions: what decision or deliverable this supports, which markets, geographies, organizations or technologies are in scope, and what time horizon matters. Your answers go into the research brief along with the question. Skipping is fine, and Farside makes reasonable choices, but answering scope and time frame is the cheapest way to get a sharper report. This step is free.
Research types
Under Optional organization and scope, the Research type menu has these options:
- Auto-detect (the default,
generalin the API) - Competitive intel
- Market research
- Market sizing
- Technical
- Content strategy
- Social strategy
- Academic
- Investigative
- Strategic
Every type produces the same report outline. The type sets the investigation tracks you see on the review screen (for example, Technical plans around architecture, constraints, benchmarks and operational risk, and Market sizing around definitions, top-down estimates, bottom-up estimates and sensitivity). It also decides which documents go into a dossier, either the one included with Deep research or one you add later. If you're not sure, leave it on Auto-detect.
Steering the sources
- Prefer domains: comma-separated domains to prioritize, up to 30. In the first round the engine runs targeted searches on up to three of them.
- Exclude domains: comma-separated domains to drop, up to 50. Results from these domains (and their subdomains) are filtered out.
- Reference files: upload PDF, DOCX, text, Markdown or CSV files (up to 25MB each) to include private evidence. Wait for each file to finish extracting before you start the run. The form blocks the start until they're ready.
Good and vague, side by side
Vague: "Tell me about observability tools." Better: "How are observability vendors pricing per host versus per GB in 2026, and which model costs less for a team ingesting about 2 TB of logs a month?" The better version names the comparison, the time frame and the workload that decides the answer.
Vague: "Is SQLite good?" Better: "Should a two-person team use SQLite or PostgreSQL for a read-heavy SaaS backend on a single server? We expect under 50 writes per second for the first year." Now the report can give a recommendation instead of a list of trade-offs.
Vague: "What's happening in EV charging?" Better: "Which EV charging networks in Germany expanded fastest between 2023 and 2026, and what did their pricing do over the same period?" Region and time frame keep figures from other markets out of the tables.
Vague: "Competitor analysis for our product." Better: "How do the three largest self-serve survey tools position against each other on pricing, integrations and data residency for EU customers? This is for our Q3 pricing review with the product council." The decision and audience go in their own fields; the question stays focused on what to find.
Make it reusable
If you run the same kind of brief often, save it. On a finished report, open the ... menu next to Export and choose Save as reusable brief, or create one from scratch under Workflows, in the Briefs section. A saved brief keeps the decision, audience and research standard. You can run it again, re-run it on a schedule as a monitor, or use it as the research step in a workflow.