Prompt like an operator

Filters are a precise way to narrow a list, but they can only express conditions that already exist as fields. The agent lifts that limit by reading a plain-language description and assembling the search for you, which means the quality of your results now rests on the quality of your request. This lesson covers how to write those requests well, so you can describe an audience the way you would explain it to a colleague and get back a list worth working.

Where filters stop being enough

Structured filters work well when your target is a value that already exists as a field, such as industry or headcount. They reach their limit the moment your audience depends on meaning or on a combination no one stored as a checkbox. A described search goes further because the agent reads your intent and assembles the query, including steps a filter cannot phrase.

The parts of a strong prompt

A clear request reads like a short brief, and the more specific you are about each part, the closer the first result lands.

• Who the people are, by role, seniority, or function.
• What kind of company they work for, by industry, size, geography, or technology.
• Any recent event that matters, such as funding, hiring, or a leadership change.
• Anything to exclude, such as current customers or accounts already in your CRM.

VP of Sales at US-based B2B SaaS companies that raised funding in the last 6 months and use [competitor], excluding our current customers.

Refine across turns

Treat the agent as a thread rather than a single shot. Read what comes back, then adjust your description in the same conversation, since each turn keeps the context of the last. Narrowing, widening, or correcting in place produces a better list than starting over, and it is how operators converge on the right audience quickly.

Give the agent your ICP

When you tell the agent your definition of a good account, everything it returns sharpens. Share your ICP, your best customers, or the traits that matter to you, and the agent works from your standard rather than a generic one. The more context you provide up front, the less you have to correct later.

Common mistakes and the fix

• Vague requests return vague lists, so name the role, the company type, and the event.
• Asking for too much at once buries the signal, so build in steps and refine across turns.
• Leaving out exclusions repeats accounts you already have, so tell the agent what to drop.
• Describing the tool instead of the outcome limits the agent, so say what you want, not how to get it.

Try it in Landbase

1. Take an audience you would normally build with filters and describe it to the agent in two or three sentences.
2. Cover who the people are, what kind of company they work for, and any recent event that matters.
3. Review what comes back, then refine your description in the same thread.
4. Stop when the list is one you would be happy to act on.

Prefer the CLI? The same described search runs from your terminal:

landbase-cli search "VP of Sales at US B2B SaaS companies that raised in the last 6 months, excluding current customers"