How to write qualification questions an AI agent can answer

When an AI agent runs outbound, the most important thing you write is not the email. It is the short list of questions the agent answers about every account before anyone gets contacted. Good questions make the agent’s research cheap, its decisions explainable and its first lines specific. Vague ones make all three worse.

What makes a question answerable

• It has a verdict. "Are they hiring salespeople?" can be yes, no or unknown. "What is their sales culture like?" cannot.
• It can be checked in public. Careers pages, the company’s own site, press releases and public profiles are fair game. Guesses about private numbers are not.
• It asks for evidence. Every answer should carry the link or quote it rests on, so a person can check it in seconds.
• It ages. Some answers stay true for a year; hiring and funding change in weeks. Decide how old an answer can be before it is researched again.

Good questions, bad questions

• Bad: "Is this a good fit?" Good: "Do they sell to other businesses?"
• Bad: "Are they growing?" Good: "Have they posted a sales or marketing role in the last 90 days?"
• Bad: "Who should we contact?" Good: "Who leads sales, and do they post on LinkedIn?"
• Bad: "What is their tech stack?" Good: "Does their site run a chat widget or a booking page?"

Keep the list short

Five to ten questions is plenty. Each extra question is research the agent repeats for every account. Put the questions that rule an account out first, so the agent stops early on accounts that will never fit.

Separate fit from personalization

Some questions decide whether to contact an account at all. Others only exist to give the first line something true and specific to say. Mark which is which. A "no" on a fit question removes the account; a "no" on a personalization question just means the email uses a different opening.

Turn answers into copy

An answer with evidence is the raw material for a first line that could not have been sent to anyone else. If the answer to "Have they posted a sales role in the last 90 days?" is yes, with a link, the email can mention the role by name. In Harp, answers to qualification questions resolve as variables in your email templates, and an email whose variable has no answer yet is held rather than sent with a gap.

Refresh only what is stale

Once answers carry their age, the agent does not need to research an account from scratch. It asks for the gaps: questions with no answer yet, or answers older than you allow. That keeps the daily research small even as the account list grows. See Research gap in the glossary.