An AI column is only as good as its question. The answer has to be findable in a person's LinkedIn profile or a company's website, and the question has to say exactly what you want back.
Ask about something visible
Good questions point at something a page or a profile would show:
| Instead of | Ask |
|---|---|
| Is this a good company for us? | Does the company sell to hospitals or clinics? |
| Are they growing? | Does the careers page list open sales roles? |
| Do they have a budget? | Does the pricing page show an enterprise plan? |
| Is this person technical? | Has this person held an engineering or IT role? |
If a human couldn't answer the question from the website or the profile, the AI can't either, and you'll get no data.
One question per column
"Does the company have an API and a mobile app?" is two questions. If one is yes and the other no, the answer is meaningless. Make two columns.
Match the answer type to the question
- Yes / No for questions with a clear yes or no. Phrase it so "yes" is the interesting answer: Does the company offer a free trial?
- Number for counts: How many locations does the company list?
- Pick from list when you'll filter or sort by the answer: Pricing model with options Per seat, Usage-based, Flat fee, Not shown. Add an option for "not shown" so the AI doesn't force a guess.
- Short text for a short description: Which industries does the company serve?
Lists you'll filter on work best with Yes / No or Pick from list. Free text is harder to filter.
Say what counts
Add a short definition when the word is ambiguous:
- Is the company B2B? Count it as B2B if most of its customers named on the site are businesses.
- Is this person a decision-maker? Count founders, C-level, VPs and heads of a department.
Person or company?
A question about the person is answered from their LinkedIn profile and work history. A question about the company is answered from its website. The line Will be answered using under the question shows which. If it picked the wrong one, rephrase so the subject is clear: Does this person… or Does the company….
Test, then fix
Test on 3 rows runs the question on three rows before you save. For each, read the answer and its evidence:
- Wrong answer with good evidence? The question is ambiguous. Add a definition.
- no data everywhere? The information isn't on the profile or site. Ask about something visible instead.
- Right answers? Save and fill the column.
It's much better to catch a vague question on three rows than after filling a list of 2,000.