The most valuable thing AI does in searching is not writing your messages. It is telling you the twelve other names the job you are hiring for goes by.
Why this matters more than search technique
A perfectly built search for one job title finds people with that job title. The same job is called something else at a company two miles away, something else again in another country, and something else again by people who came into it from a related field.
Those people are not hidden by a weak search. They are invisible to a narrow one.
This step is where your pool actually grows. Everything else in this module is about searching well inside a pool you have already defined.
Three lists worth asking for
Job titles. Ask for versions used at startups, big companies, and in at least three countries. Startups squash job levels together; large companies add more. Include the jobs people hold just before moving into this one.
Skills and tools. One idea becomes a set: the tool, its main competitors, the short form, the long form, the general category. Someone who has done the same work with a competing tool is usually a strong candidate, and a search naming only your tools misses all of them.
Employers. Which companies produce people who can do this? Direct competitors, other industries with the same problem, and the agencies that serve them. This feeds the map in the next lesson.
Asking for search words, not an essay
By default you get a friendly list with explanations, which you then have to reformat. Ask for the finished thing instead:
- Say exactly what format you want. "Give me one OR group, multi-word titles in quotes, no commentary."
- Give it the limits — region, seniority, industry — so you get relevant variations rather than every possibility.
- Ask for a number. "Fifteen variations" gives you a usable list. "All variations" gives you padding.
- Ask it to mark anything it is unsure about. Those are the ones to check first.
Check before you trust
AI invents job titles, tools and company names, and does it fluently.
An invented title returns nothing, which is harmless. A real but rare title also returns very little — and looks identical. So you throw away a good search word.
The check takes two minutes. Run the new words, look at ten results for each, and sort them into three piles:
| What you see | Verdict |
|---|---|
| Relevant people | Keep it |
| Nothing at all | Probably invented, or genuinely obsolete. Drop it |
| Lots of the wrong people | The word means something else too. Drop it |
That last pile is the expensive one. A word that returns the wrong people in quantity makes your whole list worse, and it is the mistake most likely to go unnoticed, because the number of results looks healthy.
Do not skip the check
This step gives you more for less effort than anything else in the module. It is also where an unchecked invention does the most damage, because it enters right at the top and everything after it inherits the problem.