The real power of the assistant is in how it searches, and in how you can push it deeper.
This article explains what happens under the hood and how to steer it. It assumes you have already started a list (see Building your first media list with the assistant).
How the assistant searches
A broad opening search. It starts by searching journalists across your topic, and by location if your story is place-anchored. This surfaces not only bylined reporters but editors, presenters, columnists, and anchors who do not write under a byline.
Article searches for evidence. It then searches recent articles, by default the last twelve months, to find reporters who have actually covered your angle. This gives byline-level evidence with exact publish dates.
Breaking-news search. For a reactive newsjacking story, it searches the news in near-real-time so it catches reporters already on the story.
Expanding coverage of an event. When it spots a related news event or headline in the results, it will expand outward to find everyone who covered that specific event.
A closer look at borderline picks. If a journalist is a near miss, the assistant can inspect them for deeper context, such as their full job history and more of their articles, before deciding.
Lookup by name. If you already have someone in mind, you can ask for them by name and the assistant will add them to the list. You can search multiple names at once.
Every journalist it keeps is saved to your list with article evidence and a written rationale tied to your brief.
Digging deeper, which is where the value is
A single search rarely gives you the best list. The strongest lists come from steering the assistant into specific, deeper searches. Try asking it to:
Search a specific angle or sub-topic you care about, one idea at a time.
Do more of a particular type of search (e.g. Google News Search).
Provide it a specific search you want it to make.
Find everyone who covered a specific named event or report.
Pull more reporters from a particular kind of outlet, such as trade press or national tabloids.
Tighten to a recency window, for example coverage in the last three months.
Push for more candidates on a niche beat if the first pass came back thin.
Tips for sharper prompts
Keep it to one concept per prompt. The assistant works best with a single, focused task. If you bundle several requests together, it may half-do all of them rather than fully doing the one that matters.
Be explicit about what you want. Wording matters. If you're too vague, the assistant may interpret your prompt differently from how you intended.
Be aware of how strict your wording is. The flip side of vagueness is over-strictness. The assistant takes your rules literally, so the exact phrasing of an instruction changes the result.
For example, you might say:
"Remove any journalists who only covered the royal wedding."
If the assistant finds any evidence that a journalist covers material beyond the royal wedding, it will keep them - because your instruction said "only." A reporter who's written about the royal wedding and one unrelated story no longer qualifies for removal.
Whereas if you say:
"Remove any journalists who appear on this list mainly because they covered the royal wedding."
This gives the assistant room to make a judgement call. It will weigh up why each journalist matched in the first place and drop the ones whose relevance rests largely on that single event — even if they've written about other things too.
The takeaway: strict, literal wording ("only," "exclusively," "every") gives you precise, predictable filtering. Looser, intent-based wording ("mainly," "mostly," "primarily") lets the assistant use judgement.
Neither is better - pick the one that matches what you actually want, and if a filter behaves unexpectedly, check whether your wording was stricter (or looser) than you meant.
When the list is taking shape, move on to Refining, segmenting, and exporting your list.