Prompting recipes

Build a list without prospecting your customers

Upload our customers from this CSV as an exclusion list that always applies. Then find machine builders in Germany and Austria that sell to battery cell makers, and export everything above 0.6 similarity.

create_list (kind exclusion, auto_exclude: true) → search_companies with two clauses (production machinery manufacturer on summary/offerings, lithium-ion battery cell manufacturers on customers) and filters: {country: ["DE","AT"]} → start_export with the same arguments → get_job.

Lookalikes of your best customers

Here are our ten best customers. Find 50 companies like them in the Nordics, excluding our customers.

find_similar with the ten domains (or a seed_list) and filters: {country: ["SE","NO","DK","FI"]}. Seeds never come back; with seed_list the whole list is excluded too.

Sharpen a search with feedback

These five are exactly right, these three are not — refine.

search_companies again with like and not_like. The response says which ten companies to label next (feedback.label_next); eight right and eight wrong examples typically lift precision well above the first ranking.

Size before you pull

How many software companies are there in Hungary, by size?

count_companies with filters: {country: "HU"} and group_by: ["headcount"] — free, and it tells the agent whether an export is 500 rows or 50,000.

One message per segment

Split this list into groups I can write different e-mails to.

segment_companies with list_id (or search). Rule-based segments come with a rule that re-selects the segment in search_companies / start_export.

Tips for agents

• Give each dimension of a question its own clause: what they do and who they sell to are two clauses.
• Never guess a filter value: call list_filter_values first. Values are case-sensitive.
• Prefer literal exclude phrases (is a webshop, to consumers) over semantic exclusion for roles.
• For more than 100 results use start_export, not repeated searches.