Use Custom Prompts
The #Column System lets you add custom AI-powered columns to any table - scraped from Maps, imported from CSV, or pasted from anywhere. Ask any question, and an AI agent answers it for every row.
How It Works
- Click the
+on a column header (or the Enrichment button) to open the preset picker. - Choose Custom Prompt (under Have Idea or Have Website).
- Write your prompt and mention columns with the
#system - pick a column and it becomes a{{Column}}chip in your prompt, so the AI reads that value for each row. - Run it with Run first 10 to test, then Run all. A new column appears with the results.
The Prompt Editor
- Mentions: click/type
#to reference a column → it inserts a{{ColumnName}}chip. The agent substitutes each row’s actual value at run time. - Single or Multi column mode: ground your prompt on one column or several.
- Character limit: 1,000.
- Behind the scenes, your prompt drives a tool-calling agent - it can look up the record, scrape the website and subpages, extract emails, and search Google as needed to answer.
Popular Use Cases
- Founder Name Extraction: find the CEO/founder/owner’s name from each website
- Service Specialization: does this clinic do implants, cosmetic work, or aligners?
- Technology Stack Detection: WordPress, Shopify, Webflow, or custom React?
- Anything at all - if a human could answer it from the row’s data + the website, the agent can try
Example Prompt
Prompt / Code
Find the owner or founder of {{Name}} from {{Website}}.
Look for their full name and any direct email address.
If no direct email exists but a name is found, suggest
the most likely pattern (e.g. first.last@domain) and mark
it as "pattern" instead of "verified".
Ready-Made Prompt Library
Don’t start from scratch - the AI Prompts category has ready-to-paste prompts for the most common jobs:
| Article | What’s inside |
|---|---|
| #Column Data Extraction Prompts | Owner/founder names, contact & social discovery, service specialization, tech stack, lead scoring |
| Cold Email Templates | Full cold email sequences built from your enriched columns |
| WhatsApp Scripts | WhatsApp-first outreach scripts for verified numbers |
Where Results Land
- A new column (AI columns can be renamed, moved, and exported like any other)
- (i) tooltip on filled cells shows the agent’s reasoning trail
- Failed rows stay empty (vs “None” for checked-and-not-found) - see Run Controls
All the Numbers (Technical Limits)
Every hard limit in DataMint, in one table:
| Limit | Value |
|---|---|
| Custom Prompt length | 1,000 characters |
| LLM run pacing | 3 rows concurrent, ~500ms between batches |
| WhatsApp check pacing | 1 at a time, 0–60s delay (default 15s) |
| Serper results per query | up to 10 |
| Reoon timeout | Quick 15s / Power 70s |
| Sub-page crawl (deep scrape) | max 2 contact pages, only if homepage has no email |
| Website Workflow | max 4 iterations / 6 tool calls per row |
| Support Chat context | last 6 messages |
| Table soft limit | ~1,000 rows before a performance prompt |
| Session heartbeat | every ~1 hour |
| Version-check cache | 12 hours |
Related
- AI Enrichment Presets - the full preset list
- Cold Email Templates · WhatsApp Scripts - the rest of the prompt library
- Manage Data Table - how runs fill the columns