# Manage Audience Columns > Define columns on an audience, including AI research columns. Defining is free; enrichment runs in ZoomInfo. **Date:** 2026-09-25 **Source:** https://gtm.ai/docs/mcp/tools/manage-audience-columns --- ## Overview A plain column holds values you set yourself. An **AI column** holds a question you asked once, written as a prompt, to be answered for every row in the audience. Ask whether each company recently moved to AI-based pricing, and that question is now a column across all 4,000 rows. This tool defines both kinds. **Defining a column and running it are separate steps.** You configure the column here, then open the audience in the ZoomInfo web app to run the enrichment. Writing a good column prompt is the part that benefits most from a conversation. Describe what you want, read the prompt your AI writes, and adjust it before anything runs. The run itself happens in ZoomInfo, across every row. ## What You'll Get Column definitions with their `columnId` values. Plain columns are ready for values immediately. AI columns are configured and waiting; open the audience in ZoomInfo to run them across your rows. ## Best For - Defining an AI research column to run across an entire audience - Setting up scoring or tiering that synthesizes columns you already have - Configuring first-party call and email context onto account rows - Setting up personalized outreach copy for every contact in the audience - Adding plain columns for values you'll set yourself, like owner or priority - Renaming a column To set cell values, use [Manage Audience Rows](/docs/mcp/tools/manage-audience-rows); to find an existing column's `id`, [Get Audience](/docs/mcp/tools/get-audience). To run the columns you've defined, open the audience in ZoomInfo. ## How It Works **Create.** Omit `columnId` and provide `name`, `dataType`, and `agentInstruction`. **Update.** Provide `columnId`. Renaming is the only supported update, and `dataType` is immutable after creation, silently ignored if you pass it again. Creates and updates can be mixed in one call. ### AI column capabilities `agentInstruction` is what makes a column an AI column, and it must state what each cell should contain and which capability produces it. Four are available today, with more planned: - **Web research.** ZoomInfo's own GTM-tuned research agent, built for go-to-market questions: pricing model changes, job postings, product launches, positioning shifts. Ask a question about a company and it researches and returns an answer. - **Conversation intelligence.** Questions answered from your team's calls and emails with that account. What came up in the last 90 days, whether they raised pricing concerns, what the champion asked for. First-party accounts only. - **Data analysis.** Synthesizes columns already in the audience into a new one, like turning three enrichment columns into a High/Medium/Low score. No external lookup; it reasons over what's already there. - **Email writing.** Personalized outreach copy for each contact, written from that row's context rather than a template with merge fields. The other capabilities become the raw material: research a company, pull what was said on calls, then write the email that reflects both. Whether you specify the scoring logic or let the agent propose it is up to you. Both work. ## Technical Details ### Credits Defining a column over MCP is free. No credits are consumed until you open the audience in ZoomInfo and run the enrichment. The numbers below are planning information: what a column will cost when you run it, so you can size the job before defining a dozen of them. AI columns are charged per cell, so the cost of a run is rows times columns. Cost varies with prompt complexity; typical averages: | Column type | Typical cost per cell | | --- | --- | | Data analysis | 1 to 2 credits | | Email writing | 1 to 2 credits | | Web research | ~3 credits | | Conversation intelligence | ~8 credits | Data analysis and email writing are cheap because they work from what's already in the row. Web research goes out and looks things up. Conversation intelligence is the expensive one; it processes large volumes of call and email transcripts for every row. These per-cell figures apply to enrichment runs in ZoomInfo; a standalone [Conversation Intelligence](/docs/mcp/tools/conversation-intelligence) call over MCP is priced per call, with a 9-credit minimum. Plan for that. A web research column across 4,000 companies runs to roughly 12,000 credits, and a conversation intelligence column across the same audience costs several times more. Two habits worth building: point expensive columns at the subset you actually care about rather than the whole audience, and get the prompt right before the full run. Most prompts want a revision or two, and ZoomInfo lets you check output before committing the audience to it. On a ZoomInfo subscription, your admin sets per-user AI credit limits in Admin Portal under User Management. See [Credits & Billing](/docs/mcp/credits). ### Immutable fields `dataType` cannot be changed after creation, so choose carefully. Omitting it on create is an error; passing it on update is silently ignored. ### Requirements Conversation intelligence columns need your organization to have connected at least one call, meeting, or email source to ZoomInfo. An admin sets that up once for everyone. These columns only return values for accounts your team has actually talked to. Web research and data analysis columns have no first-party dependency and work on any audience, including rows that never matched a ZoomInfo record. Email writing columns are person-level, so they belong on a CONTACT audience. Quality depends on what else is in the row; an email column with no research or conversation columns beside it has little to personalize from. `agentInstruction` is required on every entry, create and update. It's per-column and self-contained: the downstream agent reads this one field with no conversation access. Vague instructions produce vague columns. "Company info" is not a column. "Did this company change its pricing model in the last 12 months? If so, describe the change in one sentence" is. ### Automatic Tool Selection The AI reaches for Manage Audience Columns when you ask something per row rather than about the set. "Tell me which of these changed their pricing" is a column. "How many of these changed their pricing" is a question about a column that already exists, so the AI answers from the audience instead of defining anything new. ## Example Prompts > *"Add a column that finds out whether each company has moved to usage-based or AI-based pricing in the last year."* > *"Set up a column that scores each account High, Medium, or Low based on the intent and hiring columns."* > *"Add a column asking what each of these customers said about our pricing on calls in the last 90 days."* > *"Add a column that writes each contact a first-touch email referencing the pricing change and their role."* > *"Rename 'Col 3' to 'Pricing Model Change'."* Once the columns have run, ask questions across them with [Audience Analysis](/docs/mcp/tools/audience-analysis).