AI
AI integrations are direct API connections to OpenAI, Anthropic, Perplexity, and BrowserUse.
They give you an AI step as a column: it reads values from the row, does the work, and writes the answer back into the cell.
What each provider is used for
| Provider | Used for |
|---|---|
| OpenAI | Summaries, emails, research insights, and images |
| Anthropic | Deeper, context-aware reasoning and safer outputs |
| Perplexity | Real-time web research and up-to-date answers |
| BrowserUse | Multi-step browser automations driven by AI |
What you run them for
- Drafting — cold emails, LinkedIn messages, and summaries written from the row’s own context
- Classification — sort rows into categories from messy text: industry, company type, buying stage
- Web search — pull current facts about a company or person that no static dataset has
- Scoring — rate a row against your ICP so later, more expensive columns only run on the good ones
- Images — generate visuals per row for personalized assets
The link to AI Agents
AI Agents are pre-built research and summarization actions — web research, earnings summaries, product launch updates, short email writers. AI integrations are the model providers behind them.
The practical difference:
- Use an AI Agent when a pre-built one already does the job. No prompt to write.
- Use an AI integration when you want your own prompt, your own output format, or a provider-specific capability.
Bring your own key
You can run these on Sync GTM-managed keys and pay in credits, or connect your own key for the provider.
When you bring your own API key, that action uses no Sync GTM credits.
Add the key while setting up the Action or Integration, and optionally set it as the default for that provider across the workspace. See API Keys for the steps and Credits for how managed usage is billed.
Getting usable output
An AI column is only useful if the column after it can rely on the answer.
- Pass specific columns as dynamic variables, not everything you have. Type
/to insert one. Precise input, precise output. - Constrain the output when something downstream reads it. For classification and scoring, ask for one value from a fixed set — a label or a number — so a trigger column can act on it.
- Test on one row first. Run the play button on a single row, read the answer, fix the prompt, then run the column.
- Cost scales with rows. An AI column runs once per row. Put it after the cheap qualifying columns, not before them.
Setup
The six generic steps for adding any integration are on the Integrations page.
Authentication differs per provider. Each Integration’s own page in the product has the tutorial.