How Credits Work
Before you start: research and messages come out of a single generation, and what a lead costs depends on the mode you generate it in.
Base — 1 credit per lead.
High — 2 credits per lead.
Ultra — 10 credits for the first message step, then 2 for each additional step.
Re-generating the same lead within the same campaign and mode is free. Iterate as much as you want once a lead is generated.
Upgrading a lead to a stronger mode charges that mode's cost, because Twain runs the deeper research and writes again.
Generating the same lead in a different campaign counts as a new charge.
You set the mode per lead, so one campaign can hold Base, High and Ultra side by side. See How Modes Work: Base, High, and Ultra.
1. Add Your Leads
To get started, click the + button in the top left and select Campaign from the dropdown. This opens the campaign builder where you'll work through the steps below.
Start by adding the leads you want to reach out to. Twain gives you three options:
Upload CSV — upload a custom lead list from LinkedIn Sales Navigator, Apollo, or another source
Integrations — connect directly via your sales tools (e.g., Clay, HubSpot)
Add manually — paste LinkedIn URLs one by one
When uploading a CSV, map these fields:
Work Email — required for delivery
LinkedIn Profile — used for research-based personalization
Company Name, Title, First Name — optional, but the more context you give, the better the personalization
💡 Lead import runs in the background. You don't have to wait for the import to finish before moving on to the next setup step — Twain will keep pulling in your leads while you configure the rest of the campaign.
💡 If you have a large list, consider splitting it in half. It makes A/B testing easier later without extra setup.
💡 Clean, well-mapped data ensures Twain personalizes correctly and prevents errors during export.
2. Set Up Your Agent
Agents are the brain behind your campaign. They research your leads, apply your personas and rules, and write personalized sequences. Once created, an agent can be reused across multiple campaigns — you don't need to rebuild it every time.
If it's your first campaign, click Create Agent. You'll fill in three sections:
General — Base knowledge and context
Enter your company website and add a short description of the agent's purpose and audience. You can also upload files (e.g., a one-pager or brochure) for additional context. Twain uses all of this to understand your product and value proposition.
The General tab also contains:
Signals — specific data points or triggers you want the agent to look for when researching leads. Twain already researches key lead data by default; signals let you go deeper.
Warnings — disqualifiers or red flags that should stop a lead from being contacted. Twain flags common issues by default; add your own on top.
Personas — Audience and value prop
Twain suggests buyer personas based on your General setup. Each persona has detailed fields: Company/product name, Pain points, Cost of inaction, Solutions, Objections, Competitive advantages, and Social proof. Review the suggestions, edit what doesn't fit, and add new ones if needed.
Personas help the agent tailor messaging to different audience types within the same campaign.
Rules — Research and warnings
A summary view of your Signals and Warnings before you finalize the agent. Once you're happy, click Create Agent.
💡 If you've already built an agent for a previous campaign, select it from the list instead of creating a new one. A well-configured agent improves over time — the more you use and refine it, the better it represents your brand.
Fine-tuning your agent after campaign creation
Once your campaign is live, you can access extended agent settings via the ⚙️ icon next to "Agent" in the campaign view. This opens three tabs: General, Personas, and a new tab that isn't available during initial setup:
Writing style
Control how the agent writes across all campaigns it's used in:
Tone — three sliders: Formality (Casual to Official), Creativity (Direct to Imaginative), Firmness (Humble to Assertive). Use Auto-detect to let Twain infer your style, or Preview to test it.
Blocklist — words or phrases that should never appear in generated copy. Twain already blocks common spam words by default; add your own on top.
Terms — company-specific vocabulary, acronyms, or product names with exact spelling and context instructions.
💡 It's worth setting up Writing style after your first campaign — it applies to all future campaigns using the same agent.
3. Complete the Campaign Brief
Instead of filling in a form, you have a short back-and-forth with Twain. By the time you hit Create, the campaign already knows the reason, the angle, the offer, the sequence structure, and the channels.
Twain walks through five questions:
Why now? Not your goal — the actual reason you're reaching out to this list this week. A funding round, a hiring spike, a competitor move, a recent event. Be specific; Twain will push back if the answer is too generic.
Which angle? Twain proposes a few framings based on your leads and agent. Pick one or type your own.
What's the offer? What does the prospect get for replying? An audit, a benchmark, a checklist, a meeting. Twain matches the offer to the audience: cold lists get nudged toward educational offers, warmer lists toward meeting-style offers.
How many steps? Twain suggests a number and explains why. You can adjust with the slider. Max is 9 steps.
Which channels? Email only, LinkedIn only, or both. You can also describe a custom split in free text.
You can answer in free text, attach a document (a brief, a deck, a transcript) and let Twain pull context from it, or pick from suggested options when Twain offers them. If something doesn't fit, just say so ("that's too corporate", "focus on the economic buyer") and Twain re-proposes. Feedback carries through the rest of the brief.
At the end, Twain shows you the full brief in one block. Review it, push back one more time if anything is off, then hit Create campaign. Nothing is created until you confirm.
💡 The brief context carries through into your campaign's chat panel after creation. When you continue working in chat (refining messages, adjusting tone, asking Twain to change a step) it already knows what you set up here.
💡 The brief stays editable after creation. Change your mind about the angle later and regenerate, Twain uses the updated brief automatically.
4. Generate a Few Leads First
Before generating your entire list, generate one or two leads to see how the research and the messages look. This costs credits but saves you from generating a full list based on a setup that needs adjustment.
Pick the mode in the top right of the lead view. Base is the cheapest way to see whether your setup holds up. If the list needs sharp qualification, start on High instead.
💡 Re-generating the same lead in the same mode is free, so iterate as much as you want once a lead is generated. Upgrading it to a stronger mode charges that mode's cost.
5. Review What Twain Found
Every generated lead has four tabs. Which ones are filled depends on the mode.
Sequence — the messages, step by step.
Research — the lead and company block, the Relationship, and Signals with a source on each one. On High and Ultra you also get Absent signals, which tell you what not to assume about this lead.
Thesis — the prospect narrative, why they fit the persona, and Why now. Available on High and Ultra.
Strategy — what to raise, what to avoid, which channel, which proof points, and in which order. Available on Ultra.
The lead list also shows a Warning if Twain flagged something about the fit, for example wrong company size, wrong region, or wrong persona. Check these before you generate at scale.
Check: does the opener feel relevant? Is the tone right? Is the CTA what you intended? Do the signals actually support the angle you chose?
6. Refine Your Messages
Once you've seen how the first messages look, refine before generating at scale. You have three ways to adjust:
Framework Editor — click the pencil icon on any step to edit the underlying message structure. Use structured placeholders to redefine the narrative flow (e.g., pain-point > solution > proof, or research > question > CTA). Switch between saved frameworks or create a new one. Apply changes with Update Step. Frameworks can be saved and reused across campaigns.
💡 For a full walkthrough of the Framework Editor, see: Framework Editor: Customize Your Campaign Message Structures
Instructions and exact phrases — type directly into the campaign to shape Twain's output:
Instructions guide tone and structure (e.g., "Make this conversational and under 80 words")
Exact phrases ensure specific language always appears (e.g., "Twain injects real-time research into every message")
AI chat panel — use the chat panel (bottom right of the campaign view) to ask Twain to adjust steps directly, e.g., "Make step 2 shorter" or "Change the CTA to a softer ask."
💡 On Ultra, the model makes more of these decisions itself. The stronger the mode, the fewer direct instructions you need to give.
7. Generate at Scale
Once you're happy with the setup, generate messages for all leads.
💡 Twain only exports leads with generated content to your sequencer. Leads that haven't been generated yet won't be included.
8. Set Up A/B Tests
Duplicate the campaign and change one element. Good things to test:
CTA: "Book a 15-min call" vs. "How are you currently handling this?"
Subject line: Offer-based (e.g., "Quick idea for your team") vs. Source-based (e.g., referencing something the lead published)
Opener style: Research-led vs. pain-point-led
Mode: the same list on Base and on High, to see whether the deeper research changes your reply rate
Run both campaigns in parallel and compare open rate, reply rate, and positive reply rate.
💡 A/B testing shows what actually resonates with your audience instead of guessing.
9. Export
Export your generated leads:
Google Sheets — for review or manual editing before sending
Instantly, Smartlead, Lemlist, HeyReach, LaGrowthMachine — direct integration, exports leads with personalized copy straight into your sending sequence
Research, Thesis and Strategy export as well, so you can push the reasoning into your CRM or hand it to reps for calls and manual follow-ups, not just the finished message.
10. Best Practices
Don't skip the agent setup: The quality of your agent — especially personas, signals, and writing style — directly determines personalization quality. It's worth the upfront time.
Reuse and refine your agent: A good agent is an asset. Use it across campaigns and update it as you learn what works, rather than building new ones from scratch.
Be specific in your campaign brief: Vague answers produce generic output. Describe your signal, research intent, and targeting rules as concisely and specifically as you can.
Match the mode to the list: run volume on Base, then upgrade the leads worth upgrading. If qualification matters, start on High, because Base does not catch every fit problem.
Test before scaling: Generate a few leads, refine the idea, agent, or framework, then generate at scale.
Be opinionated: The AI should sound like you. Use your own vocabulary, CTAs, and point of view.
A/B test early: Even a small list split gives you data to work with for the next campaign.