Pular para o conteúdo
NovoVeja como
Novo

One AI brief that every campaign actually follows

Every campaign's writing style, samples, and instructions now live in a single versioned brief, with a preview of what the AI will actually use before it drafts.

Your campaign used to pull writing guidance from more than one place, and there was no way to see exactly what the AI would read before it started drafting. Now every campaign has one brief, and you can preview its exact contents before a single email goes out.

What changed

Writing style, sample emails, and campaign instructions now live together in a single brief attached to the campaign. Before this change, these pieces of guidance were spread across different parts of setup, and it was hard to know which version the AI would actually pull from when it drafted an email. If you updated your instructions in one place but the AI was reading from another, you had no way to catch it until a draft came back wrong.

The brief is now versioned. When you change your writing style, add a new sample, or edit an instruction, that change becomes the new version of the brief, and the campaign starts drafting against it going forward. You are no longer guessing whether an edit "took." The brief you see is the brief the AI reads, and because it is versioned, you can also look back at what an earlier version said if you need to understand why a batch of drafts from last week reads differently than what the campaign produces today.

Alongside the brief itself, there is now a preview that shows exactly what the AI will use. This is not a summary or an approximation. It is the literal content, in the order the AI receives it: your writing style notes, the samples you uploaded, and the specific instructions you gave for this campaign. If something looks wrong in the preview, it will look wrong in the draft too, and now you can catch that before the campaign sends anything.

This matters most for accounts running more than one campaign at a time. Before, if you had a campaign for outbound to enterprise accounts and a second one for a different offer, it was easy to lose track of which instructions belonged to which campaign, especially after a few rounds of edits. With one brief per campaign and a version history behind it, each campaign's guidance stays scoped to that campaign, and you can always trace back to what changed and when.

It also removes a specific class of quiet failure. Previously, an instruction you thought you had updated could still be read from an older, cached version somewhere in the pipeline, and there was no way to prove that from the outside; you just saw drafts that did not reflect your latest edit. Now the brief and the preview are the same object, so there is nothing hidden between what you wrote and what the AI drafts from.

How to use it

Open the campaign you want to check and go to its AI Instructions section. You will see your writing style, any samples you have attached, and your instructions laid out together, followed by the preview of what the AI will actually use.

If you want to change how the campaign writes, edit any of the three pieces (style, samples, or instructions) directly in that section. The edit saves as a new version of the brief immediately. There is no separate publish step and no delay before the campaign starts drafting against the update.

Before you trust a campaign to run unattended, open the preview and read it the way the AI will. Check that your tone notes are current, that the samples you meant to include are actually there, and that any instruction you added recently (like a compliance note, a phrase to avoid, or a required disclosure) shows up correctly. If a draft ever looks off, this is the first place to check: does the preview match what you expect, and if not, which piece of the brief needs fixing.

If you manage campaigns for a team, the versioned brief also gives you a way to sanity check work someone else set up. You do not have to ask a teammate what instructions they gave a campaign. You can open the brief and read it directly, and if you need to know what changed between two points in time, the version history gives you that answer without having to ask around.

It is also worth reading the preview from a prospect's point of view, not just your own. Ask yourself whether the tone, the samples, and the instructions together would produce an email you would actually want to receive if you were the one being pitched. Because the preview shows the literal inputs the AI works from, it is a much better place to catch a mismatch between what you intended and what will actually go out than reading finished drafts one at a time after the fact.

Why it matters

Cold email campaigns live or die on consistency. If the AI sometimes writes from stale guidance and sometimes from current guidance, your replies get uneven, and it is hard to tell whether a bad draft is a model problem or a setup problem. Consolidating writing style, samples, and instructions into one brief removes that ambiguity. When something reads wrong, you now have exactly one place to look, and the preview tells you in seconds whether the issue is in your guidance or somewhere else in the drafting pipeline.

This also lowers the cost of iterating. Because every edit becomes a new version immediately, you can test a change on a small batch of contacts, read the drafts, and adjust your instructions again within minutes, instead of wondering whether your last edit is even live yet. For a team running several campaigns at once, that speed compounds: less time spent debugging why a campaign is not following instructions means more time spent actually improving what those instructions say.

It also changes how you can respond when something in the market shifts. If you land a new customer with a specific title or industry and want to fold their language into how you pitch a similar segment, you can update the brief once, check the preview to confirm it reads the way you want, and know the next batch of drafts will reflect that change immediately rather than trickling in over the next few days as older guidance gets phased out somewhere in the pipeline.

Availability

The unified brief and preview apply to Autopilot campaigns going forward. If you have older campaigns, opening the AI Instructions section for the first time will show you the current, consolidated view, including whatever writing style, samples, and instructions were already attached to that campaign, so there is nothing to migrate or re-enter by hand.

Todas as novidades

Prospecção que continua quando você desconecta.Leads, emails, follow-ups e respostas, resolvidos.

Conecte caixas ilimitadas e deixe os agentes de IA encontrarem leads, escreverem cada email e fazerem follow-up. Ou envie seus próprios modelos no modo manual. A partir de $29/mês.

$1 por 3 dias · Cancele quando quiser · Sem contrato

Instalar com IA

Seu agente de código configura a CLI e as skills do FirstSales para você.

Prompt para Claude Code

I'm setting up FirstSales (https://firstsales.io), an AI outbound sales platform: campaigns, contacts, a reply inbox, deliverability and lead signals. Help me get my workspace sending real, good emails end to end, from Claude Code.

Read these first so you know how FirstSales works:
- Developer docs and API: https://developer.firstsales.io
- CLI: https://www.npmjs.com/package/@firstsales.io/cli
- Skills for coding agents: https://github.com/firstsalesio/firstsales-skills (read README.md and GUIDE.md)

Setup
1. Install the CLI: npm install -g @firstsales.io/cli
2. Install the FirstSales skills:
   claude plugin marketplace add firstsalesio/firstsales-skills
   claude plugin install firstsales@firstsales
3. Ask me if I already have a FirstSales account. If not, send me to https://app.firstsales.io to sign up.
4. Ask me to create an API key under Settings → API → API Keys, with only the scopes we need.
   Never ask me to paste the key into this chat. Tell me to run this in my own terminal, so the key stays hidden:
   read -rs FS_KEY && firstsales auth login --api-key "$FS_KEY"; unset FS_KEY
5. Run `firstsales whoami` and tell me which workspace and organisation we are connected to.

Then ask me about my business: what I sell, who I sell to (roles, company size, countries), my best customers, and what I use today for outbound (tools, lists, mailboxes). Walk me through:
1. Teaching FirstSales my business: website, offer, proof points and case studies in the knowledge base, so the AI writes like someone who knows the product.
2. Connecting my sending mailboxes and checking my sending domain: SPF, DKIM, DMARC and warm-up. If anything is unhealthy, tell me the exact DNS record to add.
3. Bringing in contacts: import my CSV or CRM export into a clean list, remove duplicates and bad emails, and block domains I must never email (customers, competitors, my own).
4. Building my first campaign: the right list, senders, steps and timing. Write the sequence with me, show me the drafts, and do not launch until I say yes.
5. Setting up Signals if it fits: find people engaging with LinkedIn posts about topics my buyers care about, and turn them into leads.
6. Inviting my team and giving each person the right access.

Once campaigns run, show me how to work the reply inbox from here: which replies need an answer, draft responses in my voice, and move good conversations into deals. If I use Cal.com, connect it so follow-ups can offer my booking link (not the first email).

Throughout, as you learn about my business, proactively suggest and help create:
- Reply templates for the questions and objections my prospects are likely to raise, based on my product, site and codebase if you can see them.
- Campaign ideas and segments worth testing, with the reason for each.

Safety rules:
- Show every change as a dry run (--dry-run) first and wait for my yes. Never send emails, launch campaigns or delete anything without my yes.
- Tell me before anything that spends credits, and how many.

Be opinionated. If the docs give a sensible default, just use it. Don't make me choose between five options.