How Stage 1 uses Claude
Where Claude does the work in Stage 1, from drafting openers and reading replies to building the Brain and answering help questions, and where it doesn't.
· 3 min read
Every sales tool says it uses AI now. Few say where, or what happens when the AI gets it wrong. Stage 1 uses Claude, from Anthropic, for a short list of jobs, with rules around each one and a person approving what goes out.
Drafting the first message
When you approve an account, Claude drafts the opener for each person Stage 1 plans to reach. It works from the 3x3 research on the account, the signal that put it on the Board, and what your Brain knows about your buyers and your voice.
The default draft is a Hypothesis Opening. Four parts, under 70 words. A research hook, a trigger tied to the person's seat, a guess at the problem they have, and a question that's easy to answer either way. The instructions rule out a pitch, your product or company name, proof points, flattery and the usual openers buyers delete on sight.
When a message is drafted from a different construct, Claude follows that recipe instead. If you've told Stage 1 how you know someone, Claude is told to use what you said as given and never embellish it.
Reading replies
When someone writes back, Claude reads the reply and gives it one label. Interested, not now, referral, objection, not interested, out of office or stop. The label decides what happens next. An out of office closes itself. A stop puts the person on your opt-outs.
Drafting the next move
Then Claude drafts your answer, with a goal for each label. Interested proposes two times. Not now names a month to come back. Referral asks for the intro. An objection gets your firm's own answer when the Brain has one that fits, then one question. Claude is told never to invent clients or numbers.
Building the Brain
The Brain is what Stage 1 knows about how you sell. Claude reads your website and the documents you add and builds it. Your buyers, their problems, your proof, your competitors and your voice.
The rule is to use only what your sources support. No invented client names, metrics or competitors. The one exception is a short list of firms a buyer might compare you with, shown as suggestions for you to keep or remove. More in inside the Brain.
New constructs to test
A construct is a recipe for a first message. Claude proposes variations of the current champion that change exactly one thing, like the hook, the exit or the length. Each one waits for approval before it tests.
It looks at your own sends too. When your rewrites of drafts start winning, or enough of your messages earn a positive reply or a meeting, Claude names the pattern and proposes it as a construct for your firm alone. Those are never shared with other firms.
Answering help questions
Type a question in the help panel and Claude answers it from Stage 1's help topics and the screen you're on, in a few plain sentences. If the topics don't cover it, it says so and points you to the closest one.
When Claude isn't there
Every call has a fallback. Without Claude, openers use the rule-based Hypothesis Opening, replies are labeled by rules, next moves come from templates, the Brain gets a basic read, and help returns the closest topics. Stage 1 keeps running. When a call fails, the app records why.
Why this matters for a services sale
A $300K engagement doesn't start with a clever message. It starts with a message that's right. Right about the account, right about the person's seat, and honest about what you don't know yet.
That's why the guardrails are where they are. Claude drafts. You approve the account, read the draft, change it, and send it. Your name is on it, so the last word is yours. For why that matters, read owning every send, and for the shape of the opener, lead with a hypothesis, not a pitch.
Claude does the reading and the first draft. You decide what goes out.