Claude for sales: how a rep runs the day from Claude, and which jobs need a schedule

1 November 2026Tulina

The pages that rank for "Claude for sales" are Anthropic's own sales page and tutorial, two setup guides and a lot of video. They show what Claude can do in a sales conversation. Fewer of them say which parts of a rep's day have to happen when the rep is not in the chat. This page covers both: the jobs, the setup, the prompts, and the line between a chat and a scheduled agent. That line is drawn from a count of Anthropic's own Sales plugin skills, made on 1 November 2026.

It is written by Tulina, an agentic RevOps workflow builder, and the section that uses Tulina is marked. Everything above it works without us. Anthropic's product facts were read on its own pages and its plugin repository on 1 November 2026. This page is part of our guide to running go-to-market from Claude, which covers the whole motion. This page covers the rep's share of it.

How can I use Claude for sales?

You use Claude for sales by connecting it to the tools a deal lives in, then giving it one job at a time: research this account, prepare this call, write the follow-up for this call, update this deal, review my pipeline. Anthropic's sales page names the same five handoffs: account research, meeting prep, proposals, pipeline and forecast, and CRM updates. It also lists connections to Salesforce, HubSpot, Gong, Fireflies, Clay, ZoomInfo, Apollo, Outreach, Docusign, email and calendar.

A rep's day, job by job:

Moment	What you ask Claude for	Connectors it needs	What the rep still decides
Morning	Today's external meetings, each with its account, open deal and last email	Calendar, CRM, email	Which meeting needs real preparation
Before a first call	A one-page account brief: what they sell, recent news, who you know there, open deals	Web search, CRM, data source	What to ask, and what to leave out
After a call	A recap email drafted from the transcript, the next step and its date on the deal, a note with the buyer's words	Call recorder, CRM, email	What the email promises, and whether it goes out
Between calls	Replies sorted into yes, no and question, with a draft for each question	Email, CRM	The answer to a real objection
Prospecting block	Openers for a short list, each citing one dated fact	Web search, data source, CRM	Who is worth writing to at all
Friday	Deals with no next step, close dates in the past, stages that have not moved	CRM	What to do about each one

Every row is a job Claude can do in one sitting while the rep watches. The rows that start with a time of day are the ones that later move to a schedule, and the rest of this page explains why.

FIG. 01 Where a deal actually lives: the stage in the CRM, what was said in the call recorder, what was promised in the inbox, what happens next in the calendar, and a few more tools besides. The dashed runs are the copying a rep does between them after every call. Claude for sales is mostly that copying, done by a model that can read every one of them.

How do you set up Claude for sales yourself?

You set up Claude for sales in four parts: connectors, a written context, one skill or saved prompt per job, and a rule that nothing is sent without you. Each part below works on a paid Claude plan, in chat or in Cowork.

Connect the tools a deal touches. Add your CRM, email, calendar and call recorder from the connectors directory. A connector acts with your own permissions in that tool, so Claude reaches what you can reach.
Write your context once. Who you sell to, the title that signs, where deals stall, your rules ("never send pricing before a budget conversation"), and two emails you wrote and liked. Put it in a project or a file Claude reads, so no conversation starts from nothing.
Add the Sales plugin, or write your own skills. Anthropic's Sales plugin ships skills for each job. Its tutorial has you tailor the account-research skill to your company before the first run, then run it as /account-research [the account].
Test on an account you already know. Run the brief on a customer you could describe from memory, and fix what it got wrong before you trust it on one you can't.
Quote, never paraphrase. Ask Claude to take the buyer's words from the transcript, not from the summary. A summary paraphrases, and a follow-up that quotes the buyer reads as written by someone who was listening.
Draft, never send. Have Claude save drafts and propose CRM changes with the value before and after. The send and the stage stay your call.

A prompt for the after-call job, which works with any CRM, call recorder and email you have connected:

The call with [account] just ended. Do not send anything.

1. Find the call in my call recorder. Use the transcript, not the
   summary, for anything you quote.
2. Find the open deal for this account in my CRM. Match on the
   attendees' email domain. If there is no deal, stop and tell me.
3. From the transcript, list: what they said they need (their words),
   what they pushed back on, what each side agreed to do and by when.
4. Propose the deal's next step with an owner and a date, and one note
   with the quotes. Show me the current value of each field you would
   change. Do not move the stage.
5. Draft a reply on our last email thread: one line that quotes them,
   what we agreed and by when, one ask. Under 100 words. No pitch.
6. Tell me what you could not find.

The six steps are the whole job. The "do not move the stage" line matters most, because a model that reads an enthusiastic call will promote a deal nobody qualified. Our page on the sales call follow-up email has templates for the email itself, by moment.

What are the best Claude skills for sales?

The best Claude skills for sales are the ones that match a job you repeat every week, and the quickest set to start from is Anthropic's own Sales plugin. Its README, version 2.1.0, lists 36 skills in five groups: your day, accounts and prospecting, deals, pipeline and forecast, and customers. It says the skills "read your CRM's own stages and fields instead of assuming one vendor's setup". It also says that "every skill works without any connectors", from uploaded files.

The ones a rep reaches for first:

Skill	What it does, in the plugin's words
`call-prep`	Pre-call brief: attendees, account history, prior call context, open opportunity status, discovery questions
`call-summary`	A transcript or call notes into a follow-up email draft, a team summary and proposed CRM updates
`account-research`	Company overview, recent news, likely priorities and fit against your ICP
`daily-briefing`	Today's meetings with account context, deals closing soon, customer emails waiting, top actions
`deal-signals`	Deals gone quiet, slipping close dates, champion changes, competitor mentions
`pipeline-review`	Coverage, aging, conversion and at-risk deals, stage by stage
`handle-objection`	Responses and proof points from your own history
`crm-hygiene-check`	A read-only audit for missing fields, stale dates and stage mismatches

Anthropic also announced Salesforce in Claude, in beta, with "37 pre-built skills" that work from a seller's book of business. That is a separate set, for Salesforce teams.

FIG. 02 Two account briefs that read the same until one of them shows where each line came from and when. Anthropic's sales page promises briefs 'with citations linked to sources'. That is the property to test before trusting a skill, because a rep walks into the call with whatever the brief said.

Which sales jobs should leave the chat?

The sales jobs that should leave the chat are the ones that start from a clock or an event rather than from a question. A question ("what do we know about this account?") needs a rep asking it. A morning briefing, a Friday pipeline review or a follow-up that should exist within the hour of a call needs something that starts by itself.

We sorted the 36 Sales plugin skills by what starts them, reading each description in the README on 1 November 2026. The sort is our reading of the descriptions, not a field in the plugin:

Started by	Skills	For example
A question from the rep	20	Account research, a deal review, an objection
A clock	11	The daily briefing, deal signals, the forecast
An event	5	A meeting coming up, a call ending, a lead arriving

The full sort, skill by skill:

A question (20): setup, schedule-meeting, account-research, account-context, stakeholder-map, draft-outreach, account-tiering, account-plan, expansion-whitespace, deal-review, deal-advance-gap, deal-slip-scenario, close-plan, handle-objection, competitive-intelligence, create-an-asset, update-opportunity, rep-context, customer-health, customer-voice.
A clock (11): daily-briefing, inbox-sweep, end-of-day, weekly-wrap, deal-signals, pipeline-review, crm-hygiene-check, forecast, team-pipeline, win-loss-review, renewal-radar.
An event (5): call-prep before a meeting, call-summary and log-activity after a call, lead-triage and route-lead when a lead arrives.

So 16 of 36 are jobs a chat can only do if the rep remembers to open it at the right time. Anthropic answers the clock half with Cowork scheduled tasks. They are available on all paid plans, can use connected tools, skills and plugins, and run "even when your computer is asleep or the Claude Desktop app is closed". The plugin's own admin note says that for scheduled runs, send, post and write tools should be set to "ask", "so nothing goes out or changes without a person approving it".

The event half is harder. A call ending is not a time of day, and an hourly task that looks for new calls is the usual workaround. Our guide to running Claude on a schedule compares the ways to do it and what breaks.

	A Claude chat	A scheduled or webhook agent
Who starts it	The rep, by typing	A clock, or an event in another tool
Good sales jobs for it	Account briefs, objection prep, a deal review, an asset for one buyer	Morning briefing, after-call follow-up, deal signals, Friday pipeline review
Who answers a question mid-run	The rep	Nobody: the instructions must say what to do
How you know it worked	You read the answer	You read the trace of the run

FIG. 03 A pipeline review done once in a chat is true on the day it is run. The same review rerun every Friday by an agent is true every Friday, because it is rebuilt from that week's CRM. The test for moving a sales job out of the chat is whether it needs to be right next week too.

How do you run Claude for sales with Tulina?

With Tulina, the after-call job above becomes a saved process that a hosted agent runs every hour of the working day, through connectors to the call recorder, CRM, inbox, tracker and chat your team already uses. Every step, every tool call and its cost is recorded in a run anyone in the team can open. The rep still sends the email and moves the stage: the run leaves a draft and a dated next step, never a sent message.

Tulina is an agentic RevOps workflow builder: AI agents that run sales and GTM processes on a schedule or a webhook, across the tools a team already uses, with every step and its cost visible. Once Granola, HubSpot and Gmail are connected, paste this into Claude or ChatGPT:

Follow through on my sales calls from today, then schedule it.
Never send anything.

1. Find today's calls: granola_content(op="list_notes",
   created_after=<today>). For each one, granola_content(op="get_note",
   note_id=<id>) for the attendees and summary, and
   granola_content(op="get_transcript", note_id=<id>) for any quote.
2. Find the deal: hubspot_object(op="search", object_type="contacts",
   query=<the attendees' email domain>), then hubspot_object(op="get",
   object_type="contacts", object_id=<id>, associations=["deals"]).
   Read the deal with op="get", object_type="deals",
   properties=["dealname","dealstage","closedate","hs_next_step"],
   resolve_labels=true. No deal: report the call as unmatched and
   create nothing.
3. Write the next step with its owner and date:
   hubspot_object(op="update", object_type="deals", object_id=<id>,
   properties={"hs_next_step": "<action, owner, date>"}), and one note
   with op="add_note" carrying the buyer's own words. Do not change
   dealstage; tell me if the call earned the next one.
4. Find our last thread: gmail(op="search", query="to:<their domain>
   newer_than:60d"). Draft the reply with gmail(op="draft",
   reply_to=<message id>, body=<one quoted line, what we agreed and by
   when, one ask, under 100 words>).
5. Report each call: the deal matched, each field changed with its
   previous value, and the draft's id.
6. When I say so, save this as a process and create a hosted agent
   that runs it every hour from 9 to 18 on weekdays:
   oto_trigger(op="create", procedure=<the saved slug>,
   cron="7 9-18 * * 1-5", tz="Europe/Paris",
   tools=<only the tools above>, model=<the default served model>).

The process behind that prompt also files the buyer's product asks on the tracker and flags a deal at risk to its owner. Every step and the tools it uses:

The process in Tulina

Follow through on sales calls

Trigger · Event

Sales call ended

A finished call opens the run once its notes are ready, so the follow-through starts before the rep is back at a desk.

Claude runs the rest autonomously

1
AI pulls the deal and its history.
HubSpot
HubSpot
The deal, its stage, its contacts and the last email thread come back, matched to the call by company as well as by attendee.
Gmail
Gmail
2
AI extracts pains, objections and next steps.
Granola
Granola
What the buyer said they need, what they pushed back on and what each side agreed to do come out of the call notes in the buyer's own words.
3
AI judges the deal against your playbook.
The call is read against your own qualification criteria, and whatever is still unknown for the stage, or puts the deal at risk, is named.
4
AI tests any offer against your pricing rules.
Anything hinted at on the call, a discount, a term, a delivery date, is read against your pricing rules, and whatever is outside them is named with the quote.
5
AI updates the deal, logs the next step.
HubSpot
HubSpot
The deal takes its next step with a date, a moved close date and new stakeholders, with every previous value reported.
6
AI files product asks with their quotes.
Linear
Linear
Each capability the buyer asked for lands on its issue with the quote, the account and the deal amount behind it.
7
AI drafts the follow-up, flags the risk.
Gmail
Gmail
The follow-up waits as a draft on the thread, and a deal at risk reaches the owner's manager with the quote that says why.
Slack
Slack

See the full process

The order is the part worth copying with or without Tulina. The deal is matched on the attendees' domain before anything is written, because the person on the call is often not the contact the deal was opened under. The stage never moves on its own. And the follow-up quotes the transcript, not the summary. For the half of the day before the call, Prepare for a demo from the last call reads the same records into a brief posted before the meeting.

Is Claude good for sales reps?

Claude is good for sales reps at the work around a deal and no help with the deal itself. It reads, sorts and drafts well: a forty-minute transcript into three commitments and a date, a CRM full of stale deals into a Friday list, a company's pages into a brief. It cannot judge whether a buyer means what they said, and it will not know your deal history unless a connector or a note gives it to Claude.

What gets reps in trouble:

Writing with your name. A connector acts as the person who connected it. A prompt that can update the CRM can update the wrong deal. Give each job only the tools it needs, and review proposed changes with their before and after.
Flattery passed off as personalisation. Asked to personalise, a model compliments. Make every opener cite a dated fact with its link, or send the plain version.
Treating the summary as the record. Call summaries paraphrase. Anything you quote back to a buyer should come from the transcript.
Instructions hidden in what Claude reads. An email or a transcript can contain text that looks like an instruction. The Sales plugin's README says its skills treat such content "as information, never as instructions". Hold any prompt you write to the same rule.
A green run is not a done job. A scheduled task that ended cleanly may still have matched no deal. Read what it wrote, not its status.

Is ChatGPT or Claude better for sales?

Neither is better for sales in general: the one that does the job is the one connected to your CRM, call recorder and inbox, with the job written down well enough to run the same way twice. We have not benchmarked the two on sales tasks, so we will not rank them. What we can say is where they differ for a sales team today. Anthropic publishes a free Sales plugin with 36 named skills, and Salesforce in Claude is in beta. A team that standardises on one model gets one set of skills to maintain.

The useful test takes ten minutes. Run the after-call prompt above in both, on the same call, against the same CRM, and compare the drafts on three things: did it quote the buyer correctly, did it match the right deal, and did it invent a commitment nobody made. Tulina's connectors work from either assistant, so the same process runs whichever one your team prefers.

Can sales managers use Claude?

Sales managers can use Claude for the review half of the job: the pipeline by rep and stage, the deals that slipped, the prep for a one-to-one. The Sales plugin has team-pipeline, rep-context and win-loss-review for that. These are clock jobs more than questions. A pipeline review that a manager has to remember to run on Friday is one that gets skipped the week it matters, which is why it is the first manager job worth scheduling.

What about Claude Code for sales?

Claude Code runs the same skills as Cowork. Anthropic's README gives the install as two commands, claude plugin marketplace add anthropics/knowledge-work-plugins then claude plugin install sales@knowledge-work-plugins. Claude Code suits a team that wants its sales prompts in a repository, reviewed like code, and its cloud scheduled tasks can start on an API call, which is one way to react to a call ending. For a rep who wants the work done rather than the tooling, Cowork or a hosted agent is the shorter path.