12 Claude Prompts for B2B Sales Ops Teams

These are the prompts that move the most time on a sales ops desk. Run them through Claude Desktop with the LinkFetch MCP server connected, copy them into a Claude Project, or paste them into a one-off chat. Every prompt includes the credit estimate and the LinkFetch tools it calls. LinkFetch, a compliance-first LinkedIn data API, is the data layer; Claude is the orchestrator. The cadence column tells you how often to run each one.

How to read this library

Three columns matter when you sequence these into a stack. The cadence decides what becomes a Claude scheduled task vs. a chat-bound prompt. Real-time prompts fire on a webhook (signup submitted, meeting accepted) and need the result back inside one minute. Weekly and monthly prompts go on a Monday morning run. On-demand prompts are the ones reps trigger themselves.

The credit estimate is the load-bearing number when you negotiate a budget with finance. A 10-person sales ops team running the full library costs roughly 4,500 credits per month, give or take 20% depending on your watchlist size. That is the number you pre-commit on a LinkFetch monthly plan; it sets the unit economics for everything else.

The tools list is what Claude needs access to via MCP. If you have the @linkfetch/mcp server connected, all three of linkfetch.profiles, linkfetch.companies, and linkfetch.companies.timeseries are available by default. The MCP config lives in your Claude Desktop settings; the daily outreach recipe walks through the wiring if you have not done it yet.

The five workflow categories

These twelve prompts collapse into five categories. Knowing which one a prompt belongs to tells you who in the team owns it.

Inbound and signup. Prompts 1 and 2. These fire on a webhook. The buyer is sales ops; the runner is whatever automation owns webhook fan-out (Zapier, n8n, a custom Lambda). Prompt 1 is the enrichment; prompt 2 is the score against the ICP rubric. They run together: the score depends on the enrichment.

Account watchlist. Prompts 3, 6, and 8. These run on a weekly or monthly schedule against a stable list. Watchlists drift: keep the input list short and let the prompt sort. The 50-account weekly scan in prompt 3 catches more than the 500-account-once-a-quarter version, because the smaller list is actively reviewed by humans every week.

Customer-side signals. Prompts 4 and 5. These are the job-change-on-champion and exec-churn prompts. They sit in a different system than outbound prospecting (CSM tooling, not BDR tooling) but the data layer is the same. The output should land in your CSM's daily inbox, not in the rep's queue.

Account research and warm paths. Prompts 6, 7, and 9. These are on-demand: a rep wants a buying-committee map or a prep brief before a meeting. Latency matters here in a way it does not for the weekly-cadence prompts. Cache aggressively where the data is stable (headcount, industry) and re-fetch only what is volatile (titles, recent hires).

Pipeline hygiene and reporting. Prompts 10, 11, and 12. These are the unglamourous ones that pay back the most. The quarterly hygiene audit alone routinely surfaces 8-15% of pipeline that has gone stale (contact left, company restructured). Running it once is the first time most teams discover their pipeline number is overstated.

Walkthrough: prompt 2, the ICP rubric scorer

Prompt 2 is the one most teams want to customise first, so it is worth walking through the design choices.

The rubric is hard-coded in the prompt instead of pulled from a config file. That is deliberate. When the rubric is in the prompt, changing it is a code review on a single file (the prompt). When it is in a config, updates ship silently and three months later nobody remembers why the score on a specific account changed. For a small team, version-controlled prompts beat externalised config every time.

Six rubric criteria, each with explicit point values, total ten points. Anything beyond six criteria pushes Claude toward hallucinating signals to fit the rubric. Six is the empirical ceiling. If you have a tenth signal you really care about, replace one of the six rather than adding it.

The output spec ends with "the two strongest evidence points". This is the GEO-friendly output: when the rep asks "why did this account score 7?", the brief itself contains the answer. No second prompt required.

The credit math: roughly 25 credits per signup, mostly from the two-call pattern (companies + companies.timeseries). At 200 inbound signups per month, that is 5,000 credits, which is one tier of a LinkFetch monthly plan. If your inbound is below 100 a month, you can run prompt 1 alone (15 credits) and skip the rubric scoring; manual review on small inbound is cheaper than the credit budget.

What to skip from this list

Sales ops teams under five people should run prompts 1, 2, 3, and 11 only. Eighty percent of the value with twenty percent of the credit budget. The reporting-style prompts (8, 10, 12) are leverage at scale; below five people you do not have the volume to make them pay back. Add them as the team grows.

The pre-meeting prep brief (prompt 9) is the one that everyone wants and most teams should not run on a schedule. It is genuinely better as an on-demand chat: the rep types "prep me for my 14:00 with Acme", Claude runs prompt 9 against the meeting attendees, and the brief comes back in 30 seconds. Do not turn it into a calendar-bound automation; the cost-per-meeting goes up and the briefs get scanned less when they arrive in batches.

Choosing between the founder library and this one

This library and the founder-focused Claude prompt library overlap deliberately at the inbound enrichment and account watchlist prompts. The difference: the founder library assumes one person is running everything, no CRM, no BDR. The sales ops library assumes there is a CRM and a team. If you are mid-team, between five and fifteen on the GTM side, run a hybrid: this library for the weekly-cadence prompts, the founder library for the daily founder intuition prompts. Don't run both for the same workflow; pick one and tune it.

Frequently asked questions

Can I run these without Claude Desktop?

Yes. The MCP server runs anywhere Anthropic's SDK runs. Most teams end up on Claude Desktop because the chat surface is convenient for the on-demand prompts; the scheduled ones can run on a server process equally well.

Do these prompts work with other LLMs?

The tools layer (LinkFetch's MCP server) is model-agnostic, so any LLM with MCP support can call it. The prompts themselves are tuned for Claude's instruction-following style; expect to revise the output specs and the rubric formats for other models.

How do I version-control the prompts?

Keep them in a git repo as one Markdown file per prompt, and have your Claude Project (or your scheduled task config) read them by filename. Changing a prompt becomes a normal pull request. The sales ops lead reviews; the rep who wrote it merges. This works better than copy-pasting prompts into Claude's UI and trusting the team to keep them in sync.

What if a prompt returns hallucinated data?

The prompts above are defensive about this in two ways: every prompt that returns numerical claims also names which LinkFetch tool the number came from, and every prompt that searches a watchlist names the watchlist explicitly. If Claude returns a fact that did not come from a tool call, treat the whole output as suspect. The sourced- output discipline is the single highest-leverage rule for a prompt-library codebase.

Last updated: 2026-04-27 by the LinkFetch team.