Every LinkedIn workflow here, and the three disciplines that have none
Four sequences are published and three disciplines are empty. A workflow here is the whole job in order — 22 steps across the four, each naming what it takes in and what it hands the next one, with the LinkedIn plan it leans on and what that plan charges this month. Sales, recruiting, content and job search have one. Networking, lead generation and personal branding do not, and they are printed empty rather than filled, because a sequence nobody has run is the thing a library like this exists to be an alternative to.
A step that cannot name what it hands over is a diagram, not a step
Six things are published for each sequence below, and the list is short on purpose: every one of them is checkable by opening the workflow rather than by trusting the page. Where a figure has not been measured, that is said instead of estimated.
Read the In and Out lines down a sequence. Where the Out of one does not appear as the In of the next, the workflow has a gap. That is the test, and it is why these are written as pairs rather than drawn as a flowchart.
Three of the four lean on a paid LinkedIn seat at some step. The figure beside each is the US list price read from LinkedIn's own purchase flow, with the day it was read, and it moves on this site the day LinkedIn moves it.
21 of the 22 steps name no model at all. The exception is a Project instruction, which is a different object from a pasted prompt and is filed as one rather than split into a Claude version and a ChatGPT version that differ in the first line.
None of the four calls a server or needs a vendor signup, which is also why none of them can get a LinkedIn account restricted. 13 MCP servers and 42 third-party tools exist for the same work; what they change is who holds your session.
The steps that produce a message to a named person hand it back for you to read and paste. That is the approval point, and it is not something to switch on — there is no automation in these sequences to switch it off.
Each sequence says what to look at weekly and what a bad number should make you change. None of them publishes a benchmark, because none has been measured against a held-out set of real prospects.
Not eight prompts — one thread
Each step's output is the next step's input, pasted straight in. Here is the start of the sales sequence shown that way, which is how it reads in practice rather than how it reads as a list.
Your last 10–20 closed deals, won and lost
An ICP in LinkedIn’s own vocabulary, plus one observable trigger
The ICP’s search-ready summary
A filter set and Boolean strings returning 200–1,500 results
The export from that search
Three tiers, capped at what your weekly invitation budget can actually cover
Four disciplines have a sequence. Three have prompts and no sequence.
The rule this table is built on is that an empty row stays empty. A directory padded to a round number cannot answer the one question a reader has on arrival, which is whether to keep looking; a table that prints its own holes answers it in a screen. The three empty rows each say what exists instead, specifically.
| Discipline | What you arrive with | Published workflow |
|---|---|---|
| Sales | “I need thirty people worth talking to and some of them on a call.” | From ICP to booked meeting — 8 steps |
| Recruiting | “I have an open role and a fixed number of InMail credits this month.” | From open role to a shortlist worth defending — 4 steps |
| Content | “I want to post consistently without inventing statistics.” | From raw material to a posting habit — 6 steps |
| Job search | “I am applying steadily and hearing nothing back.” | From applying into the void to conversations — 4 steps |
| Networking | “I want to open a conversation that is not a sale.” | None published. The prompt library has nothing for it either — every networking job on the map there is empty. The one piece that transfers unchanged is the comment strategy: being visible under someone’s posts before you message them is the same move whether the objective is a deal or a conversation. |
| Lead generation | “I need a list of the right companies and the right people in them.” | None published. No separate workflow. Steps 1 to 3 of the sales sequence are the list-building half — define, search, cut — and they run on their own without the five steps after them. Two prompts sit outside any sequence: the buying-signal analyser and the account research brief. |
| Personal branding | “I want the profile and the posting to argue the same thing.” | None published. No sequence connects the two halves. Step 1 of the content workflow settles the claim you want to be known for; the profile prompts rewrite the page that claim lands on. Nobody has written the workflow that makes one feed the other, so it is not here. |
The three empty rows are not a roadmap and nothing here promises to fill them. Two of them have prompts that do part of the job, which is why the row names those rather than leaving the cell blank — the map of which prompt does which job, including the jobs nothing covers, is on the prompt library.
Three of the four assume a paid LinkedIn seat. The assistant is the free part.
Read the first two rows before the rest. The cost of one of these is not the assistant — every prompt in all four is free and needs no account — it is the LinkedIn seat behind one or two of the steps, and the seat is the thing that decides whether the sequence finishes. The bottom two rows are the ones that keep this section honest.
| From ICP to booked meetingSales — 8 steps | From raw material to a posting habitContent — 6 steps | From open role to a shortlist worth defendingRecruiting — 4 steps | From applying into the void to conversationsJob search — 4 steps | |
|---|---|---|---|---|
| LinkedIn plan it assumes | Sales Navigator Core | None — free account | Recruiter Lite | Premium Career (optional) |
| What that costs | $119.99 / month | Nothing | $170 / month | $39.99 / month |
| Results a search returns | 2,500 per search | No search step | 1,000 per search | 1,000 per search |
| InMail credits in that seat | 50 a month | No seat | 30 a month | 5 a month |
| Assistant | Any assistant | Any, except one step — Claude or ChatGPT | Any assistant | Any assistant |
| Hardest step in it | Working | Advanced | Advanced | Starter |
| Steps you set up once | 2 of 8 | 3 of 6 | 1 of 4 | 1 of 4 |
| Needs an MCP server or a third-party tool | No | No | No | No |
| Acts on LinkedIn on your behalf | No | No | No | No |
US list prices in USD, excluding tax. LinkedIn bills in local currency in most markets and the converted figure is rarely a straight exchange-rate conversion, so treat this as the US baseline and confirm in your own billing page before committing to an annual term. Sales Navigator prices were last read from LinkedIn on 12 September 2026, Recruiter on 10 September 2026 and Premium on 10 September 2026. No end-to-end time is published for any of the four: the per-step times are the ones that were written down, they are printed beside each step below, and a total assembled from them would be a figure nobody has run a stopwatch on.
From ICP to booked meeting
“I need to find thirty people worth talking to and get some of them on a call.” The complete outbound workflow. Each step takes the previous step’s output as its input, so nothing has to be re-explained and nothing gets lost between tools. Run top to bottom the first time; after that you live in steps 4–7.
Define who20 min, once a quarter
- In
- Your last 10–20 closed deals, won and lost
- Out
- An ICP in LinkedIn’s own vocabulary, plus one observable trigger
Find them10 min, then 3 min to build
- In
- The ICP’s search-ready summary
- Out
- A filter set and Boolean strings returning 200–1,500 results
Cut the list5 min per few hundred rows
- In
- The export from that search
- Out
- Three tiers, capped at what your weekly invitation budget can actually cover
Understand the top tier3 min each
- In
- One prospect’s pasted profile and activity
- Out
- A brief ending in a value hypothesis and an opening line
Write the first touch2 min each
- In
- The brief from step 4
- Out
- A message at the right length, built on a stated hypothesis
Plan the cadence15 min, once
- In
- Your seat count and plan
- Out
- Three to four touches inside the platform ceilings, with a stop rule
Handle what comes back2 min each
- In
- The thread, with dates
- Out
- A classified reply and three calibrated responses
Prepare the call2 min
- In
- The thread and the profile
- Out
- A one-page brief and three questions, one of them disqualifying
Step 2 builds a Sales Navigator filter set. Without a seat that step becomes a regular LinkedIn search, which stops at 1,000 results rather than 2,500, and the Boolean it writes has fewer fields to write against. The other seven steps need no paid plan at all.
Steps 5 and 7 produce text addressed to a named person. Nothing sends it. The message sits in the assistant until you read it and paste it into LinkedIn yourself, which is the only approval point there is and also the reason there is nothing to switch off.
The invitation ceiling, at roughly 100 a week on every plan including the most expensive one. It sits at the bottom of the funnel and therefore sizes everything above it, which is why step 3 caps the list before step 4 spends three minutes researching a person you were never going to reach this month.
Acceptance rate on the invitations sent each week, and replies per ten first messages. A falling acceptance rate is a step 1 problem, not a step 5 one — and LinkedIn lowers the weekly ceiling for accounts whose invitations are not accepted, so it compounds. Replies without acceptances means the list is right and the message is not.
Steps 3 and 6 both size themselves against the weekly invitation budget. This is where that figure comes from.
Step 2 assumes Sales Navigator. Current prices, dated, if you are deciding whether to buy it.
Price read from LinkedIn on 12 September 2026. The full sequence, with its artefact list and its own questions: From ICP to booked meeting.
From raw material to a posting habit
“I want to post consistently about something worth being known for, without inventing statistics.” Content fails at the material stage, not the writing stage. This workflow starts with the position, then builds the capture habit that feeds it, and only then writes — which is why it produces posts in week six as well as week one.
Decide the claim30 min, once a year
- In
- Your last 5–10 posts and what you sell
- Out
- One defensible claim, three themes, and the not-list
Set up capture20 min, once
- In
- The claim and themes
- Out
- CONTENT-ENGINE.md, naming where material comes from and who logs it
Set up the assistant10 min, once
- In
- BRAND-VOICE.md and the claim
- Out
- A Project that drafts in your voice and refuses to invent facts
Draft from material5 min per post
- In
- A call, a result, a mistake, an analysis
- Out
- A post, two alternative openings, and the pinned comment
Get more from one source15 min per source
- In
- A long piece: webinar, report, transcript
- Out
- The posts the source honestly supports — an honest count, not ten
Be visible without reach10 min daily
- In
- Posts from the accounts your audience reads
- Out
- Comments that earn a reply from the author
No paid plan. None of the six steps touches a search cap, an InMail allowance or anything else a subscription lifts — posting, commenting and reading your own material are free-account work, and a Premium badge changes none of it.
Step 4 hands back a draft, two alternative openings and a pinned comment, and where the material does not support a claim it says the material is too thin to post rather than filling the gap. That verdict is the approval point: the prompt is built to refuse, and the refusal is only useful if you read it.
Nothing on LinkedIn’s side. The constraint is material, which is why step 2 exists at all — a content workflow with no capture habit produces posts in week one and silence in week six, and the drop is the commonest way this one fails.
Posts published against posts drafted, monthly, and replies from people you have never met. Drafts far ahead of posts means step 1 never settled the claim, so every draft gets argued with. Posts far ahead of replies means the material is real and the opening is not — step 4 returns two alternatives for exactly that.
Every figure sourced, every limitation stated in the piece rather than in the replies.
The full sequence, with its artefact list and its own questions: From raw material to a posting habit.
From open role to a shortlist worth defending
“I have an open role and thirty InMail credits this month.” Four steps, run in order, with the credit budget treated as a first-class constraint rather than something you discover in week three.
Decide where to look20 min, once per role
- In
- The job description and the geography
- Out
- Source archetypes, the titles the role is called at each, and a pool verdict
Build the searches5 min each
- In
- One archetype at a time
- Out
- A tight and a broad Boolean string per archetype
Shortlist with evidence10 min per batch
- In
- The job description and 20–40 pasted profiles
- Out
- A ranked shortlist with evidence per requirement and a screening question each
Reach out3 min each
- In
- One shortlisted candidate
- Out
- Subject line, message, and the reply for "not looking"
The whole sequence is sized against one seat: 30 InMail credits a month and a 1,000-result search ceiling. Those two numbers are what steps 2 and 3 are narrowing towards, and on a plan with different ones the shortlist length changes.
Step 4 drafts the InMail. A credit is spent when you send it, not when the assistant writes it, and a credit is returned only if the candidate replies inside 90 days — so the read before sending is the step with the money in it.
30 credits a month, and the 1,000 results a Recruiter Lite search returns however you paginate it. Credits accrue to at most three months’ worth and then stop, so an unused month is not banked indefinitely.
Reply rate per ten InMails, and credits left on the day the shortlist is agreed. Credits gone before the shortlist is agreed means step 3 shortlisted too loosely; the evidence column is where to look. Thin replies while the credits pile up usually means step 1 mapped the wrong source companies.
The two numbers this whole workflow is sized against.
Price read from LinkedIn on 10 September 2026. The full sequence, with its artefact list and its own questions: From open role to a shortlist worth defending.
From applying into the void to conversations
“I am applying steadily and hearing nothing back.” Applications through the job board are the lowest-yield channel on LinkedIn and where most search time goes. This workflow reallocates it, and fixes the profile first because the profile is the landing page for everything else.
Write the case once30 min, once
- In
- Your CV and the honest constraints
- Out
- JOB-SEARCH.md — one story every later draft inherits
Fix the landing page20 min
- In
- Your current profile text and one goal
- Out
- A ranked change list with the top three sections rewritten
Build the target list20 min
- In
- What you want next and what you have done
- Out
- Twenty companies in three tiers, the role to contact at each, and a weekly routine
Write to a human3 min each
- In
- One role and one named person
- Out
- Three messages: hiring manager, recruiter, referrer
Optional, and the workflow says which part of the tier it is buying: 5 InMail credits a month, for the hiring managers you cannot reach any other way. Without it the sequence routes through connection requests instead, which the weekly ceiling covers many times over for one person’s search.
Step 4 writes three messages to three named people — the hiring manager, the recruiter and the person who could refer you. All three come back as drafts, and the one you would be most embarrassed to send unread is usually the referral.
The 5 credits, if you buy them. Otherwise nothing platform-side binds a single person’s search: one job seeker does not approach the ~100 invitations a week that constrains an SDR.
Replies per ten messages sent to a named human, against replies per ten applications through the job board. The comparison is the point. If the direct channel is not several times better, step 2 has not fixed the profile that every one of those messages sends people to.
Five InMail credits a month is the feature that matters here. The rest of the tier mostly is not.
Price read from LinkedIn on 10 September 2026. The full sequence, with its artefact list and its own questions: From applying into the void to conversations.
These are prompts. The ceiling they run into belongs to the account, not the assistant.
None of the four reaches LinkedIn. They produce text, and a person carries it across. So the limits that bind them are the member-side ceilings, which are the same on every plan including the most expensive one, and which no subscription and no software lifts. The third column is the step each one binds first.
| Ceiling | Value | Which step it binds |
|---|---|---|
| Connection invitations | ~100 per week Applies on every plan including Recruiter Corporate. LinkedIn adjusts it per account based on acceptance rate, so a low-acceptance account gets less, never more. | Sales, step 3. The qualifier cuts the list to what a week of invitations can cover, so the research in step 4 is not spent on people you cannot reach until next month. |
| InMail credit expiry | 90 days Unused credits expire. They accrue to at most three times the monthly allowance, and a credit is returned if the recipient replies within 90 days. | Recruiting, step 4. A credit comes back if the candidate replies inside the window, so the allowance rewards reply rate rather than send volume. |
| Commercial use limit | Monthly search cap Free accounts hit a monthly ceiling on profile searches. Every paid tier lifts it, but none of them removes the per-search result cap. | Lead generation, before step 2. It is the reason the search step is the one that assumes a seat: a free account runs out of profile views partway through a month of list-building, and finds out by being shown an upsell instead of a result. |
| Profiles per search | 1,000 / 2,500 Regular LinkedIn search stops at 1,000 results; Sales Navigator stops at 2,500, paginated 25 at a time. Both are per search, not per day. | Sales, step 2. The filter set is written to land inside the cap rather than to return everything, because results past the cap are not paginated to — they are absent. |
LinkedIn publishes neither the invitation ceiling nor the commercial use limit as an exact number, and both flex with account age and behaviour — these are the ceilings the platform enforces in practice, not contractual values. The full table, per plan, is on the limits page.
Section 8.2 bars developing, supporting or using software that scrapes LinkedIn or copies profiles and data, and bars bots and other automated methods of access. It does not distinguish between a slow tool and a fast one, and it carves out nothing for well-paced automation. Nothing in these four sequences automates anything, so the clause does not reach them. It reaches the moment the sending is handed over, which is a decision worth making deliberately rather than discovering. linkedin.com/legal/user-agreement ↗
The Prohibited Software and Extensions policy names browser extensions and third-party automation services and states that LinkedIn may restrict or permanently restrict accounts that use them. The vendor keeps its subscription revenue either way. 18 of the 42 tools in the directory act under your own name and 12 hold your session on their own machines. linkedin.com/help ↗
Four workflows, and no target published for any of them
The convention in this category is to stamp a reply rate on a prompt and move on. We have not benchmarked these against a held-out set of real prospects, so what is published instead is the number worth watching and what a bad one should make you change — which is the part a benchmark would not have told you anyway.
| Workflow | Measure weekly | What a bad number means |
|---|---|---|
| From ICP to booked meeting | Acceptance rate on the invitations sent each week, and replies per ten first messages. | A falling acceptance rate is a step 1 problem, not a step 5 one — and LinkedIn lowers the weekly ceiling for accounts whose invitations are not accepted, so it compounds. Replies without acceptances means the list is right and the message is not. |
| From raw material to a posting habit | Posts published against posts drafted, monthly, and replies from people you have never met. | Drafts far ahead of posts means step 1 never settled the claim, so every draft gets argued with. Posts far ahead of replies means the material is real and the opening is not — step 4 returns two alternatives for exactly that. |
| From open role to a shortlist worth defending | Reply rate per ten InMails, and credits left on the day the shortlist is agreed. | Credits gone before the shortlist is agreed means step 3 shortlisted too loosely; the evidence column is where to look. Thin replies while the credits pile up usually means step 1 mapped the wrong source companies. |
| From applying into the void to conversations | Replies per ten messages sent to a named human, against replies per ten applications through the job board. | The comparison is the point. If the direct channel is not several times better, step 2 has not fixed the profile that every one of those messages sends people to. |
Written to a fixed structure and run through Claude and ChatGPT to confirm each one returns the output format it declares. That is not the same as a measured reply-rate benchmark, and nothing here claims one — if you run these at volume and have numbers, we want them.
The questions people arrive with, answered directly
What is a LinkedIn AI workflow?
A named sequence of steps where each step’s output is the next step’s input, rather than a folder of prompts you assemble yourself. Four are published here, 22 steps in total, 20 of them backed by a prompt and 2 by a context file you attach once. The test of whether one is real is printed on the page: read the In and Out lines down a sequence, and where the Out of one does not appear as the In of the next, it is a diagram rather than a workflow.
Which workflow should I start with?
The one matching the number you have to hit. If that is meetings, start with the sales sequence and run it top to bottom once — the first pass produces the ICP and the cadence that the later passes reuse, so it is slow once and fast afterwards. If you are hiring, start with the recruiting one, because it is sized against a credit allowance that a false start burns through.
Do I need Sales Navigator to run the outbound workflow?
For one step of the eight. Step 2 writes a Sales Navigator filter set, and a Core seat is $119.99 a month, read from LinkedIn on 12 September 2026. Without it that step becomes a regular LinkedIn search, which stops at 1,000 results rather than 2,500 and has fewer fields to filter on. The other seven steps run unchanged on a free account.
Can an AI agent run one of these end to end for me?
Not as written, and the reason is worth knowing before you buy something that says it can. Every step here ends in text you read; nothing in the four touches LinkedIn. The moment the sending is handed to software, the arrangement changes — Section 8.2 of LinkedIn’s User Agreement bars automated access to the service by any means, and enforcement lands on the member account rather than on the vendor. 18 of the 42 third-party tools we have records for act under your own name and 12 hold your session on their own machines. What each one holds is on the tool directory.
Why is there no networking workflow?
Because nobody has written one, and printing an empty row is more useful than filling it. Networking failed the same test in the prompt library: every job under it — introductions, event follow-up, investor outreach, partnership outreach — is empty there too. The material input a networking message needs is a conversation you already had, and that is the one input none of these prompts asks for.
Do these work with ChatGPT, or only with Claude?
Both, and Gemini. 21 of the 22 steps name no model at all, because a prompt with a role, constraints and an output schema is not model-specific and splitting one into a Claude version and a ChatGPT version that differ in the first line is how a library ends up with two thin pages for one job. The exception is the ghostwriter step in the content workflow, which is a Project instruction rather than a pasted prompt — a different object, and it names Claude and ChatGPT because those are the two with the feature.
How long does a full pass take?
Every step publishes its own time and they are printed next to the steps, so the arithmetic is yours to do rather than ours to claim. What is worth reading instead is the split: the sales sequence sets up 2 of its 8 steps once, the content one 3 of 6. The one-off steps are where the hour goes, and they are also the ones people skip.
Have these been tested?
Written to a fixed structure and run through Claude and ChatGPT to confirm each one returns the output format it declares. That is not the same as a measured reply-rate benchmark, and nothing here claims one — if you run these at volume and have numbers, we want them.
Start with the sequence for the number you have to hit
Four workflows, 22 steps, each one naming what it hands the next and what the seat behind it costs. Three disciplines have none, and they say so.