AI for LinkedIn Webinar March 6th

This session was a full hour on AI automation for LinkedIn: managing your inbox with Claude Chrome, extracting leads from other people's posts with zero risk to your account, and building a content analytics system on top of everything you have ever published. If you do outbound or content on LinkedIn and you worry about its strict terms and conditions around automation, this recap is for you. The whole approach is either manual-speed automation or fully cookieless scraping, and I am explicit about where the residual risk sits.

One Main Tool for the Quick Wins: Claude Chrome (3:26)

For the first time in these webinars I built the quick wins around a single tool, the Claude extension for Chrome, which I just call Claude Chrome. It is great at exactly one category of work: manual tasks that do not need speed or volume, on platforms that either have no API or have strict terms around automation. LinkedIn is the textbook case. Anything related to messaging and inbox management works fantastically well with it, and the same trick carries over to Instagram and Facebook.

For the advanced builds later in the session I added Apify and Unipile, driven from Claude Code. Antigravity works as an equivalent to Claude Code for everything shown here, and Nano Banana Pro came up on the content side.

Quick Win: Inbox Management (6:06)

The live prompt was embarrassingly simple. I told Claude Chrome to go into my messaging, extract the last ten messages, then go into a spreadsheet and log every message I received with the name of the person. It works across tabs: one tab is LinkedIn messaging, the other is the spreadsheet, and it walks message by message and logs the interactions. When it finished (17:31), I had everyone I messaged that day, what the replies were, all logged without me touching anything.

The same pattern handles enrichment (8:38). The week before, we ran a webinar where we never asked people for their LinkedIn profile or company. Claude Chrome went through every signup, name by name, validated the company, and found each person's profile. That is how we enriched a database of 125 people using only this extension. It is slow by design, but you can fire off several instances at once, each working on a different database.

Teach Mode, Tasks, and the Risk Question (10:00)

Someone asked whether Claude needs a special skill to understand LinkedIn. It does not. What it does have is the Teach Claude button: it screen-records you doing the task manually, Claude analyzes the video, and turns it into a task, which is effectively the skill for this version of Claude. Tasks can also be scheduled (13:02). Type the slash command, pick schedule a task, give it the prompt and the page it should start browsing from. I have one that likes people's comments on my profile at ten in the morning, every morning, and it requires nothing from me.

On the inevitable risk question (11:26): there is always some risk, everything in this life is a gamble, but it is very hard for LinkedIn to track this kind of usage. My personal rules are to automate only what I could do manually at manual speed, to automate data collection, and to never automate LinkedIn messaging. You can also ask Claude Chrome or Claude Code to insert random time intervals between actions when sequencing through many items, which reduces the risk further. Some use cases can be made literally zero risk, which is where Apify comes in.

On pricing (11:02): I am on the roughly 90 euros per month plan, and it is the best hundred euros or dollars you will spend for the near future.

Unipile: Your LinkedIn Account as an API (16:17)

Unipile is a connector. You plug in accounts you own, LinkedIn, WhatsApp, Instagram, email, and drive them through a simple API key. In the live demo (18:32) I gave Claude Code a profile URL from someone on the call and told it to use Unipile to send him a message saying hi. After sorting out a settings hiccup from a recent account change, the message went out (24:05). Once connected, you can programmatically read your messages, your posts, your interactions, and even your network.

Unipile costs 55 dollars per month. If you do not have the budget for a tool like PhantomBuster at 75 to 80 dollars per month, or you do not need all its features, this is a clean way around it. Be clear-eyed though: it is a third party connected to your account, so it is not zero risk. It is a big company here in France and the approach keeps working.

A Living Database of Every Post You Have Written (20:48)

My content management system for YouTube and LinkedIn lives in Notion, fed by Unipile. Every post I have ever created is in there with its engagement rate, impressions, reactions, the post image, and the link. One prompt syncs it. During the session a fresh post got added to the database while we were talking.

The point is not the archive, it is the analysis (21:37). Once the sync completed I asked Claude Code to benchmark my content: what are my best posts, and what do they have in common, the theme, the hook, or the visual? If you want to push this further (23:35), snapshot the metrics daily so you can compute the growth rate per post instead of a single static number. You can apply the same system to other creators, scrape their best performing content, and study how to adapt it for yourself.

Zero-Risk Lead Extraction with Apify (25:00)

Apify is a library of scrapers. The distinction that matters is cookieless versus cookie scrapers (25:42). A cookieless scraper never touches your account. Someone else's account does the scraping, so your risk is zero. The playbook is to combine scrapers on a post: comments plus reactions plus reposts, because those are the people who interacted with a lead magnet. Then a no-cookies profile scraper pulls the full profile data so you can qualify each person. Your outreach practically writes itself: I saw you commented on this precise post two weeks ago, how did you implement this in your company?

Live Build: From One Post to Qualified Leads (29:59)

I picked one of Nick's posts with a healthy amount of interactions, asked Claude Code which Apify actors it had in its tools folder, and had it pull all the comments from the post. Next instruction: combine this with the profile scraper and scrape all of those people's profiles. From there, qualification is one more prompt, for example which of these people are CEOs of companies located in Europe (35:45). At about three dollars per thousand profiles, Apify can get expensive if you are careless, but for what it returns the pricing is honestly quite good (36:02).

The Parasite System and the Profile Viewer Demo (34:00)

Why stop at one post? Expand the build: give the system one creator's profile, scrape every post from their last 10, 15, or 30 days, and run the reactions, reposts, and comments scrapers on each one. Then flag the people who interact with that content frequently. Instead of one intent signal you get many, and your message targeting improves accordingly. That is the parasite lead system, and it is what my LinkedIn Observer application is built on.

The final demo (41:01) was the profile viewer workflow, built with the Playwright MCP, an MCP server that gives Claude Code or Antigravity control of the browser. It scrolls through everyone who viewed my LinkedIn profile, collects the profiles, and serves a dashboard showing who looked at your profile and when, with optional enrichment for phone and email. If you have five to ten thousand followers and a hundred or two hundred profile visits a week, that list alone is immense value.

I also flagged PhantomBuster (37:46) for the asynchronous side: triggers when someone accepts your request, likes a post, or follows you back. Everything it does can be rebuilt on Unipile, but you would have to develop it yourself.

Two Tools, Two Machines (46:41)

The closing blueprint needs exactly two tools: Apify, plus Claude Code or Antigravity.

The lead machine: use an Apify LinkedIn search scraper to find lead magnet posts with very high engagement in your niche, scrape the engagement (reactions plus comments plus shares), scrape the profiles and companies behind it, and have Claude Code qualify them. End to end, it costs about 8 to 10 dollars per thousand leads, and these are people who already showed intent for something adjacent to what you sell.

The content machine: same skeleton, but instead of scraping engagement you collect the highest quality lead magnets into a database. That tells you what performs in your niche, and Nano Banana Pro covers the visual side of the posts. One caution from me: I am against generating content from the ground up with AI. There is a lot of value in a human voice right now, and you are better off improving your own tone than automating someone else's.

Q&A Highlights

The cost question (53:36): for the lead magnet flow, roughly three dollars per thousand profiles plus three to five dollars for the reactions and comments, so around eight dollars per thousand intent-rich leads. Not spray and pray, these people asked for the thing someone else was selling.

On what happens after scraping (52:13): you can hand the resulting database to PhantomBuster and it will send the messages asynchronously, meaning without you pushing a button. The real work is deciding your precise workflow first, then picking the tools.

On bringing this into companies (54:22): I see AI adoption at two levels. Level one is a separate AI infrastructure with one person running it for everyone else. Level two is everyone in the organization working directly with tools like Claude Code and Antigravity. Smaller teams do better at level two, larger organizations usually start at level one.

And for the Facebook question (56:07): automated comment replies are a perfect Claude Chrome use case. Show it how you respond once, then set up a task. No other tool needed.

If this recap was useful, the full recording and every other session are on the webinars page.