Why your LinkedIn connections export goes stale so fast

What is actually in the file

When you request your data from LinkedIn and pick connections, you get a file called Connections.csv. Each row is one connection and holds a handful of columns: first name, last name, profile URL, email address, company, position, and the date you connected.

That is a genuinely useful list. It is the only place many people have every professional they have ever accepted, and the profile URL is a stable identifier that does not change when someone moves jobs. If you want to answer a single question, such as who do I know at Stripe, it is a fine place to start.

The trouble is that people treat it as a database, and it is a snapshot.

The three gaps

Email addresses. LinkedIn includes a connection's email only if that person has allowed connections to see it. Most have not, so most rows come back with that column empty. You cannot send an introduction, a catch-up or a follow-up to a name. Anything you want to do with the list beyond reading it needs addresses from elsewhere.

History. Nothing in the file says how well you know anyone. A former cofounder and a stranger who sent a request after a webinar look identical. The one fact that decides whether to ask someone a favour, when you last spoke and how often, lives in your inbox and calendar, not in LinkedIn.

Time. The company and title are whatever LinkedIn showed on the day you downloaded. From the next morning the file starts to drift. People change jobs and keep their old title in your copy. You accept new connections and they are not in it.

How fast it drifts

People change jobs, get promoted and start companies all the time, and in a list of hundreds or thousands of connections some of that is happening every week. So a file that was accurate when you downloaded it gets less accurate every month, and you cannot tell which rows are wrong by looking at it.

That last part is the real cost. A stale row looks exactly like a current one. An assistant reading the file will happily tell you someone works at a company they left a year ago, and suggest an introduction on that basis.

Downloading again does not fix it on its own

The obvious answer is to download a fresh copy every month, and it helps only halfway. Now you have two files that disagree, and nothing tells you how.

To get value from a second export you have to compare it with the first. Match rows by profile URL, which is the one column that stays stable. Then look at what changed: new rows are new connections, a different company is a job change, a different title at the same company is a promotion. That comparison is where the useful information lives. A job change is one of the best reasons to get back in touch, and what to say when someone changes jobs covers how.

The comparison also has to be joined to everything else you know about the person, their email address from your inbox and the meetings in your calendar. Otherwise the job change is a fact about a stranger.

What to do with it instead

Use the export for what it is good at and keep it out of the role it is bad at.

• For a one-off question, hand the file to an assistant and ask. Finding warm introductions with a LinkedIn export shows how far that goes.
• For a network you use every week, treat the export as one input among several. Import it into something that merges it with your email, calendar and contacts, matches rows by profile link, and records what changed between imports.

NetworkOS works that way. Each LinkedIn import, from the CSV, the full data archive or the Chrome extension (which reads your connections list in your own browser at a slow pace; the export is the option with no automation at all), is merged with Gmail, Calendar and Google Contacts, and NetworkOS remembers the company and title LinkedIn last reported for each connection. The next import is compared against it, so a move is recorded in the contact's role history, and the people you are closest to who changed jobs come back as suggestions with a congratulations already drafted. The limit is that detection only sees what LinkedIn reports, so you still have to re-import every month or so; nothing watches LinkedIn between imports.

The wider point applies whatever tool you use. A personal agent with only the export is reasoning about a photograph. Why that matters, and what the agent needs instead, is in can a personal AI agent run your CRM on its own.