Replacer Replacer
← Onboarding
1 Onboarding step

Export your LinkedIn archive

A 5-minute self-serve export gives Replacer the highest-quality voice signal it will ever see — your real published writing.

The single best source of training data for your voice is the writing you've already published — LinkedIn posts, articles, comments. Drive folders are noisy; a Google Doc named "agenda" isn't your voice. Your published feed is.

LinkedIn lets you download a complete archive of your own data in a few clicks. No third-party tool, no API permissions, no scraping. You stay in control.

Why this matters

Most "voice training" pipelines silently mix three categories of content together:

  1. Authored prose you actually wrote (blog posts, op-eds, LinkedIn updates)
  2. Work artifacts you saved (CSVs, agendas, contracts, handoff docs)
  3. Third-party content in your folders (recommendations, articles, attachments)

Only category 1 should train your voice. The other two pollute the signal — every CSV that gets treated as "Mike's writing" pulls the model toward something Mike never wrote. Replacer's authorship verifier filters most of this out automatically, but the cleanest input is the one we don't have to filter: your actual published feed.

Step 1 — Request your archive

  1. Visit linkedin.com/mypreferences/d/download-my-data (or Settings → Data Privacy → "Get a copy of your data")
  2. Select "The works: All of your individual data files" — or, if you want to be selective, just "Posts," "Articles," and "Comments."
  3. Click Request archive.

LinkedIn will email you a download link. Fast archives arrive in ~10 minutes; full ones can take 24 hours.

Step 2 — Upload to Replacer

When the email arrives:

  1. Download the ZIP file. Don't unzip it.
  2. Go to Settings → Data Sources in your Replacer dashboard.
  3. Click Upload LinkedIn archive and select the ZIP.

Replacer will extract Shares.csv, Articles.csv, and Comments.csv, run each entry through the authorship verifier, and pull the qualifying writing into your gold corpus. Reposts, polls, and one-line comments don't qualify; substantive posts and articles do.

What happens next

Within ~10 minutes you'll see your voice score rise as the new authored writing gets factored in. You'll also see a fresh row in the Voice Fidelity panel showing how much published-output gold you now have. More gold = more eval power = more confident proof that your trained voice beats a generic LLM.

Privacy note

The archive never leaves your account. We extract the text into your gold corpus and discard the original ZIP. We don't share, sell, or train shared models on your writing — your voice profile is yours.