How to deduplicate contacts across Gmail, Calendar and LinkedIn

Why one person becomes four records

Pick someone you work with closely and search for them across your tools. You will probably find them in your phone under a first name, in Gmail under a personal address from an old thread, in your calendar under a work address from a meeting invite, and in LinkedIn with a full name, a title and no email at all.

Each copy is correct and none is complete. The phone has the number, Gmail has the conversation, the calendar has the meetings, and LinkedIn has the job. Pull them all into one list without merging and an assistant thinks you know four people. It cannot see that you met this person ten times, because the meetings sit on one record and the emails on another.

So deduplication is the step that turns several contact lists into one network. It is also the step most likely to go wrong.

The rule that matters most

A merge combines two records into one and, in almost every tool, cannot be undone. That asymmetry decides everything else.

Leave a duplicate unmerged and the cost is clutter: the person appears twice and their history is split. Merge two different people and the cost is corruption: one person's emails, notes and meetings now belong to someone else, and an introduction can go to the wrong address. The second mistake is much worse than the first.

So the rule is to merge automatically only what you can prove, and to ask about everything else.

What is safe to merge automatically

Two signals really do prove the same person, and one more is strong enough with guards.

The same mailbox. Two records with the same email address are one person, but only after you normalise the address the way the provider does. Gmail ignores dots before the @, so jane.doe@gmail.com and janedoe@gmail.com are one inbox. Most other providers treat dots as significant, so j.smith@company.com and jsmith@company.com may be two people. Some providers also deliver jane+news@ to jane@, and others do not. Apply each rule only where the provider guarantees it.

The same LinkedIn profile. Two rows with the same profile URL are the same person, whatever names they carry.

An exact name match with a compatible company. This is not proof in the strict sense, but it is the only way to join a LinkedIn row that has no email to an inbox contact that has no title, and leaving those apart splits the network in half. It is safe enough to automate only with guards: skip it when both halves of the name are common, skip it when the two records point at different LinkedIn profiles, and skip it when the companies disagree.

What needs a human

Everything else is a suggestion.

The classic case is one person at two addresses, an old work email and a new one after a job change. The name matches and the addresses do not. That is consistent with one person who moved, and equally consistent with two people who share a name. Only someone who knows them can say.

The same goes for slightly different spellings, such as Andrew Hummel and Andrew Hummell, and for a LinkedIn row whose name matches one inbox contact exactly but has no other link. These are worth showing to a person in a short list with the evidence beside each pair. They are not worth merging blind.

A good review step also remembers the answer. If you say two records are different people, they should never be suggested again.

Where AI helps and where it should not decide

An AI model is good at noticing likely duplicates. It knows Bob and Robert are the same name and that a person who moved from Acme to Globex probably did not stop being themselves. That makes it an excellent source of suggestions.

It is a poor judge of when to apply them, for the same reason a person would be if they had never met the people in question. The model is confident about guesses. So let deterministic rules apply the provable merges, let the model propose the rest, and keep a human on the approve button. That applies to a personal agent working through your contacts too: it should draft the merge list, not execute it.

Tools that do this

Google Contacts has a Merge and fix view that finds duplicates inside your Google account, which is a sensible first pass if most of your network lives there. Dex merges duplicates across the sources it syncs, including email and LinkedIn.

NetworkOS splits the work into the same two tiers. It merges automatically only on the same mailbox under each provider's own address rules, the same LinkedIn profile, or an exact name with a compatible company and the guards above, and it sends lookalikes such as one name at two addresses to a review list where you confirm or reject each pair. A rejection is remembered. Merges there, as in most tools, are permanent, which is why the automatic tier is kept narrow. The limit is the one any careful system has: a person who shows up under two different names with nothing else in common stays as two records until you merge them yourself.

Doing this by hand once is manageable. Doing it every week as new emails, meetings and imports arrive is the reason a personal agent with a pile of files cannot keep a clean network; can a personal AI agent run your CRM on its own covers the rest of that picture.