Publish consistently under one name
The single most effective habit, and the cheapest. Decide how your name appears — full middle name, initial, or neither — and use exactly that form on every submission for the rest of your career.
This matters most for researchers whose names are transliterated, hyphenated, or common in their language of origin. Algorithmic clustering handles a consistent name well and an inconsistent one badly.
If you have already published under variants, you cannot change the past, but you can stop adding to the problem and merge what exists.
One consistent name form on every submission prevents most profile fragmentation before it starts.
A routine worth keeping
Twice a year is enough. Open your Scopus and Web of Science author records and check three things: every paper of yours is present, nothing that is not yours has appeared, and your current affiliation is right.
Check that recent papers have actually propagated to ORCID. Automatic feeds work well but not invariably, and a gap is easier to fix at six months than at five years.
Note anything wrong and submit the corrections in one batch rather than as you notice them — the review process is the same effort for one document or ten.
Around a career change
Changing institution is the highest-risk moment for a profile, and it is the moment nobody has time to think about it. New affiliation, often a shifted research area, sometimes a new email on submissions — exactly the conditions that split a record.
Do three things in the first month: update the affiliation on ORCID, make sure your ORCID iD goes on every submission from the new post, and check three months later that new papers attached to the existing profile rather than creating a new one.
Keep your own record too. A simple list of every publication with its DOI, the journal, and the databases it is indexed in takes minutes to maintain and is invaluable when someone asks for evidence at short notice — for a promotion case, a grant, or a visa petition where the documentation matters as much as the work.