Ask a team running a multilingual site whether localized pages are reviewed before they go live and the answer is almost always yes.
Ask what happens when the reviewer does not respond for four days, and the answer usually arrives more slowly.
That second question is the gate. The first is an intention.
Approval gates rarely fail because someone decided to skip review. They fail because the gate had a single owner with another full-time job, became a queue, and the queue got routed around by people with a launch date.
What “approved” has to mean before a gate can hold
A gate needs three things that a policy document does not supply on its own.
A named person, not a team. “Marketing reviews the French” distributes responsibility until it evaporates. A queue addressed to four people is a queue nobody owns.
A criterion the reviewer is checking against. Without one, review becomes preference, and preference has no endpoint — which is where the endless revision cycle comes from. Accuracy, terminology against the approved list, tone for the surface, market-specific legal wording: each is checkable. “Does this read well?” is not.
A record of the decision. Not because anyone distrusts the reviewer, but because six months later the useful question is which version was approved and against which terminology — and a nod in a meeting cannot answer it.
Three shapes a gate can take
The mistake is applying one shape to everything. Most content does not warrant a blocking gate, and the content that does warrant one is exactly where a bypass is most expensive.
| Gate | Publish is blocked until | Right for | How it fails |
|---|---|---|---|
| Blocking | the named reviewer approves | regulated, legal and safety text; pricing; brand claims | one reviewer with a day job; the queue grows and people route around it |
| Advisory | nothing — reviewer is notified, publishing proceeds | high-volume, low-risk content: help articles, release notes | the notification goes unread and drift accumulates invisibly |
| Sampled | nothing, but a defined share is reviewed after publish, with a rollback path | most marketing and product content | the sample gets picked by convenience rather than by risk |
Choosing per content type rather than per project is the decision that makes the rest workable. A single blocking gate across everything guarantees the queue that breaks it.
What tooling can enforce, and what it cannot
The question usually arrives as “is there a turnkey solution for this,” and the honest answer is: for half of it, yes.
A TMS or CMS workflow can genuinely enforce:
- that a string cannot move to published state without a recorded approval
- who is permitted to give that approval, by role and locale
- an audit trail of what was approved, by whom, against which version
- notification and escalation when an item sits beyond a threshold
That is real and worth having. Do not build it by hand.
No tool supplies:
- authority. Software can require an approval from a person holding a role. It cannot make that person the right judge of how your product should sound in Korean.
- the criterion. A workflow enforces that someone clicked approve. What they were checking is something you define, once, per market.
- availability. The most common cause of a broken gate is that the reviewer is on holiday, and no configuration screen solves that.
A team that buys the mechanism and skips the other three ends up with an audited bottleneck — which is a better-documented version of the same problem.
The four decisions that keep a gate from becoming the bottleneck
Separate “must approve” from “must be informed.” Most people currently in an approval chain are there to stay in the loop. They can be notified without holding a lock. Shortening the list of people who can block is usually the single largest reduction in queue time available.
Give every gate a deputy. A gate with one owner is a single point of failure with a calendar. The deputy needs the same criterion and the same authority, agreed in advance rather than improvised during a launch week.
Decide the default when the clock runs out — in writing. This is the one almost nobody records. If the reviewer has not responded within the agreed window, does the content publish or hold? Both answers are defensible and they suit different content. What is not defensible is leaving it undecided, because then the decision gets made by whoever is most senior in the room on the day, and it will be made differently next time.
Measure time in queue, not just pass rate. A gate reporting 98% approval and a nine-day median wait is not working well; it is being avoided or rubber-stamped. Queue time is the number that predicts whether people will route around it.
Buy the mechanism if it saves you building it — workflow state, permissions, audit trail. But the gate holds or fails on three things no tool provides: a named reviewer with real authority, a written criterion, and a decided default for when nobody answers in time.
Where this sits
This is the layer after the decisions and before the verification.
The website localization checklist covers what to define before a launch, including who holds sign-off authority for each language. This piece assumes that is defined and asks the next question: what makes the gate actually hold once content is shipping continuously.
After publish, the checks move from who approved to what is true on the page — layout under text expansion, formats, switch behaviour — which is the localization testing checklist. And for keeping versions aligned as the source changes underneath all of this, see how to prevent locale drift.
If the review layer is the part of your multilingual operation that is slowing publishing down, our multilingual web and product content service runs it as a managed layer with named accountability per language. To get a read on where your current gate is queueing, start here.