Microsoft says label effectiveness drops noticeably past five labels. Most organisations design eleven.
Sensitivity labels classify and protect content wherever it travels, applying encryption, watermarks and container controls that stay with the file after it leaves your tenant. The technology works. The deployments that fail, fail on taxonomy design, and that is the part we spend the most time on.
- FiveMain labels, per Microsoft own guidance
- PersistentProtection travels with the file
- Copilot awareLabels are honoured by Copilot and agents
- ContainersTeams, sites, groups and Loop workspaces
Four things to get right before you publish a single label.
Labels are easy to create and genuinely awkward to unwind, because protection persists on content that already carries them.
- Deleting a label does not remove it from content. Microsoft states that if you delete a sensitivity label, it is not automatically removed from content and any protection settings continue to be enforced on content that had that label applied. A label published in error and later deleted leaves encrypted files behind that nobody can now administer through that label.
- Do not make an encrypting label the default. Microsoft advises directly that it is usually not a good idea to select a label that applies encryption as a default label for documents, because organisations routinely need to share documents with external users who may not have applications supporting the encryption or an account that can be authorised.
- Encryption with specified users and permissions caps you at 500 labels per tenant, against 1,000 or more otherwise. That ceiling is irrelevant for a sensible taxonomy and is a useful signal: if you are anywhere near it, the design is wrong.
- Guests and users from other organisations cannot see your sensitivity labels in applications. Your labels are your organisation vocabulary, so a label name that only makes sense internally is fine, but do not design the taxonomy expecting an external recipient to read and act on the label name itself.
Eight capabilities, and one design rule that matters more than all of them.
Five labels, which is Microsoft own number
Microsoft states that real-world deployments show effectiveness is noticeably reduced when users have more than five main labels, or more than five sublabels per main label, and warns that some applications cannot display all your labels when too many are published to the same user. Nearly every failed labelling project we have inherited failed here first, with a taxonomy designed by committee and understood by nobody.
Encryption that controls what people can do, not just who opens it
A label can control access using encryption for emails, meeting invites and documents, choosing which users or groups have permission to perform which actions and for how long. Microsoft gives the example of allowing all users in your organisation to modify a document while a specific group in another organisation can only view it. You can also let users assign permissions themselves at the point of labelling.
Content marking, with limits worth knowing before you design
Watermarks, headers and footers on labelled documents and emails, supporting variables such as the label name or document name. The limits are specific: watermarks are capped at 255 characters, headers and footers at 1,024, except in Excel, where the total limit is 255 including invisible formatting codes. Microsoft advises testing against your existing templates and workflows before publishing.
Container protection, which labels the space rather than the files
Labels can protect Teams, Microsoft 365 Groups, SharePoint sites, Viva Engage communities and Loop workspaces, controlling privacy settings, external user access and external sharing, access from unmanaged devices, private team discoverability and how channels can be shared. Microsoft is clear that this does not label the items inside, it controls access to the container where content is stored.
Automatic and recommended labelling
Labels can be applied automatically when content matches conditions you define, or recommended to the user through a policy tip they can accept or override. Recommended labelling is usually the better starting point, because it produces the same coverage over time while teaching people what the labels mean, and it does not encrypt something important on day one because a pattern matched wrongly.
Copilot honours labels, which is now the main driver
Microsoft 365 Copilot, Copilot Chat and Copilot agents recognise and use sensitivity labels. In conversations referencing data from multiple items, users see the label with the highest priority, typically the most restrictive. Where a label applies encryption, Copilot and agents return data from an item only if the user has been granted the copy permission, described as the EXTRACT usage right. That single mechanism is why labelling projects that stalled for years are now funded.
Protection that travels beyond your tenant
Because the label is stored in file metadata it stays with the content wherever it is saved, and because it is stored in clear text, third-party applications and services can read it and apply their own protective actions. Through Defender for Cloud Apps, content in third-party services such as Salesforce, Box or Dropbox can be detected, classified, labelled and protected even where the service itself does not support labels.
Order matters, twice, in ways that surprise people
Label order sets priority, with the most restrictive label at the bottom of the list and the least restrictive at the top, and that order defines what counts as downgrading for justification prompts. Separately, label policy order also sets priority, and where a user is covered by multiple policies the setting from the policy with the highest order number wins. Most unexpected behaviour in a labelling deployment traces to one of these two orderings.
Four things that separate a labelling project that lands from one that gets switched off.
We hold the line at five labels
Microsoft own guidance is that effectiveness drops noticeably past five main labels, and every workshop produces pressure to add a sixth for a specific department. We push back, because a taxonomy nobody can apply correctly produces mislabelled content, which is worse than unlabelled content since it carries false assurance.
We start with labelling, not with encryption
Microsoft explicitly supports labelling content without any protection settings, which gives you a visual map of data sensitivity plus usage reports and activity data, and you can add protection later. Starting visual means the taxonomy gets tested against real content and real behaviour before anything becomes hard to undo.
We test external sharing before anything is published
The reliable way to have a labelling project reversed is for an encrypted document to reach a client who cannot open it. We test the sharing paths that matter to your business first, with the specific external parties you actually work with, and design the encrypting labels around what survives that test rather than around the ideal policy.
We design for Copilot if Copilot is anywhere on the roadmap
Copilot and agents honour labels, showing the highest priority label across referenced items and returning data from encrypted items only where the user holds the copy permission Microsoft describes as the EXTRACT usage right. If Copilot is coming, the labelling design is the control that stops it surfacing content to people who should not see it, and it should be designed with that in mind rather than retrofitted.
Six UAE situations where labelling stops being optional.
An organisation about to deploy Copilot
Copilot reaches everything the user can already reach, which turns years of over-permissioned SharePoint into a discovery problem overnight. Labels are the control that limits what it returns, since encrypted items are only surfaced where the user holds the copy right. Doing the labelling work before the rollout is considerably cheaper than doing it after somebody surfaces a salary file in a chat response.
A regulated firm with client confidentiality obligations
Banks, finance companies, insurers and DIFC or ADGM entities are expected to classify information and control it according to that classification. Labels enforce it technically rather than aspirationally, and the usage reporting produces evidence that classification is actually happening, which a policy document alone cannot.
Professional services sending documents outside constantly
Law firms, consultancies and audit practices send sensitive material to clients daily, so a blanket block is not available to them. Encryption that permits view but not forward, applied selectively to genuinely confidential material, with the external sharing paths tested first, is the configuration that works in this sector.
A group with Teams and SharePoint sprawl
Hundreds of Teams and sites created over years with inconsistent privacy and external sharing settings. Container labels apply consistent controls over privacy, external access, unmanaged device access and channel sharing to the space itself, without needing anybody to label the thousands of files already inside it.
Manufacturing or engineering with intellectual property to protect
Drawings, designs, specifications and process documents shared with suppliers and contractors. Persistence is the point here: the label and its protection stay with the file after it leaves your tenant, which is the only control that still applies once a document is in a supplier hands.
Any organisation holding personal data under UAE PDPL
Knowing where personal data is and controlling who can act on it are both expected. Labels supply the classification and the enforcement, activity data shows what is happening to labelled content, and container labels stop personal data being placed in a Team that is open to external sharing by default.
How organisations classify data, and what each position actually delivers.
| Feature | Labels deployed | Policy on paper | Nothing |
|---|---|---|---|
Staff can classify content at the point of creation | Yes | In theory | No |
Protection travels with the file outside the tenant | Yes | No | No |
Encryption restricts what recipients can do | Yes | No | No |
Teams, sites and groups have matching controls | Yes | No | No |
Copilot respects the classification | Yes | No | No |
Downgrades are recorded with a reason | Yes | No | No |
Usage reporting on sensitive data | Yes | No | No |
Third-party services can act on the label | Yes | No | No |
Evidence for an auditor | Strong | Weak | None |
Frequency in the UAE market | Uncommon | Common | Common in SMEs |
What a label can be scoped to, and what that unlocks.
| Scope | What it covers | |
|---|---|---|
| Files and other data assets | Office files, Loop, Power BI, Microsoft Fabric items and Purview Data Map assets | |
| Emails | Messages, normally selected with files because attachments share the sensitivity | |
| Meetings | Calendar events, Teams meeting options and Teams chat, requires files and emails scopes too | |
| Groups and sites | SharePoint sites, Teams, Viva Engage communities and Loop workspaces, labelling the container not the items | |
| Third-party services | Salesforce, Box, Dropbox and similar, through Defender for Cloud Apps | |
| Purview Data Map, preview | SQL, Azure SQL, Azure Synapse, Azure Cosmos DB and AWS RDS schematised assets | |
| Copilot and agents | Highest priority label shown, encryption checked against the EXTRACT usage right |
Five steps, and the workshop is the one that decides the outcome.
- 1
Design the taxonomy, and defend the ceiling
A workshop with the people who actually handle sensitive information, producing at most five main labels with names that mean something to a normal employee. We test each proposed label by asking somebody outside IT to sort real documents with it, which reliably eliminates one or two labels that sounded necessary in the room.
- 2
Publish visual labels first, with no protection
Labels applied without protection settings, so people learn the vocabulary and you gather usage reports and activity data showing where sensitive content actually is. Nothing becomes hard to reverse in this phase, and the data it produces routinely changes the design before encryption is added.
- 3
Test encryption against your real sharing paths
Before any encrypting label is published, we test the external sharing your business genuinely depends on, with the parties you actually work with, on the devices and applications they actually use. This is the step that prevents the reversal, and it is the one most rollouts skip.
- 4
Add container labels and protection settings
Labels for Teams, groups, sites and Loop workspaces covering privacy, external user access, unmanaged device access and channel sharing. Then protection on the item labels that warrant it, with content markings tested against your document templates before publication rather than after.
- 5
Automate, then review
Recommended labelling first with policy tips, moving to automatic labelling once the conditions have been validated against real content. Then a review rhythm: usage reports, downgrade justifications in activity explorer, and a check on whether the taxonomy is still being applied the way it was designed.
What organisations ask about sensitivity labels.
Fifteen questions worth answering first.
Taxonomy
- How many main labels are you proposing?Microsoft says effectiveness drops noticeably past five.
- Can a normal employee explain the difference between two adjacent labels?If not, the taxonomy is too fine.
- Which labels apply encryption, and which are visual only?Labelling without protection is a legitimate first phase.
- Have you set the order most restrictive at the bottom?Order defines priority and what counts as a downgrade.
- Do you need sublabels, or is that complexity you will regret?Five sublabels per main label is the same stated ceiling.
Behaviour
- Automatic labelling, recommended, or manual to begin with?Recommended usually beats automatic on day one.
- Will you set a default label, and will it encrypt?Microsoft advises against an encrypting default for documents.
- Is labelling mandatory before saving or sending?Effective, and frustrating without a sensible default.
- Do you require justification for downgrades?Prompted once per app session in Office apps.
- Who reviews justifications in activity explorer?The reason is recorded and almost never read.
Rollout
- Do you share encrypted documents with external parties?Test this before publishing, not after.
- Have you tested content markings against your templates?Watermark and header limits are real and bite in Excel.
- Which users get which label policy?Policies publish to users and groups, not to locations.
- Is Copilot in use or planned?It changes the priority of this work considerably.
- Who explains the labels to staff?A Learn More URL can be surfaced in the Office apps.
The pages around this one.
Microsoft Purview
The wider governance and compliance platform this sits inside, including retention, audit and compliance management.
Data loss prevention
The enforcement layer that acts on classified content when somebody tries to send or copy it somewhere it should not go.
Microsoft 365 Copilot
The deployment that has made labelling urgent for most organisations, and what has to be in place before it is switched on.
Write down the labels you think you need, then count them.
If the number is above five, we will spend the first session getting it down, because that single decision predicts whether the deployment works better than any technical choice that follows it.
Related Services
Explore more solutions that work great with this service
Microsoft Purview
Data governance and compliance solutions
DLP Solutions
Microsoft Purview DLP and labels
Microsoft Copilot
AI-powered productivity with Copilot
Defender for Cloud Apps
Shadow IT, SaaS posture and the OAuth apps already reading your mail
UAE PDPL Compliance
Federal Decree-Law 45 of 2021 readiness and operations
Microsoft 365 Security Audit
Tenant review, and how far back your evidence really goes
Microsoft Security Dubai
Entra, Defender, Purview, Sentinel, and what you already own
ISO 27001 Certification UAE
The 2022 edition, and whether you should certify at all