When spam and phishing share one queue, one verdict scale and one dashboard, the SOC spends its time on newsletters while a credential phish sits unanswered in the same pile. Spam is unwanted mail; phishing is mail designed to make someone act against the organisation's interest, by handing over credentials, approving a payment, running a file or granting access. They overlap in delivery and look alike to users, but they need different filtering actions, different handling when users report them, and different measures of success. This guide covers how to design the triage pipeline so that the two separate early: how mail platforms classify each, how user reports should be routed, what the SOC does with each class, and which common metrics hide phishing inside spam numbers. It is for email security engineers, SOC leads and awareness programme owners. Report button mechanics are in the Phishing Report Button and User Reported Message Triage Guide; the phishing landscape is in the Phishing Guide.
Key takeaways
- Intent is the dividing line, not annoyance. Spam wastes attention; phishing aims at an action: a credential entered, a payment sent, a file run or an OAuth consent granted. The second is a potential incident, the first is a hygiene problem.
- Platforms already separate the verdicts. Mail filters assign distinct spam, bulk and phishing classifications with different default actions; pipelines that collapse them into "junk" throw that signal away.
- A user report is a classification attempt, not a fact. Most reports are spam, bulk or legitimate mail; design the pipeline to dismiss those automatically and spend analyst time on the phishing minority.
- Response paths diverge after the verdict. Spam ends in sender reputation, blocking or unsubscribe; phishing starts an incident: purge every copy, find who clicked, and act on the identities involved.
- Blended metrics mislead. "Threats blocked" and "reports received" mix spam with phishing; measure each class separately or a phishing surge will look like a quiet week.
Definitions that hold up in triage
Spam: unsolicited mail sent in volume, usually commercial, sometimes fraudulent in a low-effort way (fake products, advance-fee schemes). Its harm is cost, distraction and, at the edges, low-grade fraud. In law it is largely a consent and transparency issue: in the UK, regulation 22 of the Privacy and Electronic Communications Regulations requires consent for marketing email to individuals (with a soft opt-in for existing customers), forbids concealing the sender's identity and requires a valid opt-out address. The ICO notes its guidance is under review following the Data (Use and Access) Act. See the CAN-SPAM and TCPA Guide for the US rules.
Bulk or graymail: mail from legitimate bulk senders that some recipients want and others do not. Microsoft treats bulk mail, also called gray mail, as its own class scored on a bulk complaint level (BCL). Graymail is not spam in the legal sense and is not malicious, but it generates most user "junk" reports in a typical tenant.
Phishing: mail crafted to deceive the recipient into an action that benefits the attacker. MITRE ATT&CK lists it as T1566 Phishing, with sub-techniques including T1566.001 Spearphishing Attachment and T1566.002 Spearphishing Link, and separately T1598 Phishing for Information for reconnaissance. Phishing includes mail with no link or attachment at all: a payment request or a "call this number" lure is phishing because of its intent. See Spear Phishing vs Whaling vs BEC and the Callback Phishing (TOAD) Guide.
Side-by-side comparison
| Dimension | Spam and bulk | Phishing |
|---|---|---|
| Sender goal | Reach as many inboxes as possible; a sale, a click, a list signup | Make a specific person or role act: credentials, payment, execution, consent |
| Targeting | Untargeted or loosely segmented | Ranges from mass campaigns to one named executive |
| Payload | Advertising, low-quality offers, occasional crude fraud | Credential harvesting page, malicious attachment, payment instruction, OAuth consent link, phone number |
| Sender identity | Often the real sender, sometimes disguised | Impersonated brand, colleague, supplier or service; lookalike or compromised domain |
| Harm if it gets through | Lost time, storage, possible unwanted subscription | Account takeover, fraud loss, malware, data breach |
| Correct filter action | Junk folder or bulk folder; user can release | Quarantine with admin control; post-delivery removal from all mailboxes |
| What happens after a user report | Tune sender reputation, block or unsubscribe; no incident | Cluster, purge, identify clickers, open an incident if anyone interacted |
| Owner | Messaging or email administration | SOC and incident response, with identity team support |
| Framework anchor | NIST SP 800-53 SI-8 Spam Protection | NIST SP 800-53 IR-4 Incident Handling, IR-6 Incident Reporting |
Threat model: why the difference matters
Spam
What it exploits: cheap sending, harvested addresses and weak consent practices. The defender's concern is volume, user productivity and reputation of the organisation's own domains if its systems are abused to send it.
Failure mode: filters that are too aggressive. Overblocking legitimate bulk and transactional mail drives users to ask for allow-list entries, and broad allow lists are a common route by which phishing bypasses filtering later.
Architecture in words: content and reputation scoring at the gateway or in the cloud mail service, a bulk threshold, a junk folder the user controls, and feedback to the provider so models improve.
Phishing
What it exploits: trust in a sender, urgency and the user's ability to authenticate, pay or run code. Modern phishing increasingly uses legitimate infrastructure (cloud file-sharing links, compromised supplier mailboxes) so reputation alone misses it.
Failure mode: treating phishing that reached the inbox as a filtering miss to be reported to the vendor, rather than as a possible compromise. The real question is not "why did the filter miss it" but "who interacted, and what did they give away".
Architecture in words: authentication checks (SPF, DKIM, DMARC), impersonation and spoof detection, link and attachment analysis at delivery and click time, quarantine under admin control, post-delivery removal, user reporting, and an incident workflow tied to identity controls. See the DMARC, SPF and DKIM Guide.
How mail platforms classify each
Both major cloud mail platforms already keep spam and phishing apart at filtering time. Platforms are listed alphabetically; settings are as described in each vendor's documentation in November 2026.
Google Workspace: Gmail's advanced safety settings (Gmail > Safety in the Admin console) cover attachments, links and external images, and spoofing and authentication. For each setting the admin chooses one of three actions: keep the message in the inbox with a warning banner, move it to spam, or quarantine it, where users see nothing and an admin reviews it. Users can mark mail as spam or use "Report phishing", and Google states that moving a message to Spam sends Google a copy for analysis.
Microsoft Exchange Online and Defender for Office 365: anti-spam policies assign separate verdicts with these default actions: spam and high confidence spam move to the Junk Email folder; phishing is quarantined; high confidence phishing is always quarantined, and users can only request its release; bulk moves to Junk at the configured threshold. The bulk complaint level runs from 0 (not bulk) to 9 (many complaints), with a default threshold of 7, 6 in the Standard preset and 5 in Strict, where bulk is quarantined rather than moved to Junk. Zero-hour auto purge removes delivered spam and phishing after a verdict changes, and is on by default.
| Verdict class | Google Workspace action options | Microsoft default action | Pipeline treatment |
|---|---|---|---|
| Bulk or graymail | Spam classification; user can unsubscribe or move | Junk Email at BCL threshold (quarantine in Strict) | No SOC involvement; user controls |
| Spam | Move to spam | Junk Email | No SOC involvement; reputation feedback |
| Suspected phishing or spoofing | Warning banner, move to spam or quarantine | Quarantine | Analyst review of releases; cluster with reports |
| High confidence phishing or malware | Quarantine | Quarantine, release only on request | Incident check: did any copy reach an inbox? |
Microsoft's documentation adds that in cloud organisations the spam confidence level (SCL) stamped in the X-Forefront-Antispam-Report header does not set the verdict; the CAT value in the same header shows the category the filter assigned. Triage automation should read the category, not infer phishing from a high SCL.
Designing the reporting and triage pipeline
The design goal is to separate the classes as early as possible and spend people only where intent is malicious.
- Give users one button with a clear choice. Microsoft's built-in button offers Report phishing, Report junk and Report not junk; third-party add-ins tag reports with subject prefixes (1| or Junk:, 2| or Not junk:, 3| or Phishing:), and a report with no prefix is treated as phishing. Keep the "junk" path for spam so users do not inflate phishing numbers, and accept that users will misclassify in both directions.
- Route by intent, not by the user's label alone. Run every report through the same enrichment (authentication results, filter category, URL and attachment verdicts, sender history), then route: known bulk and spam to an automated close with feedback to the provider; suspected phishing to clustering and analyst review.
- Cluster phishing reports into campaigns. Group by sender, URL, attachment hash and subject so one decision covers every copy; then search all mailboxes for unreported copies.
- Close spam reports without an analyst. Auto-respond to the user, add sender or domain blocks where the volume justifies it, and point the user to unsubscribe for legitimate bulk. Under RFC 8058, one-click unsubscribe uses a List-Unsubscribe-Post header with an HTTPS URI, and a receiver should only offer it when a valid DKIM signature covers both unsubscribe headers. Gmail requires one-click unsubscribe for senders of more than 5,000 messages a day to Gmail accounts, as part of its sender guidelines in force since 1 February 2024.
- Escalate phishing with interaction to incident response. Any click, credential entry, reply, file open or consent grant opens an incident. See the Phishing Incident Response Playbook.
- Feed verdicts back. Tell the reporter the outcome (phishing, spam, no threat), submit misses to the vendor, and tune filters and allow lists from confirmed cases.
Rollout and rollback: pilot automated closing of spam and bulk reports with one department, compare automated verdicts with analyst verdicts for two to four weeks, and keep a switch that sends all reports to analysts if false negatives appear. External reporting adds a public-good path: in the UK the NCSC Suspicious Email Reporting Service accepts forwarded suspicious mail at [email protected], and states that users do not need to forward mail already in their spam folder.
SOC triage: what each verdict triggers
| Verdict | Containment | Investigation | Identity action | Closure |
|---|---|---|---|---|
| Bulk | None | None | None | Automated reply; unsubscribe advice |
| Spam | Optional sender block | None unless volume spike | None | Automated reply; provider feedback |
| Phishing, no interaction | Purge all copies; block sender, domain and URLs | Campaign scope: who received it | None, or targeted watch on recipients | Reporter thanked; indicators shared |
| Phishing, link clicked | Purge; block; isolate device if a download followed | Proxy and sign-in logs for credential entry | Revoke sessions, reset credentials, check MFA methods | Incident record |
| Phishing, consent or token granted | Purge; disable the application grant | Audit the app's API activity | Revoke grants and refresh tokens | Incident record |
| Phishing, payment or data reply | Purge; contact finance or the data owner | Mailbox rules, forwarding, replies sent | Check for mailbox compromise | Incident record; fraud reporting |
The identity columns are where phishing and spam differ most. A phished credential, a session token captured by an adversary-in-the-middle page, or an illicit OAuth consent all leave an attacker with access that outlives the email. See Consent Phishing vs Credential Phishing and the AiTM Phishing Guide.
Metrics that blur the two, and better ones
| Common metric | Why it misleads | Better measure |
|---|---|---|
| Total threats blocked | Dominated by spam and bulk; phishing changes vanish in the total | Phishing and malware blocked per week, reported separately from spam |
| Reports received | Rises with marketing volume as much as with phishing | Confirmed phishing reports, and share of reports that were phishing |
| Report rate in simulations only | Rewards reporting test mail, ignores real-world accuracy | Real phishing reported by at least one user before filter or SOC detection |
| Mean time to close a report | Fast automated spam closes hide slow phishing handling | Time from first phishing report to purge of all copies |
| Filter accuracy as one figure | A spam false positive and a phishing false negative carry very different costs | Phishing false negatives (reached inbox) and legitimate mail quarantined, tracked apart |
Add two outcome measures: users who interacted with confirmed phishing before removal, and identity actions completed (sessions revoked, grants removed) within the target time. For a wider metrics set, see the Security Operations Metrics Guide.
Decision guidance
- Small organisations with no SOC: rely on platform defaults, which already quarantine phishing, and send phishing reports to whoever owns incidents. Do not route junk reports to a person at all.
- Mid-sized organisations: automate spam and bulk closure, keep analysts on clustered phishing, and set a service level for purge time.
- Large or regulated organisations: integrate reports into the SOAR or case platform, automate identity actions for clicked phishing, and report phishing metrics to leadership separately from spam volume.
- High-value roles such as finance, executives and administrators: treat any reported message as possible phishing until enriched, because targeted lures often lack the indicators that filters key on.
Worked scenario
On a Monday the reporting mailbox receives 400 reports. Enrichment classifies 310 as bulk newsletters and 60 as spam; both groups close automatically with a verdict email. The remaining 30 cluster into two campaigns. The first, 22 reports of a fake parcel notice with a credential page, reached 900 mailboxes before the filter caught up; the copies are purged and proxy logs show 14 clicks and three password submissions. Those three users have sessions revoked, passwords reset and MFA methods reviewed. The second, eight reports of a supplier "bank details changed" email with no link, comes from a lookalike domain; finance confirms no payment was made, and the domain is blocked. The weekly dashboard shows 400 reports, 30 phishing, two campaigns, purge times of 40 and 25 minutes, and three users with credential exposure contained within the hour.
Common failure patterns
- One queue for everything: analysts work reports first-in-first-out, so a targeted phish waits behind marketing mail.
- Allow lists created to stop spam complaints: broad sender or domain allow entries added to rescue legitimate bulk later let phishing from a compromised supplier through untouched.
- Treating no-link phishing as spam: payment and callback lures carry no URL or file, so link-focused triage closes them as harmless.
- Stopping at purge: the message is removed but nobody checks who clicked, submitted credentials or granted consent, so the account takeover proceeds. The Twilio breach shows how phished employee credentials, taken over SMS rather than email, opened a path to customer data.
- Reporting a blended number upward: leadership sees "millions of threats blocked" and misses that phishing reaching inboxes doubled.
Framework mapping
| Framework | Reference | Spam | Phishing |
|---|---|---|---|
| NIST SP 800-53 Rev 5 | SI-8 Spam Protection | Primary control | Supporting |
| NIST SP 800-53 Rev 5 | SI-3 Malicious Code Protection, IR-4 Incident Handling, IR-6 Incident Reporting, AT-2 Literacy Training and Awareness | Not required | Primary controls |
| NIST CSF 2.0 | RS.MA-02 Incident reports are triaged and validated; DE.AE-02 Potentially adverse events are analyzed | Light touch | Core of the pipeline |
| MITRE ATT&CK | T1566, T1598; mitigations M1054 Software Configuration, M1021 Restrict Web-Based Content, M1017 User Training | Not applicable | Technique and mitigation reference |
| MITRE D3FEND | D3-EF Email Filtering, D3-ER Email Removal | Filtering | Filtering and post-delivery removal |
| CIS Controls v8.1 | Control 9 Email and Web Browser Protections, Control 14 Security Awareness and Skills Training | Control 9 | Controls 9 and 14 |
| ISO/IEC 27001:2022 | A.6.8 Information security event reporting, A.5.25 Assessment and decision on information security events | Not an event | Event reporting and decision |
See MITRE ATT&CK T1566 Phishing and D3-ER Email Removal.
Practitioner checklist
- Keep spam, bulk and phishing as separate verdicts end to end, from filter to dashboard.
- Offer users distinct "junk" and "phishing" report options and accept they will mislabel.
- Enrich every report automatically and route by the filter category and analysis, not the user's label alone.
- Close spam and bulk reports automatically with a verdict email and provider feedback.
- Cluster phishing reports into campaigns and purge every copy, reported or not.
- Trigger identity actions (session revocation, credential reset, grant removal) for any interaction with confirmed phishing.
- Review allow lists quarterly and remove entries added only to reduce spam complaints.
- Report phishing metrics to leadership separately from spam volume.
Standards and references
- Microsoft Learn: Anti-spam protection in cloud organisations
- Microsoft Learn: Bulk complaint level (BCL)
- Google Workspace: Advanced phishing and malware protection
- Google: Email sender guidelines
- IETF RFC 8058: Signaling One-Click Functionality for List Email Headers
- IETF RFC 5965: An Extensible Format for Email Feedback Reports
- NCSC: Reporting a suspicious email
- ICO: Electronic mail marketing under PECR
Related NHI Mgmt Group resources: Email Security Guide, Bulk Sender Requirements Guide, How to Spot a Phishing Email Guide, Teams and Slack Phishing vs Email Phishing.