Phishing simulation with an approved pretext
Brief
A phishing simulation measures how an organization responds to one convincing email: who clicks, who reports, and how fast. The pretext, the audience and the stop conditions are agreed in writing before anything is sent.
On this page
- Check first
- How do I write the approval and the stop conditions?
- How do I set up the sending domain and its authentication records?
- How do I build a landing page that records the click and never collects a password?
- How do I time the send and build the exclusion list?
- Which metrics matter and how are results given back?
- The defender sees
- Done when
- Not covered
- Tools
- Related research
- Reviewed
- Situation
- The client has approved a pretext in writing, named the audience, and given us a contact who can stop the exercise at any hour. The scope letter names the sending domain, the dates and the exclusions. We never collect a password, never send to anyone outside the approved list, and never report results by name.
- Goal
- A campaign record: every send, click and report with a timestamp, summarized by team and never by person.
- Outputs
- Representative examples on placeholder hosts, trimmed to the lines that matter. Not captured from a public fixture yet.
Check first
- The written approval: signer, pretext, audience, dates, and the one person who can stop the exercise.
- Pretexts the client has ruled out: pay, layoffs, health, real suppliers.
- The exclusion list, and who maintains it until the morning of the send.
- How users report mail today, and where those reports land.
- Whether the client's SOC knows the dates; a simulation treated as a live incident tests nothing.
Steps
How do I write the approval and the stop conditions?
The approval is one page the client signs:
Pretext: <one paragraph, the exact subject line and sender display name>
Audience: <team or list name>, <count> recipients, exclusions attached
Sending domain: sim.tester.example (ours, registered for this test, parked after)
Dates: <send window>, results by <date>
Stop: <contact and number>; we stop within 15 minutes of the call
Data: click and report timestamps only; no credential field; results by team
Expected output
Approval signed: 2026-10-20 by <role, not name>
Stop test: called the contact at 09:00, acknowledged in 4 minutes # <- the stop path works before the send does
Exclusions: 41 recipients removed, list dated 2026-10-26
We call the stop contact once before the send, so we know the number is answered. A pretext that impersonates a real third party, promises money or threatens a job is refused: it measures fear, not judgement.
Negative result. An approval with no named stopper, no exclusion owner or no date is not an approval, and the send waits.
Fix. A simulation policy the client owns: who approves a pretext, which subjects are never used, when the SOC is told, and that results never reach performance reviews.
How do I set up the sending domain and its authentication records?
We send from a domain registered for the test, never a lookalike of a real company, and it carries SPF, DKIM and DMARC like any legitimate sender: the question is whether people spot a plausible email, not whether the filter spots a broken one. The SPF, DKIM and DMARC page reads those records; here we confirm ours are live:
dig +short TXT sim.tester.example
dig +short TXT _dmarc.sim.tester.example
Expected output
"v=spf1 ip4:203.0.113.25 -all"
"v=DMARC1; p=reject; rua=mailto:dmarc@sim.tester.example" # <- our domain authenticates; a block now is a content verdict
The send runs in two phases: the first batch with no allow entry, to learn what the gateway does with a well-formed phishing email; the second registered with the gateway so it reaches inboxes and measures people. In Microsoft 365 that is the advanced delivery policy for non-Microsoft phishing simulations, keyed on the MAIL FROM or DKIM domain and the sending IP address; Safe Links still wraps the URLs without blocking them, and a user's report lands on the User reported tab without raising an alert. The entry is removed when the test ends.
Negative result. The unregistered batch quarantined or junked, DeliveryAction Junked or Blocked in the client's EmailEvents table.
Fix. If the unregistered batch reached inboxes, tune the anti-phishing policies for first-contact senders and lookalike display names before training anyone to catch what the filter should have.
How do I build a landing page that records the click and never collects a password?
The landing page is the awareness page: it records the click by the recipient id in the link and shows one paragraph saying this was an authorized simulation and how to report real phishing. In Gophish that means Capture Submitted Data and Capture Passwords both unchecked, and no form. We check it as a stranger would:
curl -s 'https://sim.tester.example/?rid=7f3a1c' | grep -ci '<input'
Expected output
0 # <- no input element on the page, so nothing a person types can reach us
A page that collects passwords, even to discard them, gives the test its own breach risk; Gophish's interface warns that captured passwords are stored in cleartext.
Negative result. Any count above zero means a form slipped in; the campaign waits.
Fix. A captured password should be worth nothing: phishing-resistant MFA, FIDO2 keys or passkeys, starting with administrators and finance, with SMS and plain push approvals retired, as the MFA field note sets out.
How do I time the send and build the exclusion list?
Exclude: anyone on leave, anyone in a role the client flags (legal, HR, crisis desk),
executives unless the approval names them, anyone who asked not to be tested
Timing: working hours, mid week, not payday or a results week, one message per person,
sends spread over 90 minutes so the first report can arrive before the last send
Expected output
09:00 batch 1 120 sent
09:30 batch 2 118 sent
10:00 batch 3 121 sent # <- 359 sent, 0 duplicates, 41 exclusions honoured
One message per person, once. The spread is for the defender: if the first report arrives at 09:12 and the SOC pulls the message from every inbox, the later batches measure the response.
Negative result. Zero matches between the send list and the exclusion list, and one delivery row per recipient.
Fix. The landing page is the first training moment: two sentences on what gave the message away. Training follows for everyone, not only clickers, and nobody is named or disciplined for a click.
Which metrics matter and how are results given back?
The click rate is the number everyone asks for and the least useful. The report rate and the time to first report show whether the organization defends itself; a team with more reports than clicks is doing well.
Click rate: clicked / delivered, per team
Report rate: reported / delivered, per team (a report after a click still counts)
First report: time from the first delivery to the first report that reached the SOC
Group size: no row for fewer than 10 recipients; smaller teams merge
Retention: per-recipient record deleted when the report is accepted
Expected output
Team Sent Clicked Reported First report
Finance 62 9 21 09:11 # <- reports above clicks: this team is the control working
Operations 140 31 12 09:48
Results go back by team with a minimum group size, so no row describes one person; the per-recipient record exists for the SOC to match its logs and is deleted once the report is accepted.
Negative result. Clicks with no reports means the reporting path is missing or unknown: the lead finding.
Fix. The report button turns one alert employee into protection for everyone: one click in every mail client, a reporting mailbox the SOC reads, and a procedure to pull a reported message from every inbox within minutes.
The defender sees
In Microsoft 365, each message is a row in EmailEvents with our SenderMailFromDomain, its DeliveryAction and, for the registered batch, the phishing simulation override in OrgLevelPolicy. Each click on a Safe Links wrapped URL is a row in UrlClickEvents with ActionType ClickAllowed, the AccountUpn and the NetworkMessageId that joins it to the message. Each press of the Report button is an entry on the Submissions page's User reported tab and a message in the reporting mailbox. Other gateways keep the same three records. The report says how long the SOC took to act on the first report and whether remaining copies were pulled. Clicks with no record in the gateway's logs is a logging finding, not a people finding.
Done when
Every approved recipient has received one message, the landing page has recorded clicks with no credential field, the report rate and click rate are counted by team, and the sending domain is parked.
Not covered
Credential capture, attachments or payloads, voice and SMS pretexts, and the client's own mail records; every output here is representative on placeholder domains.