Test Your Staff, Do Not Assume Them
You cannot know how your people would respond to an attack until you test. Here is why assumption fails, and the rules that keep testing from damaging trust.

There is a phishing lure that outperforms almost everything else. You write to an individual employee, from something that looks like HR, with a subject line about their own pay or their own job. Confidential compensation review outcome. Changes to your role effective Monday. Click rates go through the roof, because the recipient's hands are shaking before they finish the subject line.
It works. That is not a reason to send it.
Holding both of those thoughts at once is the entire discipline of testing people well. You have to accept that testing is necessary, because assumption is worthless. And you have to accept that a test which humiliates people destroys the thing it was measuring. Most organizations resolve the tension by picking one side: they either never test, or they test as though the goal were to win.
Why assumption fails, specifically
Every alternative to testing measures something other than behavior.
Policy measures intent. A document saying staff must verify payment changes by telephone tells you what the organization has decided. It says nothing about Thursday afternoon, when the supplier is chasing and the person who normally checks is on leave.
Training completion measures attendance. Ninety-four percent finished the module. That figure appears in reports as though it were an outcome. It is a record that a browser window was open.
Confidence surveys measure confidence. Ask people whether they can spot a phishing email and most say yes. The answer is uncorrelated with whether they can, in a direction that should worry anyone relying on it.
Incident history measures luck. No successful attacks in two years might mean your people are excellent. It might mean nobody has targeted you properly yet. Those two states look identical from the inside and require completely different responses.
A realistic test is the only instrument that produces behavior rather than description. It also produces something none of the alternatives can: a distribution. Not "our staff are fairly good" but operations at twenty-eight percent, finance at nine, with the first report arriving in under two minutes. One of those is a mood and the other is a plan.
What a test is actually for
This is where programs go wrong, and the error is usually in the stated goal.
If the goal is "reduce our click rate", you will eventually get there by making the tests easier, and you will have learned nothing. If the goal is "identify who is risky", you have built a surveillance exercise and people will behave accordingly.
The goal worth having is narrower: find out which lures work on which teams, so the next month's effort goes somewhere useful, and find out whether anyone reports, because reporting is the only behavior that protects the whole organization rather than one person.
That reframing changes what you measure. A person who notices a phishing email and deletes it has saved themselves. A person who reports it has given the security team the chance to warn everyone and block the sender before the fortieth recipient opens it. In a click-rate view those two people are identical. They are not worth the same, and the metric should say so.
The rules that keep it honest
These are the constraints we build into the product, and they are worth applying whatever tool you use.
Announce the program, not the campaign
Tell everyone that simulated phishing happens, why it happens, what occurs if they click, and that nothing bad follows from clicking. Then do not warn anyone that a particular email is coming, because a warned test measures nothing.
The distinction is easy to defend and easy to get wrong in both directions. Organizations that announce nothing are running tests on people who did not consent to the program. Organizations that announce each campaign are generating numbers that mean nothing.
Require a named human to authorize every campaign
Every simulation in our platform requires an administrator's approval before it sends, and the approval is recorded. This is partly governance and partly a brake: someone has to look at the email and decide it is defensible. Lures that should never have gone out tend not to survive a person reading them with their own name attached.
Never collect a credential
Some platforms put a fake login page behind the link and record what people type. The justification is that submitting a password is a stronger signal than clicking.
It is a stronger signal, and it is not worth it. You have now collected your own employees' real passwords in a system that did not previously hold them, to establish a fact you were already confident of. Our simulations measure the click and stop there, and the code that assembles an AI-written lure rejects any draft that asks the reader to type or send a credential at all.
Do not weaponize personal distress
Lures that use somebody's pay, job security, health or family to produce a click work extremely well. They also convert a security exercise into something people experience as cruelty, and the resentment outlasts any lesson. The practical test is whether you would be comfortable reading the email aloud to the recipient afterwards and explaining why you chose it.
Business pretexts give you everything you need. Password expiry, a shared file, an invoice, a parcel, a document to sign, an IT survey, a quarantine notice. These are the attacks people actually receive, which is the point.
Make the moment of failure a moment of teaching
When someone clicks, what happens in the next ten seconds determines whether the test was worth running. A page that names the specific tells in the email they just clicked, addressed to them personally, with a three-minute lesson on that attack type, converts an uncomfortable moment into the most effective teaching opportunity in the whole program. Nothing they typed is kept, and the page says so.
What must not happen: an email to their manager, their name on a list, or a company-wide message. One click is not a performance issue. It is a Tuesday.
Close the loop with the people who report
Every report should get an answer, including the ones that turn out to be nothing. People who report and hear nothing conclude that reporting is pointless, and they are not wrong to conclude it.
Where several people report the same email, treat it as one incident with several reporters rather than several tickets, and tell all of them the outcome. Making reporting a single step is half of it; answering is the other half.
Keep the difficulty honest
Label how hard each lure was, and vary it deliberately. An easy lure has visible tells. A hard one is internal, specific and plausible. Without that label, a falling click rate is ambiguous, which is precisely the problem NIST built the Phish Scale to address. A program with no difficulty record will drift toward easy tests, because easy tests produce better slides.
Let pilots be pilots
If you tell an administrator that a practice round will not count toward the organization's score, it has to actually not count. Our scoring excludes campaigns marked as not contributing. A score that silently absorbs practice data is a score people stop believing, and once they stop believing it, the measurement program is over.
Testing more does not mean testing harder
A reasonable cadence is regular and varied rather than constant. Monthly or thereabouts, across different pretexts and audiences, with difficulty rising as teams improve.
What does not work is relentless testing of the same people until they fail. Repeat targeting of one individual reads as a vendetta, whatever the intention, and it produces a specific failure mode: the person stops engaging with email thoughtfully and starts treating everything as a trap, which is its own kind of damage.
Testing by exposure is different and defensible. Finance, payroll, procurement and the people who support executives are targeted more heavily in reality, so testing them more heavily reflects their risk rather than your suspicion. Finance teams in particular are worth concentrating on, because the attacks aimed at them convert directly into money.
Before your first campaign. Write the message you will send to all staff announcing the program. If you cannot write a version you would be comfortable having quoted back to you, the program design needs changing before the emails go out.
The failure mode to watch for
There is one outcome that means your testing program has made the organization less safe, and it does not appear on any dashboard.
It is the person who clicked something real, realized it, and said nothing. They did that because the culture taught them that clicking has consequences. The security team now has an active intrusion they do not know about, which is strictly worse than the click.
The NCSC's board guidance is direct about this: a culture where people are afraid to admit mistakes produces silence, and silence is what an attacker needs. So the measure of a healthy program is not a low click rate. It is whether somebody who has just made a mistake tells you about it within five minutes.
For the operational detail, see how simulations work and running simulations without damaging trust. For the training that should follow a test, awareness training that works, and for the attacker behavior you are preparing people for, the social engineering guide. Anyone who wants a second opinion on a message in the meantime can use CyberWatch AI Scan for free.
Frequently asked questions
Why test staff at all?
Because no other method answers the question. Policy documents, training completion and confidence surveys all describe intention. Only a realistic test shows what people do when something arrives.
Should employees be told that testing happens?
Yes. Tell them the program exists, what it is for and what happens if they click. Do not announce individual campaigns, because that removes the thing you are measuring.
What lures should never be used?
Anything that exploits personal distress or financial insecurity to produce a click: fake redundancy notices, fake disciplinary letters, fake bonus or pay-cut announcements aimed at individuals. They work, which is not the same as being justified.
Should a simulation capture passwords to prove the risk?
No. A simulation measures whether someone clicks. Collecting a real credential adds no information you needed and creates a liability you did not have before.
What should happen when someone clicks?
They should learn immediately: a page naming the specific tells in that email and a short lesson on that attack type. Not a message to their manager, and not their name circulated.
What is the most important number to report?
Report rate, with time to first report alongside it. Clicks tell you about exposure; reports tell you whether your organization can respond, and speed decides whether a warning arrives in time.
Is it fair to test some departments more than others?
Yes, if it follows exposure rather than suspicion. Finance, payroll and procurement are targeted more heavily in reality, so testing them more closely reflects their actual risk.
What if the results are embarrassing?
That is the normal first result and the reason to run the test. A high first-round click rate is information you did not have, and the only thing it should produce is targeted training and a second round.
Can testing make an organization less safe?
Yes, if it is punitive. People who fear consequences hide clicks and stop reporting, which closes the detection channel the security team depends on.
Sources
- Phishing attacks: defending your organisation, UK National Cyber Security Centre
- Cyber Security Toolkit for Boards: Developing a positive cyber security culture, UK National Cyber Security Centre
- SP 800-50 Rev. 1, Building a Cybersecurity and Privacy Learning Program, NIST Computer Security Resource Center
- The Phish Scale: NIST-Developed Method Helps IT Staff See Why Users Click on Fraudulent Emails, NIST
- Recognize and Report Phishing, US Cybersecurity and Infrastructure Security Agency
- Phishing Guidance: Stopping the Attack Cycle at Phase One, US Cybersecurity and Infrastructure Security Agency
- Awareness and Cyber Hygiene, European Union Agency for Cybersecurity
- Cyber Security Toolkit for Boards, UK National Cyber Security Centre


