The phrase "spear phishing can often be identified via bad grammar" is useful, at most, for initial triage. For a CISO, IT, or compliance team in LATAM, the evidence shows that treating it as a detection rule or a core awareness criterion gives an incomplete view of human risk, because NIST asks organizations to measure message difficulty, clicks, reports, and reporting speed, while CISA, the FTC, and educational guides still keep grammar as one clue among others. In an empirical study of AI-generated and human-written spear phishing SMS messages, arXiv found that 24% of participants, 6 of 25, mentioned grammar and spelling when evaluating the messages.
Placed side by side, the sources show something that does not appear clearly in any single one of them: grammar survives in awareness materials, but the more operational sources downgrade it and turn it into a secondary signal inside a measurement and response framework.
| Source | Treatment of grammar | Other signals | What to measure | Expected action | Program implication | Source |
|---|---|---|---|---|---|---|
| CISA | Bad grammar and spelling mistakes are a traditional phishing signal, but in the age of artificial intelligence some emails can have perfect grammar and spelling, according to CISA. | CISA also recommends looking for other signals. | not specified | Do not interact with suspicious messages and report the phishing, according to CISA. | Grammar works as a triage clue, not as the only detection or decision criterion. | CISA |
| FTC | The FTC infographic lists bad grammar and spelling as an indicator. | The FTC also notes that the message may not use the recipient's name, may request personal information or passwords, and recommends not using the included link and contacting the company through a known channel. | not specified | Report the message and delete it if scam signs are identified, according to the FTC. | The useful decision is not only spotting errors, but escalating and verifying through an independent channel. | Federal Trade Commission |
| Illinois Department of Human Services | The Illinois guide includes grammar and spelling errors as a possible phishing signal. | The same guide calls for checking the sender, external domain, links that do not match their destination, attachments out of context, and emotionally urgent language. | not specified | Review multiple signals in the message. | Even in educational material that keeps grammar as a clue, the real evaluation is already multifactor. | Illinois Department of Human Services |
| HP | Bad grammar or awkward wording is no longer enough to determine whether a message is legitimate, according to HP. | not specified | not specified | not specified | A control based on detecting grammar errors loses power as the main defense against spear phishing. | HP |
| Canadian Centre for Cyber Security | AI is making it harder to detect traditional phishing features, such as bad spelling or a robotic tone, according to the Canadian Centre for Cyber Security. | Spear phishing can include the victim's interests, recent activities, or purchases, according to the Canadian guide. | not specified | not specified | Personalization and AI reduce the value of traditional language signals as the main detector. | Canadian Centre for Cyber Security |
| NIST | The NIST Phish Scale says to record grammar, spelling, punctuation, and agreement errors when evaluating a simulated email. | NIST uses these elements as part of a measure of email difficulty that provides context for click rate and report rate. | Click rate, report rate, and reporting time, plus segmentation by group or role, according to NIST. | Report the suspicious message quickly and share results in aggregate, according to NIST. | Grammar is one variable within the human difficulty of the message, not the program's success metric. | National Institute of Standards and Technology |
| ISO/IEC 27001:2022 clause 7.3 | Clause 7.3 does not prescribe detecting phishing through grammar errors, according to ISO/IEC 27001:2022 and ISACA Germany's implementation guide. | not specified | not specified | Requires awareness of the security policy, the contribution to the effectiveness of the ISMS, and the consequences of not meeting the requirements, according to ISO/IEC 27001:2022 clause 7.3. | ISO compliance does not require grammar spotting as a method or a specific course or platform. | ISO/IEC 27001:2022 |
Does bad grammar still count as a valid signal, or has it stopped working?
CISA says bad grammar and spelling mistakes are still a traditional phishing signal, but warns that in the era of artificial intelligence, some emails can have perfect grammar and spelling. HP goes a step further and says bad grammar or awkward wording is no longer enough to determine whether a message is legitimate.
Whalemate's reading is that there is no full contradiction, only a change in priority. Grammar is still useful as a cheap and quick clue, but it no longer has value as the main rule for detecting spear phishing. If a program still rewards only the person who spots wording errors, it is measuring a partial skill that can fail precisely against the most dangerous messages.
What changes when the attacker uses AI and writes better than before?
The Canadian Centre for Cyber Security warns that AI is making it harder to detect traditional phishing traits such as bad spelling or a robotic tone. CISA agrees that some emails can have perfect grammar. At the same time, a pilot study on spear smishing in arXiv found that some participants even read overly perfect grammar as a sign of AI-generated content.
Whalemate's reading is that style-based signals have become ambiguous. A message can sound bad and be malicious, but it can also sound polished and be malicious. Worse still, according to the arXiv study, stylistic perfection also does not reliably show whether the source was human or artificial. For operational defense, that pushes organizations toward measuring observable behavior instead of intuitions about writing style.
If NIST records grammar errors, why not use them as the main KPI?
The NIST Phish Scale says to record grammar errors, spelling errors, punctuation errors, and agreement errors, including in headers, body text, and subject lines. But NIST also says the scale is meant to give context to click rate and report rate through a measure of the human difficulty of the message.
Whalemate's reading is that NIST does not turn grammar into the goal of the program, but into a design and classification variable for the lure. It helps estimate how detectable a simulated email was, not decide whether the user "learned" simply because they noticed a spelling mistake.
What do educational guides still include grammar as a clue?
The FTC lists bad grammar and spelling as an indicator, but also says the message may not use the recipient's name or may ask for personal information or passwords. The Illinois state guide does something similar: it keeps grammar as a possible signal, but also requires checking the sender, external domain, links that do not match their destination, attachments out of context, and urgent language.
Whalemate's reading is that even where advice like "look at the grammar" remains, the real evaluation is already multifactor. That matters because many organizations keep only the simplest part of the message and not the full requirement to verify identity, context, urgency, links, and attachments.
Why does measuring only click rate give a poor picture of risk?
NIST says many programs use click rate as the only metric, even though the Phish Scale adds a measure of email difficulty to explain why results vary across exercises. NIST also recommends looking at how many users report a message and how quickly they report it. The document on human difficulty in phishing adds that reporting rates and reporting times matter because early reporting can improve mitigation.
Whalemate's reading is that an isolated click rate blends too many things together. It can reflect worse behavior, but also a more difficult lure or a more credible context. Without difficulty, reporting, and reporting time, the organization does not know whether the problem is human, related to exercise design, or due to a lack of an operational channel.
What goes wrong when the simulation is too obvious?
A vendor guide, Qalea, argues that a simulation with an obviously fake email, bad grammar, and an unknown sender measures very little. NIST does not frame the issue in those exact terms, but it does build the Phish Scale specifically to contextualize difficulty and avoid treating different exercises as comparable by default.
Whalemate's reading is that both layers converge. If the lure is so crude that it almost invites the user to spot the trap through wording alone, the organization gets a reassuring metric with low utility. That kind of simulation can inflate the appearance of maturity and leave untested what happens with personalized, polished emails.



