What Sentence to Say to Check for a Stroke?

There is no single magic sentence that definitively confirms a stroke, but asking someone to repeat a simple phrase is one of the three pillars of the most widely used stroke-screening tool in the world. The standard instruction is: “Ask the person to say ‘You can’t teach an old dog new tricks’” or a similarly clear, complete sentence. If their words come out slurred, garbled, or wrong, that counts as a positive sign for stroke on the FAST test (Face, Arms, Speech, Time) and the closely related Cincinnati Prehospital Stroke Scale. The speech check is just one piece, though, and the evidence around how well it works, what it misses, and why seconds matter is worth understanding.

How the Speech Check Actually Works

The sentence “You can’t teach an old dog new tricks” became a standard not because of some special neurological property but because it is long enough to reveal problems. A person having a stroke may slur the words, swap in the wrong ones, leave out parts of the sentence, or be unable to speak at all. Any of these counts as abnormal. The Cincinnati Prehospital Stroke Scale, which paramedics have used since the late 1990s, scores speech alongside two physical signs: facial droop (ask the person to smile and check if one side sags) and arm drift (ask them to hold both arms out and watch for one drifting downward). If any one of the three is abnormal, the person should be treated as a potential stroke patient.

When untrained adults were asked to use this three-part test in a research setting, they identified speech deficits with 96% sensitivity and 96% specificity, meaning they almost never missed a real problem and almost never flagged a normal person incorrectly.¹ That makes the speech check the most reliable component for an ordinary bystander to perform. Facial droop requires a judgment call about asymmetry, and arm drift can be subtle. Abnormal speech is hard to miss.

Why Speech Breaks Down During a Stroke

A stroke cuts off blood supply to part of the brain. When the affected area handles language, the result can be dysarthria (slurred or effortful speech because the muscles of the mouth, tongue, or throat aren’t coordinating properly) or aphasia (the ability to produce or understand language itself is damaged). Both show up in roughly half of all strokes caused by a clot in the brain and a similar fraction of hemorrhagic strokes. In one large registry study, speech disturbance appeared in about 53% of clot-caused strokes and 48% of bleeding-type strokes.²

Speech problems at stroke onset also carry prognostic weight. That same study found that patients with speech disturbance at onset had significantly higher early death rates across all stroke types. For clot-caused strokes, the death rate was roughly four times higher in patients with speech trouble compared to those without it. For a rarer type of stroke caused by bleeding around the brain (subarachnoid hemorrhage), the difference was even more dramatic.² This doesn’t mean speech trouble makes the stroke worse on its own. It reflects the fact that strokes affecting the language-heavy regions of the brain tend to involve larger or more critical areas of damage.

Emerging research is even exploring whether the specific sounds a person struggles with can point to where in the brain the damage is happening. A recent study designed special sentences loaded with particular consonant types, targeting the lip muscles, tongue, and soft palate separately. The tongue-focused sentence performed best at detecting speech problems linked to specific brain lesions, with strong diagnostic accuracy.³ This kind of fine-grained analysis is still experimental, but it hints that the choice of sentence could eventually be tailored for more precise stroke screening.

How Accurate Is the Full Screening Test

The Cincinnati Prehospital Stroke Scale, taken as a whole, consistently shows high sensitivity, meaning it catches most real strokes. In a study of over 800 suspected stroke patients, the scale correctly identified strokes about 93% of the time.⁴ A multicenter study across multiple hospitals found sensitivity in a similar range.⁵ A Belgian pilot study looking at the closely related Face Arm Speech Test reported sensitivity of 95%.⁶

The trade-off is specificity. These scales cast a wide net. That same Belgian study reported specificity of only 33%, and other studies show similar patterns: the Cincinnati scale’s specificity ranged from about 25% in a Ghanaian emergency department to around 52% in the Iranian study.⁷ In practical terms, this means many people who test positive on the scale do not actually have a stroke. They might have a migraine, a seizure, low blood sugar, or one of the many conditions that can mimic stroke symptoms.

This is actually by design. In emergency medicine, a screening tool for a life-threatening, time-sensitive condition should err heavily on the side of catching real cases, even at the cost of some false alarms. A missed stroke can mean permanent disability or death. A false alarm means an unnecessary ambulance ride and some extra testing. The math clearly favors over-detection.

When the Speech Test Alone Is Not Enough

The standard FAST approach works well for strokes affecting the front-of-brain blood vessels that supply the regions controlling face, arm, and language. But about a fifth of strokes happen in the back of the brain, in what’s called the posterior circulation. These strokes can cause dizziness, loss of balance, double vision, or difficulty coordinating movements without necessarily producing the classic facial droop, arm weakness, or speech problems that FAST is built to catch.

This is a known blind spot. One review found that the standard FAST test missed about 14% of strokes, and most of those misses were posterior circulation strokes.⁸ An expanded version of the test called BE-FAST adds two more checks: Balance (sudden difficulty walking or standing) and Eyes (sudden vision changes). Adding these brought the missed stroke rate down to about 4.4%, according to the same review. Training paramedics to perform a simple finger-to-nose coordination test increased their recognition of posterior strokes by 28% and shaved 21 minutes off the time to get a brain scan.

Despite these improvements, a direct comparison of BE-FAST and FAST scores in a prehospital setting found that overall diagnostic accuracy was comparable between the two scales, with similar areas under the curve.⁹ The BE-FAST version did produce a higher negative predictive value (93% vs. 86%), meaning that when BE-FAST said someone was not having a stroke, that call was correct more often. For a bystander, the practical takeaway is: if someone suddenly cannot walk straight, has sudden vision loss, or seems uncoordinated, treat it as seriously as you would slurred speech or arm weakness, even if the face and speech seem normal.

Mild Strokes Are the Hardest to Catch

Screening tools struggle most with mild strokes. A study of patients who arrived at the hospital with very low scores on the detailed neurological exam used by doctors found that only 35% of those patients ended up with a confirmed stroke diagnosis. Among that group, the sensitivity of prehospital screening scales varied wildly, from as low as 42% for one scale up to 93% for another.¹⁰ The speech, facial droop, and arm weakness that screening tools look for were still the most common symptoms in confirmed mild stroke patients, but they appeared less reliably than in more severe strokes. Meanwhile, patients whose symptoms turned out to be something other than stroke were more likely to present with dizziness, headache, and nausea.

This matters because mild strokes are not harmless. A “minor” stroke can be a warning shot before a larger one. If you notice something off, even if the person can still talk and move both arms, the safest response is still to call emergency services and let hospital imaging sort it out.

Conditions That Mimic Stroke

One reason specificity is low on prehospital stroke scales is that several common conditions produce eerily similar symptoms. Severe low blood sugar is one of the most well-documented mimics. It can cause one-sided weakness, confusion, and even a form of spatial neglect where the person ignores one side of their visual field, exactly the pattern seen in certain strokes. A published case described a patient with hypoglycemia who, during a standard neurological exam, ignored everything on the left side of a picture and skipped the beginning of sentences when reading, perfectly imitating the pattern of a right-hemisphere stroke.¹¹ The mechanism appears to involve certain brain regions being more vulnerable to glucose deprivation than others, so when blood sugar drops severely, those regions fail first and produce focal neurological symptoms.

Other common stroke mimics include seizures (which can leave temporary one-sided weakness afterward), complex migraines (which can cause speech difficulty and numbness), and inner ear problems (which cause sudden dizziness and imbalance). For a bystander, trying to distinguish a stroke from a mimic is not your job. The screening test is designed to flag anything suspicious so that a hospital with brain imaging can make the definitive call. If someone fails the speech test or any other part of FAST, call for help regardless.

Why Every Minute Counts

The urgency behind the “T” in FAST (Time to call emergency services) is backed by striking numbers. During a typical large-vessel stroke, the brain loses roughly 1.9 million neurons every minute that blood flow is not restored.¹² Over the course of an hour without treatment, the rate of neuron loss is equivalent to about 3.6 years of normal brain aging. That research was later revisited with updated imaging data, and the updated figures confirmed a similar average rate of loss, about 2 million neurons per minute, although there was enormous variability from patient to patient, ranging from under 35,000 to over 27 million per minute depending on the location and size of the blocked vessel.¹³

That variability is important context. Some people have robust backup blood supply routes (collateral circulation) that keep brain tissue alive longer, while others have almost none. You cannot know which type you or a loved one is while sitting in a living room. The safe assumption is always that every minute matters.

Treatment windows have expanded over the past decade. The clot-dissolving drug alteplase (tPA) was originally approved for use within three hours of symptom onset and is now used up to about four and a half hours. A trial showed that alteplase guided by advanced brain imaging could benefit some patients up to nine hours after symptoms began.¹⁴ For large-vessel blockages, mechanical clot retrieval (thrombectomy) has been shown to help selected patients as far out as 24 hours, largely because some patients’ collateral circulation keeps brain tissue viable much longer than average.¹⁵ But “some patients can be treated later” should never translate to “there’s no rush.” Earlier treatment still produces better outcomes across the board, and most people will not qualify for late-window treatment.

What Happens After You Call for Help

When you call emergency services and describe stroke-like symptoms, the dispatcher typically alerts the ambulance crew, who then contact the hospital in advance. This hospital prenotification step turns out to be one of the most impactful parts of the chain. A large study of stroke patients arriving by ambulance found that when the hospital was alerted ahead of time, the door-to-imaging time dropped from 31 minutes to 26 minutes, and patients were significantly more likely to receive clot-dissolving treatment within the recommended window.¹⁶ Those five minutes might sound small, but against the backdrop of millions of neurons dying per minute, they add up.

In hospitals with highly optimized stroke workflows, the time from walking in the door to receiving the clot-dissolving drug has been pushed as low as 11 to 14 minutes.¹⁷ That kind of speed is only possible when the hospital knows the patient is coming. Calling emergency services instead of driving to the hospital yourself is not just about getting medical attention during the ride. It triggers a system that begins preparing for you before you arrive.

How Public Campaigns Have Changed Outcomes

The FAST acronym was originally designed as a public health messaging tool, and the evidence suggests it has genuinely moved the needle on stroke outcomes. In New Zealand, national FAST campaigns increased public recognition of stroke warning signs from about 72% to 76%, and awareness that stroke is time-critical jumped from 8% to 32%. Daily ambulance calls for suspected stroke rose, and the rate of patients receiving clot-dissolving treatment climbed from about 8% to nearly 10%.¹⁸ In England, a time-series analysis of the FAST campaign found significant increases in thrombolysis activity after campaign phases launched.¹⁹

These numbers reflect a real shift: more people recognized stroke symptoms, more people called for help quickly, and more patients received treatment in time. The sentence-repeating test became something ordinary people could use without training, which is exactly the point. You do not need to be a paramedic to ask someone to smile, raise both arms, and repeat a phrase. If any of those three things looks wrong, you call for help and let the professionals take over.

AI Listening on the Other End of the Line

An emerging area of development is using artificial intelligence to help emergency dispatchers identify strokes from the phone call itself. A recent study tested a system that analyzed transcripts of emergency calls, combined with patient health records, to classify whether the caller was describing a stroke. The AI model achieved strong overall performance, and its accuracy improved rapidly over the first 90 seconds of a call, reaching its best performance within about a minute and a half.²⁰ The model also generated fewer false alerts than human dispatchers operating without AI support. This technology is still being validated, but it represents a future where the system might flag a stroke even before the caller knows to describe the right symptoms, picking up on speech patterns and descriptions that trained ears and algorithms catch faster than the average person on a panicked phone call.