The phonological loop holds speech for about two seconds. How the Baddeley model works, the three effects that prove it, and why talk erases numbers.
You are holding eight digits. Someone asks you a question — a small one, nothing demanding — and you answer it in words. When you turn back, the number is gone. Not faded. Gone, the way a word vanishes when you look it up and forget it on the way back to the sentence.
That gap is not a character flaw, and it is not early memory loss. It is a specific component of working memory doing what it is built to do, on a timer you can measure. The phonological loop holds speech in your head for about two seconds at a stretch. Most of what feels like "my memory is bad" about short strings — phone numbers, Wi-Fi passwords, the name you were introduced to nine seconds ago — is this buffer running out.
What follows is the mechanism, the model it belongs to, the three effects that prove it exists, and why answering a question deletes the number you were repeating.
What the Phonological Loop Actually Is
Take the name seriously. This is the part of working memory that deals in sounds, not meanings and not pictures.
The American Psychological Association's dictionary defines it as a component that holds and manipulates auditory information over short intervals, built from two halves. One is the phonological store, which keeps memory traces for about two seconds before they fade. The other is an articulatory control process that refreshes those traces by subvocal rehearsal — you re-speak the item silently, over and over. The APA entry offers a pair of images worth keeping: the store is an "inner ear" that holds speech sounds in the order they arrived, and the rehearsal process is an "inner voice" that repeats them on a loop.
That is where the name comes from. Not a metaphor for repetition, but a literal loop between a voice and an ear, both inside your head.
A 2013 review in Frontiers in Psychology by Bradley Buchsbaum adds a detail the dictionary leaves out: the store holds speech-based information for approximately two seconds per item before decay, and the two halves are mutually dependent. The store has capacity but no way to reactivate itself. The rehearsal process can reactivate but has no storage of its own. Damage either half and the system degrades.
That mutual dependence explains how the loop fails. It is not a bucket that fills up. It is a tape that has to keep being played.
Where It Sits in the Baddeley and Hitch Working Memory Model
The loop did not arrive alone. Alan Baddeley and Graham Hitch proposed the working memory model in 1974, and the loop was one of its three original components, alongside the central executive and the visuospatial sketchpad.
The model came out of patient work, not introspection. In a 2021 retrospective in Archives of Clinical Neuropsychology, Baddeley describes the evidence that forced it: patients with classic amnesia had intact short-term memory, while other patients showed the opposite pattern — badly impaired short-term memory with long-term memory preserved. Two opposite deficits are hard to explain with a single store. When healthy participants were made to show the same short-term deficit using a dual task, the result looked like a system with an overall attentional controller and separate temporary buffers for speech and for space.
A fourth component came later. The episodic buffer was added to bind information from several sources into one representation that conscious awareness can use.
| Component | Job | What it feels like from the inside |
|---|---|---|
| Central executive | Allocates attention, switches tasks, picks strategies | The effortful part; why N-Back is tiring |
| Phonological loop | Holds and rehearses speech | Your inner voice reading this sentence |
| Visuospatial sketchpad | Holds shape, location and movement | Mentally rotating a key to fit a lock |
| Episodic buffer | Binds sources into one scene | Recalling where you left your keys, as a picture |
BrainFacts, the Society for Neuroscience's public site, describes the central executive as a kind of company boss deciding what deserves attention, with the loop as the assistant handling anything spoken. The image is loose, but the division of labour matches what the clinical evidence produced.
Working memory models have multiplied since 1974 and they disagree about plenty. Almost all of them keep some version of the loop.
The Three Effects That Prove the Loop Exists
A model earns its keep by predicting what happens when you break it. Three findings do that for the loop, and all three are things you can feel in under a minute.
The word-length effect. Sets of words that take longer to say are recalled worse than sets of short words, a result Baddeley and colleagues reported in 1975 and which Buchsbaum's review treats as a cornerstone. Try six items: cat, hat, map, pen, cup, bell. Now try elephant, university, alphabet, hospital, computer, butterfly. Same number of items, same task. The second list collapses faster, because capacity is measured in time rather than in slots — long words eat the two seconds before you finish the list.
The phonological similarity effect. Items that sound alike interfere with each other. Cat, cap, map, mat, bat is much harder to hold than cow, day, pin, leaf, stone, even though both are five short words. The store encodes sound, so rhyming items arrive as near-duplicates and get confused on the way back out.
Articulatory suppression. Ask someone to repeat an irrelevant word — "hiya, hiya, hiya" — while they memorise, and recall drops sharply. The inner voice is busy saying something else, so nothing refreshes the store. This is the closest thing to direct proof: occupy the rehearsal process and the buffer empties on schedule.
The suppression work also produced a subtler result. Blocking rehearsal wipes out the similarity effect when words are shown visually, but not when they are heard. If the store were merely a by-product of silent speech, that difference should not exist. It points to a store that holds an auditory pattern, with rehearsal as the maintenance crew rather than the store itself.
Why Talking to Someone Erases the Number
The usual explanation is "I got distracted". That undersells what happened. A distraction is a competing stimulus. What killed the number is competition for the same two resources, both of which are speech.
Answering a question in words recruits the articulatory process. For the two or three seconds you spend forming a sentence, nothing refreshes the digits. The store does what it always does — decays — and by the time you turn back, the trace sits below the level where retrieval can find it. A visual distraction is a different case. It occupies the sketchpad, so it breaks concentration without competing for the same buffer.
Two things follow, and only two. First, articulation time is the budget, so the way to hold more digits is to say them in fewer syllables: chunking eight digits into four pairs shortens the spoken string and buys a little room. Second, the loss is time-based, so the value of rehearsal drops the longer you wait to use the number. Repeating it for thirty seconds before you dial does not build a stronger trace. It keeps a weak one alive.
Everything else about short-term memory — when to write it down, when to trust it, what to practise — follows from those two facts rather than from willpower.
Reading Your Own Score in Memory Games
Most people who train working memory never find out which part they are training. The drills differ, and so do the failures.
N-Back asks you to keep a stream of items in mind and judge whether the current one matches the one n steps back. The load is continuous updating, which is central-executive work. If your accuracy collapses when the rule changes but holds when the items get longer, you are watching the executive, not the loop. N-Back is free to try here.
Sequence memory with Simon-style patterns is mainly the sketchpad's task. Colours and positions arrive as a spatial sequence, and the loop contributes little unless you talk your way through it. A good score says nothing about your digit span. The sequence memory game is the one to run if spatial span is what you want to read.
Number sequence memory comes closest to a loop test: digits, order, immediate recall. This drill measures the buffer itself.
The diagnostic rule is simple. Longer items hurt you → the loop. Changing rules hurt you → the executive. Spatial patterns hurt you → the sketchpad. Three different failures, three different systems, and no single "memory score" that covers all of them.
If you want the wider picture of what training changes and what it does not, the working memory improvement guide covers the exercise evidence in more detail.
One caution is worth stating plainly, because the marketing around brain games rarely does. A 2016 review in Psychological Science in the Public Interest by Daniel Simons and six co-authors examined the studies cited by leading brain-training companies. They found extensive evidence that training improves performance on the trained task, less evidence for closely related tasks, and little evidence that it transfers to distantly related tasks or to everyday cognition. Practice makes you better at the thing you practise. That is real, and it is also the limit.
Using the Loop Instead of Fighting It
The loop is not a defect to be trained away. It is a two-second buffer that keeps speech available long enough for you to act on it, and it does that job well as long as you stop asking it to be a notebook.
Three habits follow from the mechanism:
- Count syllables, not items. When something has to survive in your head, choose the shorter version of it. Six short words beat four long ones.
- Say it once, cleanly, at the moment you receive it. Encoding is the part you can influence. The wait afterwards is mostly decay management.
- Move anything that matters out of the loop. A number you need in five minutes belongs on paper. A number you need in five seconds belongs in your inner voice, in as few syllables as possible.
You can watch the mechanism work on yourself in about a minute. Read six short words, look away, count backwards from twenty for ten seconds, then write down what you remember. Repeat with six long words. Most people lose two or three more items from the long list. That gap is the word-length effect, produced in your own head, with no equipment and no test fee.
Your inner voice is not failing you. It is a short tape loop running on a two-second timer, doing precisely what it is built to do. The skill is knowing when to stop asking it for more — and when to reach for a pen instead.
References
- APA Dictionary of Psychology — phonological loop
- Buchsbaum (2013), Frontiers in Psychology — The role of consciousness in the phonological loop
- Baddeley (2021), Archives of Clinical Neuropsychology — Developing the Concept of Working Memory
- BrainFacts / Society for Neuroscience — Sketch of a Memory
- Simons et al. (2016), Psychological Science in the Public Interest — Do "Brain-Training" Programs Work?
