Why a street vendor's shout beats your headline

A hawker on a Bengaluru road has about six metres of sound, one breath and three seconds. The physics of that constraint explains why the call works, and when a long headline wins instead.

Street vendors with carts of produce on a road in Karnataka
Photo: Shagil Kannur, CC BY-SA 4.0, via Wikimedia Commons

A street vendor's shout beats your headline because the street forces it to be a perfect message: the product, a number and a reason to act now, in three to five words, repeated, with a melody that carries over traffic. The acoustics make this compulsory. A shout of about 89 dBA falls to the level of a 73 dB Bengaluru road within roughly 6 metres, so every syllable must earn its place. Headlines that face the same noise, a 1.7-second glance in a feed, should obey the same rules; headlines read by someone who has already stopped can afford to say more.

Key takeaways
  • 89 dBA is a shout at 1 m; a normal voice is about 58 dBA (Pearsons et al., 1977).
  • 73.1 dB(A) was the 2019 daytime average at Marathahalli, Bengaluru, against a 65 dB limit.
  • About 6 m is how far a shout stays above that noise, by the inverse-square law.
  • 1.7 seconds is the average time a mobile feed post gets (Facebook IQ, 2016).
  • 32,487 Upworthy tests are in the public archive; a study of its exploratory share (Banerjee and Urminsky, 2021) found that once readers stop, longer and more concrete headlines can win.

This is a collision post: two things that look unrelated, a sabziwala's call and the headline on your landing page, solved with the same equations. The working is shown, so you can check every step.

89 dBAshouted voice at 1 m (Pearsons et al. 1977)
+16.1 dBnatural loud voice over comfortable, with pitch up 8.2 semitones (Garnier et al. 2022)
1.6%average click-through rate across Upworthy headline tests
1.15 crorestreet vendors targeted by PM SVANidhi to 2030

What does a street vendor's shout actually contain?

A street vendor's call carries four slots: the product, a number (price or quantity), a quality cue and urgency. Most calls fill two or three of them in under two seconds.

The story

In November 1741 William Hogarth issued an engraving called The Enraged Musician. A professional violinist leans out of his London window with his hands over his ears. Below him the street is a wall of sound: a milk-seller balancing her pail, a ballad-seller with a crying baby, a knife-grinder with a barking dog, a dustman ringing a handbell, a sow-gelder blowing a horn and a fishmonger shouting. Art historians note that the milk-seller is the calm centre of the picture, the one "natural" musician in the noise. The joke is that the trained musician cannot be heard, while the street sellers, who had spent their lives designing calls to cut through exactly this racket, are heard perfectly well.

Wikipedia: The Enraged Musician (Hogarth, 1741)
Hogarth engraving of a violinist covering his ears at a window above a crowd of noisy London street sellers
Hogarth's The Enraged Musician (1741): a milk-seller, ballad-seller, knife-grinder and fishmonger drown out a trained violinist.Photo: William Hogarth, CC0, via Wikimedia Commons

London's cries were printed as a genre for two centuries, from John Overton's Common Cryes of London (1667) to Francis Wheatley's series of the 1790s. Historians argue that each trade developed its own melodic cry to claim an "aural space" in a crowded street, and that cries were the main way of carrying price information before mass media.

Read the captions of the prints as copy, and the structure jumps out:

  • Two bunches a penny, primroses: quantity, price, product.
  • Milk below, maids: product, location, the audience by name.
  • A modern Indian equivalent, my illustration rather than a recorded call: Tamatar, bees rupaye kilo: product, price, unit.
SlotVendor callHeadline equivalentWhy it survives noise
Product"Primroses", "tamatar"The noun the buyer searches forA known word is recognised from fragments
Number"Two bunches a penny"Price, percentage, quantityNumbers are concrete and fast to parse
Quality cue"Fresh", "garam"One concrete proofReduces risk in a single word
Urgency"Last ones", the cart moving onDeadline, stock, seasonGives a reason to act this second
Addressee"Maids", "didi"Who it is forSalient words cut through attention filters
Anatomy of a call: one breath, about 2 secondsProducttamatarNumberbees rupayeQualitytaazaUrgencyaakhri0 s2 sPitch contour: raised and sung, which is how a voice gets loud
The four slots of a vendor call. Most calls fill two or three; the melody is physics, not decoration.

How far does a street vendor's shout carry over Indian traffic?

About 6 metres on a road as loud as Marathahalli in 2019, if the listener needs the voice to be at least as loud as the noise. In a quiet lane the same shout carries about 50 metres.

Assumptions and variables

  • \( L_1 \): voice level at 1 m. Shouted speech is about 89 dBA, normal speech about 58 dBA (Pearsons, Bennett and Fidell, 1977, as summarised by Phonak).
  • \( N \): background noise. CPCB's monitoring terminals logged 2019 daytime averages of 73.1 dB(A) at Marathahalli and 72.1 at Yeshwantpur, Bengaluru (Garg et al., 2021).
  • \( S \): the signal-to-noise ratio the listener needs. I use 0 dB as a working threshold for picking out a familiar word, and test +5 and −5 dB.
  • Free-field spreading from a point source, with no walls, wind or air absorption. This is my simplification.

The worked derivation

The inverse-square law says sound intensity spreads over a sphere, so it falls with the square of distance. In decibels that is a loss of 6 dB for every doubling of distance.

  1. Level at distance r. \( L(r) = L_1 - 20\log_{10}(r / 1\,\text{m}) \).
  2. Audible while \( L(r) - N \ge S \).
  3. Solve for range. \( 20\log_{10} r \le L_1 - N - S \Rightarrow r_{max} = 10^{(L_1 - N - S)/20} \).
  4. Shout at Marathahalli. \( r_{max} = 10^{(89 - 73.1 - 0)/20} = 10^{0.795} \approx 6.2 \) m.
  5. Normal voice at Marathahalli. \( 10^{(58 - 73.1)/20} = 10^{-0.755} \approx 0.18 \) m. A normal voice is drowned before it reaches the customer's ear.
\[ r_{max} = 10^{\frac{L_1 - N - S}{20}} \]
Setting (noise)Shout, S = 0 dBShout, S = +5 dBNormal voice, S = 0 dB
Marathahalli, Bengaluru (73.1 dB)6.2 m3.5 m0.18 m
Parisar Bhawan, Bengaluru (67.2 dB)12.3 m6.9 m0.35 m
CPCB commercial limit (65 dB)15.8 m8.9 m0.45 m
Quiet residential lane (55 dB)50 m28 m1.4 m

The sensitivity is brutal. Every 6 dB of extra traffic halves the vendor's range, and every 6 dB the listener needs for clarity halves it again. That is why vendors push carts through lanes instead of standing still: the shout covers a moving circle of a few metres, and the cart sweeps it through the market.

90 dB50 dBRoad noise 73.1 dBShout, 89 dB at 1 mNormal voice, 58 dB at 1 m~6 m1 m8 m64 mDistance (log scale; each step is x8, a loss of 18 dB)
The inverse-square law against a Bengaluru road: the shout crosses the noise floor at about 6 m; a normal voice never clears it.

How far does your call carry?

Use this to check what you just read: drop the noise to 55 dB and the area covered grows about 60 times.

Why vendors sing instead of just shouting

Because raising pitch is how the voice gets loud. In a 2022 study in the Journal of the Acoustical Society of America, speakers who got loud naturally added 16.1 dB and raised their pitch by 8.2 semitones; men went from about 115 Hz to 197 Hz. When told to hold their pitch, the same speakers managed only 8.3 dB more. The sing-song pheriwala call is the body's most efficient way of buying decibels, and a tune also makes the call recognisable from a fragment.

Lip hyper-articulation in loud voice: Effect on resonance-harmonic proximityMaëva Garnier, John Smith, Joe Wolfe · Journal of the Acoustical Society of America · 2022

Eight men and women sustained vowels at comfortable and loud levels under three instructions, with sound level, pitch (from electroglottography), vocal-tract resonances and lip shape measured. Unconstrained loud voice rose 16.1 ± 0.6 dB with pitch up 8.2 semitones; holding pitch limited the gain to about 8.3 dB. The limit: a small lab sample producing sustained vowels, not running speech in a street.

Read the paper (free)
Impact of COVID-19 lockdown on ambient noise levels in seven metropolitan cities of IndiaN. Garg, V. Gandhi, N.K. Gupta · Applied Acoustics · 2021

Data from 70 permanent CPCB noise monitoring terminals, 10 in each of Bengaluru, Chennai, Delhi, Hyderabad, Kolkata, Lucknow and Mumbai, compared 2019 averages with the 2020 lockdown. In 2019 no commercial or silence-zone site met its limit, and 70% of sites sat between 60 and 75 dB(A) by day. The limit: fixed terminals measure one spot each, not the lane a vendor works.

Read the paper (free)
Daytime noise at commercial monitoring sites, 2019 annual average
See the data as a table
L day 2019 (dB(A))
ITO, Delhi75.1
CST, Mumbai74.5
Marathahalli, Bengaluru73.1
New Market, Kolkata72.9
Yeshwantpur, Bengaluru72.1
MPCB HQ, Mumbai70.2
Bandra, Mumbai67.5
Parisar Bhawan, Bengaluru67.2
Tollygunge, Kolkata66.8
CPCB HQ, Delhi66.5
Anand Vihar, Delhi65.6
WBPCB HQ, Kolkata63.4
Civil Lines, Delhi61.7

Source: Garg, Gandhi and Gupta, Applied Acoustics (2021), CPCB terminal data · The legal daytime limit for commercial areas is 65 dB(A).

Why do short calls win when attention is noisy?

Because a noisy channel carries fewer bits per second, so the message must shrink and repeat. That is Shannon's channel capacity, and it applies to a feed as much as to a road.

Claude Shannon showed in 1948 that the maximum error-free rate through a channel is \( C = B\log_2(1 + \text{SNR}) \), where B is bandwidth and SNR is the power ratio of signal to noise. You do not need the full theory, only its shape: capacity falls as noise rises, and the fix for noise is redundancy, saying less and repeating it.

  1. Quiet lane, SNR of 20 dB (a power ratio of 100): \( \log_2(101) \approx 6.66 \) bits per unit bandwidth.
  2. Busy road, SNR of 0 dB (ratio 1): \( \log_2(2) = 1 \) bit per unit bandwidth.
  3. Ratio: the same voice carries about 6.7 times less information on the busy road. A message that fits in a sentence in the lane must shrink to a phrase, said twice, on the road.
\[ \frac{C_{lane}}{C_{road}} = \frac{\log_2(1 + 100)}{\log_2(1 + 1)} \approx 6.7 \]

The feed has its own noise floor. Facebook reported in 2016 that people spend on average 1.7 seconds with a piece of content on mobile, against 2.5 seconds on desktop. At an assumed reading speed of about 4 words a second, which is my round figure, 1.7 seconds is roughly 7 words. A scrolling reader is a pedestrian on Marathahalli road.

A scrolling reader is a pedestrian on a 73-decibel road: you get one breath, so spend it on the product, a number and a reason to stop.Applying the inverse-square and Shannon results above

The cocktail-party filter

The brain helps the vendor in one way. In dichotic-listening experiments, about a third of people hear their own name in a message they are trying to ignore: 33% in Neville Moray's 1959 study and 34.6% in Wood and Cowan's 1995 replication. Highly salient words leak through the attention filter. For a buyer walking to the market, tamatar or aloo is that salient word, which is why the product noun goes first.

The cocktail party phenomenon revisited: The importance of working memory capacityAndrew R.A. Conway, Nelson Cowan, Michael F. Bunting · Psychonomic Bulletin & Review · 2001

Forty undergraduates, split into high and low working-memory span groups, shadowed one spoken message while their own first name was played in the ignored ear. 65% of low-span listeners noticed their name against 20% of high-span listeners, and the paper confirms the earlier 33% and 34.6% rates. The limit: a lab task with headphones, not a street, and a name rather than a product word.

Read the paper (free)
Eighteenth-century print of a London flower seller with the caption Two bunches a penny primroses
"Two bunches a penny, primroses": a print from the Cries of London series. Quantity, price and product, in five words.Photo: Rijksmuseum, CC0, via Wikimedia Commons

When does a long headline beat the vendor's shout?

When the reader has already stopped. The largest set of real headline experiments found that, on average, longer and more concrete headlines got more clicks, which is the opposite of the shout rule.

Upworthy randomly tested packages of headline and image on its own site from January 2013 to April 2015. Researchers later released the archive: 32,487 tests (a later analysis counts 32,488), more than 150,000 packages, 538 million impressions and over 8 million clicks. The median test had 4 packages and 14,342 impressions.

The Language That Drives Engagement: A Systematic Large-scale Analysis of Headline ExperimentsAkshina Banerjee, Oleg Urminsky · University of Chicago working paper (SSRN) · 2021

Using the exploratory share of the Upworthy archive, they coded 36 language constructs with LIWC and other NLP tools and modelled within-test click-through differences. Average click-through was 1.6%, and most effects moved it by about ±3.1% per standard deviation. Higher word count, longer sentences, numbers and concreteness raised clicks; headlines addressed to "you" backfired. Only 7 of 21 literature-based predictions came out in the predicted direction. The limit: preliminary results on one site's audience, flagged by the authors as likely to change.

Read the paper (free)

This does not break the collision; it locates it. The shout is a first-contact message in a noisy channel. The Upworthy headline was read by someone who had already landed on a page and was choosing between stories, a quieter channel with more capacity. Two rules survive in both places: numbers and concrete nouns win everywhere, and talking at the reader ("you won't believe") loses everywhere.

Does rhyme or rhythm make a call more believable?

Yes, measurably. In a 2000 study in Psychological Science, Matthew McGlone and Jessica Tofighbakhsh found that people rated rhyming sayings as more accurate than the same sayings reworded without the rhyme. They called it the rhyme-as-reason effect: rhyme makes a statement easier to process, and the ease is mistaken for truth. The effect shrank when people were warned to separate form from meaning.

A vendor's chant and a jingle both exploit this processing fluency. It is also a warning: fluency persuades whether or not the claim is true, which is why the rule in the cost of AI slop ads applies here too. Make the rhythm carry a true number.

Engraving of an oyster seller carrying a basket from Laroon's Cryes of the City of London
An oyster seller from Marcellus Laroon's The Cryes of the City of London. Each trade owned a call, the original brand sound.Photo: Rijksmuseum, CC0, via Wikimedia Commons

How do you write a headline like a street vendor?

Fill the slots in order of salience and cut everything else. Here is the method I use, step by step.

  1. Product noun first. The word the buyer would search or scan for. Not your brand line.
  2. One number. Price, quantity, time or percentage. Numbers raised clicks in the Upworthy tests.
  3. One concrete proof. An invented example, not a real seller: Taaza, "made in Pune", "4.6 stars from 2,100 buyers". Concreteness also raised clicks.
  4. A reason for now. A real deadline or stock limit, never a fake one.
  5. Say it twice, in two forms. The thumbnail and the first line, or the hoarding and the jingle. Repetition is how noisy channels deliver a message.
  6. Match length to noise. Fewer than about 7 words where the glance is 1.7 seconds; more where the reader has already stopped.
ChannelNoise level (my rating)Words to aim forExample shape
Hoarding seen from trafficVery high3 to 6Product + number
Reel first frame, thumbnailHigh4 to 7Product + number + hook
Push or WhatsApp previewHigh5 to 8Product + deadline
Landing page H1Medium8 to 14Outcome + number + proof
Article headline in a list of storiesLower10 to 15Concrete, specific story

The same logic explains why Reels ranking rewards the first second so heavily, and why creative fatigue sets in: a call heard too often stops being salient.

So, does the shout really beat your headline?

Verdict: possible with changes

The shout wins every first contact in a noisy channel. By the inverse-square law an 89 dBA call stays above 73 dB of Bengaluru traffic for only about 6 m, and Shannon's capacity falls about 6.7 times between a quiet lane and a loud road, so the message must shrink to product, number and urgency. Once a reader has stopped, the channel is quieter and the exploratory share of the Upworthy archive (32,487 tests in all) shows longer, concrete headlines can win. Use the vendor's rules for thumbnails, hoardings and first frames; use concrete detail for the page.

The vendor also has an unfair advantage most brands forget: he is standing there. The 1.15 crore vendors targeted by PM SVANidhi sell face to face, with the product in view. Your headline has to do the work of the shout, the cart and the tomato all at once, which is a good reason to put a picture of the tomato next to it.

What if next?

Questions people ask

How many words should a headline have?

It depends on the noise. Where the glance is about 1.7 seconds (feeds, thumbnails, hoardings), aim for 3 to 7 words with a product and a number. On a page where the reader has stopped, the Upworthy data found longer, more concrete headlines did better on average.

Why do street vendors call in a sing-song voice?

Raising pitch is how the voice gets loud efficiently. Lab data show a natural loud voice adds about 16 dB with pitch up about 8 semitones, while loudness at a fixed pitch adds only about 8 dB. A tune is also recognisable from a fragment.

How loud is a shout?

About 89 dBA at 1 metre, against about 58 dBA for normal speech, according to the US EPA study by Pearsons, Bennett and Fidell (1977).

Do numbers in headlines increase clicks?

In Banerjee and Urminsky's analysis of Upworthy's randomised tests, the presence of numbers raised click-through rates, as did concrete language. Headlines that addressed the reader directly did worse.

Sources

  1. Phonak Audiology Blog: Revisiting expectations for average and soft speech levels (summarises Pearsons, Bennett and Fidell, US EPA 1977; source for 58 dBA normal and 89 dBA shouted speech)
  2. Garnier, Smith and Wolfe, JASA: Lip hyper-articulation in loud voice (2022; source for +16.1 dB, +8.2 semitones and the 8.3 dB pitch-held figure)
  3. Garg, Gandhi and Gupta, Applied Acoustics: Impact of COVID-19 lockdown on ambient noise levels in seven metropolitan cities of India (2021; source for 2019 site noise levels and compliance)
  4. Delhi Pollution Control Committee: Ambient noise standards (source for the 65 dB(A) commercial daytime limit)
  5. Conway, Cowan and Bunting, Psychonomic Bulletin & Review: The cocktail party phenomenon revisited (2001; source for 33%, 34.6%, 65% and 20%)
  6. Banerjee and Urminsky: The Language That Drives Engagement (2021; source for 1.6% CTR and headline-feature effects)
  7. McGlone and Tofighbakhsh, Psychological Science: Birds of a feather flock conjointly (?): rhyme as reason in aphorisms (2000; source for the rhyme-as-reason effect)
  8. Facebook IQ: Capturing attention in feed (April 2016; source for 1.7 s on mobile and 2.5 s on desktop)
  9. Wikipedia: The Enraged Musician (source for the 1741 print and its street sellers)
  10. Wikipedia: Street cries (source for the London cries publications and melodic cries)
  11. Wikipedia: Noisy-channel coding theorem (source for Shannon's capacity formula)
  12. Business Today: Cabinet extends PM SVANidhi scheme till 2030 (Aug 2025; source for the 1.15 crore vendors and ₹7,332 crore outlay)
  13. Tubefilter: YouTube expands Test & Compare to titles (Jul 2025)
  14. arXiv 2405.01165: How A/B testing changes the dynamics of information spreading on a social network (2024; source for 32,488 tests, over 150 thousand packages, 538 million impressions, over 8 million clicks, median 4 packages and 14,342 impressions per test)
  15. Upworthy Research Archive: Data in the archive (source for 32,487 tests and the date range)
street vendorheadline writingcopywritingsignal-to-noise ratioShannon capacityUpworthyrhyme as reasonstreet cries
Sapta

Sapta (Saptarshi Sarkar) is a mechanical engineer turned marketer in Bengaluru and a solo touring rider. He writes the seven Saptaverse blogs. Work with him or read more of Marketblog.

Keep reading in Marketblog

From the other blogs