Copywriting for voice interfaces and audio advertising
What you really encountered: voice search or audio format?
The first question you need to ask yourself: does the user read the answer out loud (for example, asks Alice for a recipe) or listen to a ready-made audio clip (in a podcast, on the radio, in an app)? These are two fundamentally different scenarios.
In voice search, an algorithm works: the system searches for a short, structured answer and voices it. In audio advertising, the listener did not ask a question — you interrupt their attention and must hold it without visual cues.
| Parameter | Voice interface | Audio advertising / radio |
|---|---|---|
| Who initiates the contact | User (asks a question) | Brand (interrupts the broadcast) |
| Content source | Featured snippet, FAQ, Schema.org | Written script, announcer / AI voice |
| Optimal length | 29 words, 10-15 seconds | 45-75 words per 30-45 seconds |
| The goal of the text | To give an accurate answer | To evoke an emotion and embed the brand in memory |
| Success metric | To appear in the zero position | Memorability, conversion via promo code |
Why voice assistants choose this particular text
Alice, Siri, and Google Assistant do not 'understand' the brand — they extract a ready-made fragment from structured data. 30-50% of voice responses are taken from the zero position of the search results, which means the content must be formatted in such a way that the algorithm can extract it entirely, without editing.
- The answer is formulated in FAQ or HowTo markup format — assistants primarily scan these blocks first.
- Schema.org microdata is not an option, but a mandatory condition for visibility in voice search.
- The answer formulation should mirror the structure of the question: voice queries are two to three times longer than text ones and sound like natural speech — 'what is the weather in Moscow today', not 'weather Moscow'.
- Optimal volume — 20-40 words, which fits into the 29-word 'golden' length that can be delivered in 10-15 seconds without losing the listener's attention.
There are differences between platforms as well: Alice tends to rely more on structured answers from Yandex and works less effectively with long, complex subordinate phrases, Siri requires stricter matching of question wording, and Google Assistant actively uses featured snippets from web search. Writing one universal text for all three systems does not work — you need to check how each of them actually voices your content.
How to make the listener remember the brand without an image
In audio, there is no visual anchor, so the entire load falls on the sound: pause, intonation, rhythm. These are sound anchors — recurring acoustic elements that the brain begins to associate with the brand even before the company name is heard.
A pause before the key message works like italics in text — it physically forces the listener to focus. Changing the speech tempo (slowing down on an important phrase) highlights it from the flow just as effectively as bold font. And the rhythmic pattern of the phrase — the very one that makes jingles catchy — gets embedded in memory even without conscious word memorization.
The McGurk effect is also evident here: brain perception of sound is distorted depending on context and expectations. In short audio formats, this works in the brand's favor — if the announcer's intonation matches the expected tone of voice, the brain 'fills in' missing semantic details on its own, and the message is processed faster and more accurately than when the same words are spoken indifferently by a neutral voice.
How much text really fits in 30 seconds
The main mistake of audio scriptwriters is writing text designed for reading with the eyes, not for speaking out loud. The standard speech rate of a narrator is 140-160 words per minute, and this limit cannot be cheated with emotional delivery.
| Timing | Text volume |
|---|---|
| 15 seconds | 25-30 words |
| 30 seconds | 45-55 words |
| 45 seconds | 65-75 words |
Inside this volume, the rule of three repetitions applies: the brand name must be heard at least three times — at the beginning, middle, and end of the ad. Without this, the name gets lost in the general flow of information and does not stick in memory. The second practical rule: the text must be spoken in one breath. If the narrator runs out of air and needs a second inhale in the middle of a phrase — the phrase is too long and must be rewritten, not just comma-separated. Checking aloud before recording is not a formality but a mandatory step: what reads smoothly with the eyes often stumbles on the tongue.
How the same brand sounds on different platforms
The tone and structure of the text change depending on the platform, and one script cannot be copied everywhere.
| Platform | Text feature |
|---|---|
| Podcast (native integration) | Personal speech of the host, conversational tone, story instead of a slogan — converts 30-50% better than a preroll |
| Online radio / preroll | Strict timing, three brand repetitions, clear call-to-action with promo code |
| IVR (voice menu) | Minimum emotion, maximum clarity, short commands, no metaphors |
| Voice assistants | Question-answer structure, FAQ format, alignment with Schema.org |
It is worth separately noting the brand persona — how it would 'speak' if it were a person: the voice age, tempo, humor or its absence, formality of address. It is precisely the persona that determines the tone of voice in all audio formats at once, and without it, sonic identity becomes a set of fragmented clips rather than a recognizable company voice. Digital audio in Russia showed growth in 2024 year according to sellers' forecasts precisely because brands began to treat sound as a system rather than one-time purchases of air time.
Typical mistakes and what actually works
Practice shows that failures are almost always of the same type:
- The text is written for reading, not for pronunciation — long subordinate clauses that cannot be spoken in one breath.
- The brand name is mentioned only once at the very end — the listener simply will not remember it.
- One script is used simultaneously for the voice assistant and for the radio spot — the SEO effect and emotional impact are both lost.
- The lack of personalized promo codes — the user's path from the airwaves to purchase is almost always lost, and attribution is impossible to calculate without a unique identifier for each channel.
Regarding the choice of tools: a jingle is good for long-term memorization and instant brand identification in the stream of airwaves, but it does not work well for conversion at this very moment. Native integration with a podcaster is more convincing than any pre-roll — due to trust in the host's personality, but it does not scale well and requires an individual script for each blogger. Sound effects (signal, click, a distinctive product sound) — the most underrated tool: they are cheaper than a jingle, quicker to read by ear, and work well precisely in IVR and applications where instant recognition without words is needed. The trend toward AI voiceover in 2024 year removed the budget issue for small businesses — the quality of synthesized voice approached live announcer recording closely, and the difference is not heard by all audiences.
What you really encountered: voice search or audio format?
The first question you need to ask yourself: does the user read the answer out loud (for example, asks Alice for a recipe) or listen to a ready-made audio clip (in a podcast, on the radio, in an app)? These are two fundamentally different scenarios.
In voice search, an algorithm works: the system searches for a short, structured answer and voices it. In audio advertising, the listener did not ask a question — you interrupt their attention and must hold it without visual cues.
| Parameter | Voice interface | Audio advertising / radio |
|---|---|---|
| Who initiates the contact | User (asks a question) | Brand (interrupts the broadcast) |
| Content source | Featured snippet, FAQ, Schema.org | Written script, announcer / AI voice |
| Optimal length | 29 words, 10-15 seconds | 45-75 words per 30-45 seconds |
| The goal of the text | To give an accurate answer | To evoke an emotion and embed the brand in memory |
| Success metric | To appear in the zero position | Memorability, conversion via promo code |
Why voice assistants choose this particular text
Alice, Siri, and Google Assistant do not 'understand' the brand — they extract a ready-made fragment from structured data. 30-50% of voice responses are taken from the zero position of the search results, which means the content must be formatted in such a way that the algorithm can extract it entirely, without editing.
- The answer is formulated in FAQ or HowTo markup format — assistants primarily scan these blocks first.
- Schema.org microdata is not an option, but a mandatory condition for visibility in voice search.
- The answer formulation should mirror the structure of the question: voice queries are two to three times longer than text ones and sound like natural speech — 'what is the weather in Moscow today', not 'weather Moscow'.
- Optimal volume — 20-40 words, which fits into the 29-word 'golden' length that can be delivered in 10-15 seconds without losing the listener's attention.
There are differences between platforms as well: Alice tends to rely more on structured answers from Yandex and works less effectively with long, complex subordinate phrases, Siri requires stricter matching of question wording, and Google Assistant actively uses featured snippets from web search. Writing one universal text for all three systems does not work — you need to check how each of them actually voices your content.
How to make the listener remember the brand without an image
In audio, there is no visual anchor, so the entire load falls on the sound: pause, intonation, rhythm. These are sound anchors — recurring acoustic elements that the brain begins to associate with the brand even before the company name is heard.
A pause before the key message works like italics in text — it physically forces the listener to focus. Changing the speech tempo (slowing down on an important phrase) highlights it from the flow just as effectively as bold font. And the rhythmic pattern of the phrase — the very one that makes jingles catchy — gets embedded in memory even without conscious word memorization.
The McGurk effect is also evident here: brain perception of sound is distorted depending on context and expectations. In short audio formats, this works in the brand's favor — if the announcer's intonation matches the expected tone of voice, the brain 'fills in' missing semantic details on its own, and the message is processed faster and more accurately than when the same words are spoken indifferently by a neutral voice.
How much text really fits in 30 seconds
The main mistake of audio scriptwriters is writing text designed for reading with the eyes, not for speaking out loud. The standard speech rate of a narrator is 140-160 words per minute, and this limit cannot be cheated with emotional delivery.
| Timing | Text volume |
|---|---|
| 15 seconds | 25-30 words |
| 30 seconds | 45-55 words |
| 45 seconds | 65-75 words |
Inside this volume, the rule of three repetitions applies: the brand name must be heard at least three times — at the beginning, middle, and end of the ad. Without this, the name gets lost in the general flow of information and does not stick in memory. The second practical rule: the text must be spoken in one breath. If the narrator runs out of air and needs a second inhale in the middle of a phrase — the phrase is too long and must be rewritten, not just comma-separated. Checking aloud before recording is not a formality but a mandatory step: what reads smoothly with the eyes often stumbles on the tongue.
How the same brand sounds on different platforms
The tone and structure of the text change depending on the platform, and one script cannot be copied everywhere.
| Platform | Text feature |
|---|---|
| Podcast (native integration) | Personal speech of the host, conversational tone, story instead of a slogan — converts 30-50% better than a preroll |
| Online radio / preroll | Strict timing, three brand repetitions, clear call-to-action with promo code |
| IVR (voice menu) | Minimum emotion, maximum clarity, short commands, no metaphors |
| Voice assistants | Question-answer structure, FAQ format, alignment with Schema.org |
It is worth separately noting the brand persona — how it would 'speak' if it were a person: the voice age, tempo, humor or its absence, formality of address. It is precisely the persona that determines the tone of voice in all audio formats at once, and without it, sonic identity becomes a set of fragmented clips rather than a recognizable company voice. Digital audio in Russia showed growth in 2024 year according to sellers' forecasts precisely because brands began to treat sound as a system rather than one-time purchases of air time.
Typical mistakes and what actually works
Practice shows that failures are almost always of the same type:
- The text is written for reading, not for pronunciation — long subordinate clauses that cannot be spoken in one breath.
- The brand name is mentioned only once at the very end — the listener simply will not remember it.
- One script is used simultaneously for the voice assistant and for the radio spot — the SEO effect and emotional impact are both lost.
- The lack of personalized promo codes — the user's path from the airwaves to purchase is almost always lost, and attribution is impossible to calculate without a unique identifier for each channel.
Regarding the choice of tools: a jingle is good for long-term memorization and instant brand identification in the stream of airwaves, but it does not work well for conversion at this very moment. Native integration with a podcaster is more convincing than any pre-roll — due to trust in the host's personality, but it does not scale well and requires an individual script for each blogger. Sound effects (signal, click, a distinctive product sound) — the most underrated tool: they are cheaper than a jingle, quicker to read by ear, and work well precisely in IVR and applications where instant recognition without words is needed. The trend toward AI voiceover in 2024 year removed the budget issue for small businesses — the quality of synthesized voice approached live announcer recording closely, and the difference is not heard by all audiences.
Recent articles
MegaFon, BMW, VTB and more 240+ companies. See how it looks in the showcase reel:
Our branding agency is approached to order:
Branding agency services
