Synthetic audiences:
test sound concepts before they sound
A sound logo is usually only tested once it already sounds finished. By then composition and production are paid for, a panel has to be recruited, and every extra variant costs more. If the direction fails, the work starts over. AI-simulated audiences move that question forward, to the point where nothing has been produced yet.
The short version
Synthetic audiences are simulated respondents that an AI model builds from the characteristics of a real audience. Research shows they often get averages right and underestimate how much real people vary. In sonic branding we therefore use them for direction decisions before production, and for big launches we test the finalists with real people afterwards.
Contents
1. The problem with the late test
A classic pre-test measures how real people react to finished candidates. That is valuable, but it has three built-in drawbacks. It comes late, because there has to be something to test. It needs recruitment and fieldwork. And it gets more expensive with every extra variant, because every variant needs its own respondents.
In practice, many brands do not test at all, or only pit one variant against another that already goes in the same direction. The question that matters most, whether the direction is right in the first place, is decided on gut feeling before production.
2. What synthetic audiences are
Synthetic audiences, called silicon samples in research, are simulated respondents. An AI model is given the characteristics of an audience, such as age, life situation, attitudes or media use, and then reacts to a concept the way people with those characteristics probably would.
The result is a forecast. It needs no recruitment and no fieldwork, and it can be run for several variants side by side. That is exactly what makes it interesting for the early phase, when it is still being decided which direction gets produced at all.
3. What research shows
The method is young but well studied. Three papers give a fair overview:
| Source | Finding | What it means |
|---|---|---|
| Argyle et al., Political Analysis, 2023 | Properly conditioned language models reproduce the response distributions of many population groups remarkably closely. The authors call this "algorithmic fidelity". | Simulated groups can meaningfully approximate real groups on many questions. |
| Brand, Israeli and Ngwe, Harvard Business School, 2023 | Simulated customers make purchase decisions for products like toothpaste, yogurt or laptops that follow basic patterns of real demand. | The method is useful for market research, not just opinion polls. |
| Bisbee et al., Political Analysis, 2024 | Averages match a large real survey well. But variance is too low, results depend on wording and change over time. | Yes for directions, no as a substitute for measurement. |
4. Where the limits are
The most important limit is in the third study: synthetic audiences agree with each other more than real people do. Outliers, strong rejection or the one subgroup that hears something completely different are easily lost.
- Averages yes, extremes less so: the tendency of a group can be estimated well, the edges less so.
- Wording matters: how audience and concept are described affects the result. The brief has to be clean.
- No substitute for sound: a forecast at concept stage assesses the described direction, not the finished sound. How a particular melody sticks only shows in a test with real people.
Anyone using synthetic audiences honestly says so. That is why every one of our reports states that it is a forecast based on simulated audiences and that statements in it are modelled.
Your sound direction, checked in advance
With Sonic Swarm, the AI system we developed ourselves, we test concepts for your sound logo or brand sound against simulated audiences before anything is produced. Bookable with any sonic branding project.
More on Sonic Swarm →5. How we use them in sonic branding
For exactly this early phase we built a system of our own: Sonic Swarm. The name is not an industry term and it is not an off-the-shelf tool; it is our own development, and we keep developing it. Sonic Swarm comes in at two points of a sonic branding project:
- Starting point: if you like, we first forecast how the brand's current sound and its competitors' sound land with the audiences.
- Concept forecast: before anything is produced, we test the sound directions from the kick-off workshop against the simulated audiences.
The result is an assessment per audience: whether the direction carries the intended values, where it comes too close to the competition, in which channels it works, which risks should be settled before production and which direction we recommend. How we build and weight the audiences remains our craft.
The main benefit is avoiding detours. Fewer variants need to be produced, and if a test with real people makes sense afterwards, it becomes smaller and more focused.
6. When you need real people
In our projects this runs in three steps:
- Forecast with Sonic Swarm: first we check the sound directions internally with our own system and go through the result with you.
- Production: on that basis we produce, depending on the project, music, sound design or your sonic branding with sound logo and brand sound.
- Test with real people, if needed: after that we can test the finished candidates with real people from your target group. Whether and to what extent is your decision, based on need, wishes and budget.
For a big launch, a relaunch of a well-known brand, or when a lot of media budget sits behind the sound, we recommend the third step. The measurement design comes from us, and we run large panels together with specialised research partners. How that works is on the sonic audit page, and why measuring pays off is in the article Sonic branding ROI.
The order is what matters: first the forecast narrows the direction, then the real test measures what actually ends up sounding.
Research, source selection, and editorial responsibility: Rahim Erbil. We used AI while drafting and researching. Every sentence was checked by hand before it landed here. Made with Herzblut and AI.
Related reading:
The same name on every line.
Concept, composition, sound design, mix, master: at Campera Media one name is responsible for every project, from the first briefing to the last second of the fade-out.
Meet Rahim →Frequently asked questions about
AI audiences.
Simulated respondents that an AI model builds from the characteristics of a real audience. They answer questions or react to a concept the way people with those characteristics probably would. The result is a forecast, not a measurement.
No, and we do not claim they are. Research shows that averages are often matched well, but the range of real opinions comes out too narrow. That is enough for direction decisions, not for securing a big launch.
Above all before production: to choose between several sound directions, rule out misses early and see where a direction comes too close to the competition.
Mostly it saves detours. If weak directions are ruled out before production, fewer variants get produced, and a later test with real people becomes smaller because only the strongest candidates go in. We do not quote fixed percentages, it depends on the project.
Yes, with a system of our own. Sonic Swarm is developed by Campera Media, and we keep developing it. You book it as an add-on with any sonic branding project. More on the AI audience analysis page.
Ready for sound
that impacts?
Tell us about your project, free and without obligation. We will get back to you within 24 hours with concrete ideas for your audio identity.
