What exactly is machine-learning as it relates to artificial intelligence?

Machine learning, at the intersection of technology and behavioral research, is a field of computer science that gives computers the ability to learn without being explicitly programmed. AI is broader. Anything that can make a machine or software do something that would usually require a thinking person is artificial intelligence.


What is a Synthetic Respondent?

A Synthetic Respondent is a machine learning model of a targeted population segment trained to have content consumption interests that match that segment. Once created, Synthetic Respondents access the Internet where they keep abreast of content that is of interest to the segment they represent.  This ability allows clients to spot changing trends within their target market and ensures that the responses of Synthetic Respondents to client message tests always reflect current preferences of the target market.


What is the relationship between Synthetic Respondents and humans?

Synthetic Respondents are created from data obtained from research studies using human population segments, so they mirror the behavioral patterns, expressed desires, and preferences of the people within those segments. Once created, Synthetic Respondents are plugged into the Internet, where they read and view material appropriate to their interests. The output from Synthetic Respondents is more accurately representative of population segments than the output from individual people because it relies on large data sets, so clients get a profile of behavior that cannot be obtained from respondents who may not be representative of their respective universes.


What is an important difference between a human and a Synthetic Respondent in terms of the way each behaves?

An important difference is the large volume of information a Synthetic Respondent reviews every day. While an individual human reads a small subset of the Internet, a Synthetic Respondent examines a representative selection of the content that the entire population segment it has been modeled to represents reads every day. PersonaPanels provides clients with access to the articles that each Synthetic Respondent tags as interesting, helping you spot trends within the market segment that the Synthetic Respondent represents.  This ability ensures that when you use Synthetic Respondents for your advertising tests, new product tests, or other message-based research tests, you receive responses that reflect current trends and interests of large segments of the population encompassing hundreds, thousands, or even millions of people.  This gives you a better understanding of the preferences of your target market than is possible with individual people or small panels.


What is an important similarity between a human and a Synthetic Respondent in terms of the way each behaves?

Just like humans, Synthetic Respondents’ current interests and reading material influence what they look at in the future. Consequently, their interests can change over time, as happens with humans. Sometimes, as is also the case with humans, Synthetic Respondents are relatively stable since there may be little variance in their reading. At other times, Synthetic Respondents get “excited” about new topics that are related to their interests. As this exciting topic starts to age, Synthetic Respondents, like humans, move on to other things.


Can data from a traditional research effort be used to create Synthetic Respondents?

Yes. PersonaPanels can create Synthetic Respondents from a variety of sources, including a client’s own segmentation descriptions or information obtained from other research efforts.


Can different data sets be combined and then used to construct Synthetic Respondents?



Does PersonaPanels have existing panels of Synthetic Respondents that can be used or must panels always be custom created?

PersonaPanels currently has a US Generational Synthetic Panel available for immediate use.

The US Generational Synthetic Panel is comprised of 4 distinct generations within the US population: Generation Z, Millennials, Generation X, and Baby Boomers.  Each of these generations is comprised of 1-2 Synthetic, depending on the variations we saw in the data when the generations were segmented.  You can use any or all of these Synthetic Respondents in your research, depending on the market you are trying to target.

In addition to this already-created panel, we can custom create Synthetic Panels around your specific customer segments.

We will work with you to determine the type of panel that is best for you. If you would like to read more about these panel options, please visit the Our Panels page of our website.


What does a Synthetic Respondent consist of?

A Synthetic Respondent consists of a set of interest topics constructed in a taxonomy format based on the specific information used to build it.  A visual representation of this can be seen in Question 10.


Can you provide a visual representation of the taxonomy behind Synthetic Respondents?

Below is a visual representation of a simplified taxonomy connected to a hypothetical Synthetic Respondent named Environmental Millennials.



How do Synthetic Respondents stay current with market trends?

Once Synthetic Respondents are created, they begin to access the Internet where they evolve in real time through a machine learning process that integrates relevant information into their programming.  The more aligned a piece of Internet content is with a Synthetic Respondent’s interests, the more likely the Synthetic Respondent is to be influenced by it.  The newly integrated information, in turn, informs their interests going forward, ensuring they stay current with market trends.


Does PersonaPanels designate the articles that Synthetic Respondents read or do Synthetic Respondents operate on their own once they are given a foundation of sites?

PersonaPanels directs Synthetic Respondents to sites that are representative of the sites visited by their human segment equivalents.  For most consumer marketing efforts, these are the Alexa 500 Top Sites for a country, but for Synthetic Respondents designed around segments with specialty knowledge, the sites may fall outside the Alexa 500 list.  When your research requires Synthetic Respondents to access specialty online publications, we will discuss publications options with you.  Once on a site, a Synthetic Respondent’s machine learning capabilities take over, allowing it to decide which articles hold interest for its represented population.  


Why is it important to know what a Synthetic Respondent is viewing and how interested it is in the material?

On its own, what a Synthetic Respondent is reading reveals current customer trends. More importantly, though, Synthetic Respondents incorporate the things they find interesting into their personalities, just as human beings do. This ensures that when we expose Synthetic Respondents to your research copy, they will give responses that reflect each Synthetic Respondent’s current view. This also ensures that we can conduct research with Synthetic Respondents at different periods of time to see if attitudes toward your messages and concepts are changing.


How does a Synthetic Respondent measure its interest in the Internet content it reads?

Synthetic Respondents can be built based on multiple data sets covering everything from demographics to purchase habits to survey responses and more. These data sets come together and form an array of interest graphs for each Synthetic Respondent. Once Synthetic Respondents are created, the machine learning process scrapes tens of thousands of sources each day from the Internet and connects the most relevant content to the Synthetic Respondent, given the Synthetic Respondent’s particular interest graphs. An interest score for every piece of content is created by matching the content against the interest graph. The higher the interest score, the more closely it matches up against the interest graph.


Do Synthetic Respondents access social media sites?

No, these sites are not accessed because the posts are not reliably long enough to provide adequate context for the Synthetic Respondent to evaluate.


Can Synthetic Respondents process sounds and images?

No, not at this time, but it is a future goal. However, if images can be described in words, Synthetic Respondents can read the descriptions.


Can Synthetic Respondents access YouTube videos?

Yes, but the videos will need transcriptions. If transcriptions are present, Synthetic Respondents can read them.


Can Synthetic Respondents access sites that require logins or subscriptions?

PersonaPanels has access to multiple subscription-based sites and can obtain access to others as long as doing so does not violate the site’s Terms of Service.  Please keep in mind, though, that if content from a restricted site is reproduced or quoted on other popular sites, as often happens with Internet content, the Synthetic Respondents will run across it during the normal course of their reading.


Can Synthetic Respondents distinguish between advertisements and articles when they access a website?

Yes. Synthetic Respondents are able to make this distinction and will not read advertisements.


Can Synthetic Respondents be directed to specific content on the Internet?

We will ensure that the material a Synthetic Respondent accesses on the Internet is reflective of its represented population segment. For example, a study among Synthetic Respondents representing average consumers will access “popular” Internet sites that have high consumer traffic while a study among Synthetic Respondents representing a specialty occupation will include specialty sites that the average consumer would not typically access (for instance, Synthetic Respondents representing medical specialists could be directed toward specialist journals and forums).


As Synthetic Respondents mature, do they become less interested in what they read on the Internet or are shown by clients?

No.  A Synthetic Respondent’s interest in the material it reads on the Internet is only impacted by the content of the material and how aligned that content is with the Synthetic Respondent’s own programmed and evolving characteristics, not by how long ago the Synthetic Respondent was created.  This is also true of a Synthetic Respondent’s responses to a client’s message or concept tests. This is one of the benefits of Synthetic Respondents – they do not become static over time.


Do Synthetic Respondents ever go to a site and then decide not to review something?

No, if they access it, they review it. However, just because a Synthetic Respondent accesses a site, it doesn’t necessarily have interest in every article that appears on the site. Synthetic Respondents will provide you with a list of the specific articles they found interesting, along with a score that tells you how interested or uninterested they were in the article.  Information within articles that received high interest scores becomes integrated into the Synthetic Respondent’s memory through its machine learning capabilities. 


Do Synthetic Respondents evolve at an even pace or, like humans, is there great variance in terms of the degree to which they integrate information?

Synthetic Respondents are affected by what is going on in the wider web. If a lot of activity is happening one day and less another within a Synthetic Respondent’s areas of interest, there will be variance in how the Synthetic Respondent is affected.


Do Synthetic Respondents' interests evolve over time?



Why does a Synthetic Respondent's interests evolve over time?

The interests of a Synthetic Respondent are influenced by what it reads given the set of interest topics associated with it. This means that a Synthetic Respondent can start from one area of interest and be led to another area of interest until that topic and its many sub-topics fade from popularity. For example, a Synthetic Respondent with interest in gaming may showing growing interest in a new game release if it’s encountering Internet excitement in the months prior to launch.  If the game receives negative reviews after its launch, the Synthetic Respondent’s interest could start to wane.


Once Synthetic Respondents are created, how quickly do their interests begin to evolve?

Synthetic Respondents can begin to integrate new information and evolve new interests within twenty-four hours of being created.


Once Synthetic Respondents are created, how quickly can a client's new concepts and message tests be evaluated by them?

If a client would like us to custom-create Synthetic Respondents representing their specific customer base, we recommend that the Synthetic Respondents be allowed to access the Internet for two weeks before testing begins. This ensures that they have had enough time to integrate current trends and information into their knowledge base. If you would like to use our existing US Generational Synthetic Panel, you can begin new concept and message testing immediately.


If what a Synthetic Respondents reads can influence its reaction to a client's message tests, does what a Synthetic Respondent read 3 months ago have as much impact as what it read last night?

Synthetic Respondents are like humans in the sense that their core values and personality traits will always have the greatest impact on their reaction to the concepts and messages that clients put in front of them.  But, also like humans, information that Synthetic Respondents read can reinforce, expand, and/or nuance their opinions. This process, when applied to Synthetic Respondents, is known as machine learning.  If a Synthetic Respondent learned something of relevance 3 months ago that has been repeated and reinforced by other articles it has accessed since then, the impact of the article it read 3 months ago will remain high. If, however, that information stopped appearing in articles or was negated by more recent information, then the impact of the article accessed 3 months ago will have faded and have less of an impact than information read more recently.


When a Synthetic Respondent is exposed to a client's message or concept test, what is the “overall interest score” that is produced?

Just like humans, a Synthetic Respondent’s interest in a message or idea never occurs in isolation, but instead is influenced by all its experiences and preferences. In other words, a Synthetic Respondent’s interest in a message or idea occurs within the context of its “overall interests.” For example, when a Synthetic Respondent is exposed to a marketing campaign, the Synthetic Respondent takes into account all its interests before it decides if and how much it likes the campaign.


Can you provide a visual illustration of how the client's message and the Synthetic Respondent's overall interest areas interact?

In the simplified example below, a hypothetical client has created a message (“Message 1”) for a hypothetical Synthetic Respondent to evaluate (“Environmental Millennials”), The Synthetic Respondent accesses the entirety of its programmed interests and the interests acquired by it during its machine learning phases and determines the percentage of alignment between the client’s message and it’s own interests.  This becomes the Synthetic Respondent’s “Overall Interest” in the client’s message.  In this illustration, the Synthetic Respondent’s overall interest score would be 40% because only 40% of the information in the message overlapped with the interest areas of the Synthetic Respondent.



When a client’s concept is put in front of a Synthetic Respondent and the Synthetic Respondent gives the concept an overall interest score, is the score cumulative in nature (reflecting everything the Synthetic Respondent has experienced throughout its life)?

Synthetic Respondents are composed of both permanent and transitory characteristics.  The permanent characteristics consist of attributes that have been programmed into the Synthetic Respondent as part of its personality.  These traits remain consistent over time.  Synthetic Respondents also read material on the web and are influenced by this material.  This material tends to have transitory influence over a Synthetic Respondent unless the content is repeatedly reinforced by additional material over a prolonged period of time.  When a Synthetic Respondent evaluates a client’s message or concept, it pulls from both its permanent and transitory influences, resulting in message scores that reflect whatever is going on with a Synthetic Respondent at that particular moment, similar to a human when survey questions are being answered at that moment-in-time. However, scores can be captured at specific time intervals to see how the Synthetic Respondent’s preferences are changing or staying consistent over time.


Can the Synthetic Respondent's interest in a concept be compared with a human's interest in the same concept?

Everyone involved in research knows that concept tests run between different human respondent groups or run at different periods in time among similar respondent groups seldom receive identical results and are most usefully employed for directional analysis.  The same is true of research run between human respondent groups and Synthetic Respondents.  PersonaPanels has conducted validation tests and found that when human respondents and Synthetic Respondents both represent the same population segment and are both employed during the same period of time, then results are directionally similar.


When a Synthetic Respondent is reviewing a concept, it not only produces an overall interest score for that concept, it also reveals a topic that is resonating with the Synthetic Respondent as the concept is being reviewed. How does a topic score relate to the overall interest score that a Synthetic Respondent gives a client's concept?

A Synthetic Respondent is constructed as a hierarchy of topics and sub-topics built to represent the mind of that Synthetic Respondent. When a score is reported for one topic area, that shows how much overlap there is for just that topic.  A visual representation of this process can be seen in FAQ #34.


Can you provide a visual illustration of how the interest score of the primary topic is determined when a Synthetic Respondent reviews a client's messages or concepts?

In the simplified example below, a hypothetical client has created a message or concept (“Message 1”) for a hypothetical Synthetic Respondent to evaluate (“Environmental Millennials”).  The Synthetic Respondent accesses the hierarchy of topics and subtopics built into its programming and acquired by it during its time on the Internet (“Topic 1”, “Topic 2”, “Topic 3,” “Topic 4”).  It then selects the topic that best aligns with the message (in this example, “Topic 3”).  The degree of overlap between the client’s message and the topic is the Synthetic Respondent’s Topic Interest Score for the client’s message.



Are there any other statistical measures used in concept testing beyond Overall and Topic interest?

Yes.  When analyzing interest in your new product ideas, advertising messages, or any other text-based content such as websites, customer letters, email campaigns, and brochures, PersonaPanels will always begin by providing you with Overall interest scores (see FAQs 29 & 30 for an explanation of Overall interest). There are circumstances where this will be all the information you will want or need — such as when you are trying to rank order interest in multiple new product ideas before you develop them for market or when you want to decide which advertising message has the greatest chance of success before you spend money on art and media. 

There are other circumstances where you also will want to know Topic scores – such as when your new product ideas or advertising messages are not performing well and you want your target audience to provide other, related topics that might increase their interest.  With Topic scores, you are provided suggestions beyond, but related to, your message that you can consider incorporating into it in order to increase your target audiences’ interest (see FAQs 33 & 34).  Overall interest scores and Topic interest scores are always available within the pricing of any of our services, with the exception of PersonaPanels Monitoring, which tracks Internet trends.

Sometimes, however, you will want to know not only if your new product idea or new messaging campaign is interesting to your target audience, but why it is or isn’t interesting.  This is when you will want to use an additional statistical measure that we call Importance Testing.

Importance Testing begins with your product description, advertising message, or other customer communication and separates the content into component parts.  For example, a message may naturally lend itself to 3 components:  introductory section, informational section, concluding remarks.  Using this example, we would separate the message into those 3 sections and test each section individually to produce interest scores.  We then divide each individual interest score by the Overall interest score, producing a percentage total for each section.  This technique allows you to know what information within your message is driving interest.  You then can reformulate your message to accentuate positive drivers and minimize negative ones.


What are some of the advantages of conducting research through Synthetic Respondents as opposed to traditional research methods?

Let’s start with the important fact that Synthetic Respondents do not get fatigued so the number of messages and concepts they’re willing to respond to and the length of time they’ll devote to an exercise need not be a factor in your research design. Next, unlike people, Synthetic Respondents do not have ebbs & flows in concentration levels. Third, we know that respondents can answer questions in one way and behave in another. This is never the case with Synthetic Respondents, which respond honestly and exhibit no tendency to try to “please the questioner.” Fourth, sometimes humans have issues that cause timeline delays in projects, resulting in missed deadlines. This will never happen with Synthetic Respondents. Fifth, using Synthetic Respondents for your research needs allows you to maximize privacy since sensitive topics are not being reviewed by humans. In traditional research, these sensitive topics can leak out in unintended ways, causing problems for the client company. Sixth, most research engagements represent a point in time when data was acquired through human respondents. More often than not, the data ages quickly and its value to decision-making fades. Synthetic Respondent’s machine learning capabilities mean that they are constantly evolving to reflect current market trends, resulting in data that is always up-to-date and relevant. Seventh, panels of Synthetic Respondents can be accessed as often as a client desires, with no expensive incentives needed for respondent participation. Finally, Synthetic Respondents have no response bias, evaluating each message or concept independently of others that have already been put in front of them. This, combined with Synthetic Respondents’ quick response time, allows you to make as many minute changes in your messaging or concept tests as you want without going beyond your budget. 


What company is behind the technology used by PersonaPanels?

PersonaPanels works with Tanjo, a dynamic leader in the field of artificial intelligence.


Has PersonaPanels established any partnerships at this time?

Yes, PersonaPanels has entered into partnerships with several companies to help maximize our value to clients.  The Collaborative is a “one-stop” shop that coordinates health sciences research across multiple companies with diverse specialties, all under one supplier MSA.  Relative Insight provides text analytics software that allows clients to quickly and easily evaluate the data that is generated by PersonaPanels Monitoring.  FastFocus has a mobile platform that brings insights learned through our KnowNow Messaging service to Millennial human panelists.  Ironwood Insights Group designed and supports our user-friendly client dashboards. And Tanjo.ai developed the machine learning algorithms that underpin our system.  See “Our Partners” for more details.


What types of traditional research lend themselves to the use of Synthetic Respondents?

The best traditional application of Synthetic Respondents is concept and message testing.  In addition, Synthetic Respondents have a “non-traditional” ability to monitor Internet material that is appealing to their market segment in order to help clients identify evolving customer interests.


Are there certain types of research that would not be appropriate for Synthetic Respondents?

Yes. One example would be message recall work with respondents who saw sales representatives. This would not be an appropriate application for Synthetic Respondents.

If you have a specific question not covered by this FAQ, please use this contact form. Someone from our team will reply with your answer.

4 + 10 =