Key Takeaways
- Town halls and conventional surveys reach the loudest voices and the most available people — two groups that rarely represent a full community.
- Sonderly is an AI community research tool that conducts asynchronous qualitative interviews at scale, then returns thematic analysis with demographic segmentation across all respondents.
- For campaign strategy teams, AI interviews surface what communities actually care about and how they frame it — not just which issues poll well in a binary question.
- For program efficacy assessment, AI interviews surface the gap between intended outcomes and actual experience on the ground — the implementation friction, the unintended consequences, and the community perspectives that administrative data and satisfaction surveys cannot reach.
- A study can be live in under an hour. The analysis shows not just what themes emerged, but how they differ across population segments.
Why Conventional Listening Methods Fall Short
Sonderly is an AI community research tool designed for public sector teams — campaign organizations, government agencies, and NGOs — who need to understand constituent and community voice at a scale and depth that conventional methods cannot reach. This post covers two of the most common use cases: campaign strategy research and program efficacy assessment. Both are served by the same underlying capability, and both face the same structural limitation in how public sector teams currently gather insight.
The limitation is participation. Town halls, public comment periods, and community meetings consistently attract a narrow cross-section of any population: those with the time, confidence, and proximity to show up. Sherry Arnstein's foundational 1969 analysis of citizen participation in the Journal of the American Institute of Planners noted that nominal participation mechanisms often produce the appearance of community input without its substance — a critique that has only become more relevant as the populations public sector teams serve have grown more geographically distributed and time-constrained.
Conventional surveys improve on reach but impose their own constraint: they capture what respondents are willing to select from a predetermined list. They cannot capture the reasoning behind a choice, the framing a community uses to describe a problem, or the barrier a person encountered that prevented them from engaging at all. Both of these gaps are costly — for campaign teams trying to understand what an electorate actually cares about and how to speak to it, and for program teams trying to understand whether their work is reaching the people it was designed for.
AI-powered constituent interviews address both. By conducting asynchronous, open-ended qualitative conversations at scale — and returning structured analysis across the full respondent set — they extend the depth of a qualitative interview to the reach of a survey, without the constraints of either.
Use Case 1: Campaign Strategy
The research problem campaigns routinely underestimate
Campaign strategy depends on a clear model of what a community cares about, how it frames those concerns, and what it needs to hear to trust a candidate or policy position. The standard inputs — polling data, focus groups, and advisor judgment — are genuinely useful but each carries a systematic gap.
Polling data measures issue salience but not issue framing. A poll that finds broad support for "improving public safety" does not tell a campaign whether constituents are thinking about violent crime, traffic enforcement, neighborhood lighting, or school security — four very different concerns that call for different messaging and different policy commitments. Treating them as interchangeable is how campaigns end up talking past the communities they are trying to reach.
Focus groups solve the framing problem but do not scale. A focus group of eight people in one district cannot tell a campaign how sentiment differs across geography, age, or income — and those differences often matter more than the aggregate finding.
Constituent AI interviews fill this gap: they are open-ended enough to surface how communities actually describe their concerns, scalable enough to reach across demographic segments, and fast enough to be useful on a campaign timeline.
Setting up a campaign strategy study
A campaign strategy study in Sonderly is typically designed around one of two research questions: understanding which issues a community prioritizes and how they frame them, or testing how a specific policy position or message lands with a target constituency. Both are structured as open-ended qualitative interviews rather than preference polls.
For an issue-priority study, the interview guide should open without naming specific issues — asking respondents to describe the concerns that most affect their daily life or their community before any issue framing is introduced. This prevents the study from confirming the campaign's assumptions rather than testing them. Follow-up probes should then explore what the respondent means, why a particular concern feels urgent, and what they would want a candidate to understand about it that they might not already.
For a message or position test, the structure mirrors what Blog 2 describes for brand concept testing: present the stimulus in the respondent's own session, capture unguided reaction first, then probe comprehension, credibility, and resonance. The public sector specific consideration is tone — constituents respond differently to policy language versus lived-experience language, and surfacing that difference is often one of the most actionable findings.
Sonderly operates on a bring-your-own-audience model. Campaign teams recruit respondents through voter file outreach, volunteer networks, canvassing lists, or community organization partnerships. Sonderly provides the interview link and handles all sessions asynchronously.
What the analysis reveals in practice
To illustrate the type of finding this kind of study surfaces: imagine a campaign team preparing to make public safety central to their platform in a mid-sized district. Pre-existing polling data shows "public safety" as a top-three concern for a majority of surveyed residents. The team runs a Sonderly study with 45 constituent interviews, using an open protocol that does not mention specific issues upfront.
The analysis across those sessions might return something like the following:
Theme: "Public safety" meant pedestrian and traffic safety, not crime — The majority of respondents who raised safety concerns described it in terms of road conditions, crosswalks near schools, and speeding through residential streets. References to violent crime were present but less prevalent than the polling data implied, and respondents who raised it often described it as a problem elsewhere rather than in their immediate neighborhood.
Theme: Institutional trust was the precondition, not the issue — A recurring pattern across respondents who were skeptical of political messaging was that the issue itself mattered less than their confidence that any candidate would follow through. This theme did not appear in any polling question but surfaced consistently as the explanation for why respondents felt detached from local politics.
Theme: Younger respondents framed safety through housing and stability — Respondents under 35 were significantly more likely to connect safety concerns to housing affordability and neighborhood change — framing economic insecurity as a public safety issue in ways that older respondents did not.
Each of these findings reshapes the campaign's approach in a different direction. The third theme, in particular, is the kind of finding that a standard age-crosstab on a survey would not surface — because no survey option would have connected housing to safety in the way respondents themselves did.
(Note: The above is illustrative of the type of output Sonderly's analysis produces. It is not drawn from a specific published study.)
Use Case 2: Program Efficacy Assessment
The research problem program evaluations routinely miss
Program efficacy assessment — whether in disaster response, public health, infrastructure delivery, social services, or any other domain — is ultimately a question about the gap between intended outcomes and actual experience on the ground. The standard measurement tools are administrative data (service delivery counts, coverage metrics, response timelines) and satisfaction surveys distributed to program contacts. Both inputs have a shared structural limitation: they capture what reaches the measurement system, not what the measurement system cannot see.
Administrative data tells you what was done. It does not tell you how it was experienced, whether it reached the communities it was designed for, or where the gap between delivery and impact actually opened up. A disaster response agency may have distributed supplies to a thousand households and have no visibility into the neighborhoods where residents didn't know assistance was available, couldn't navigate the registration process, or distrusted the response channel enough not to engage.
Satisfaction surveys compound this by capturing the experience of those already in contact with the program — a population that, by definition, overcomes the access and awareness barriers that often determine whether a program achieves its purpose. The perspectives that matter most for program improvement are often held by people who are hardest to reach through conventional feedback mechanisms.
Qualitative interviews designed with this gap in mind surface what administrative data and satisfaction surveys cannot: the implementation friction that shaped who was reached, the unintended consequences that emerged in practice, and the variation in experience across communities, geographies, or respondent types that aggregate metrics flatten into a single number.
Setting up a program efficacy study
A program efficacy study in Sonderly is designed around the research questions the evaluation team actually needs to answer — which vary significantly by program type. A disaster response evaluation asks different questions from a public health outreach assessment or a social services program review. The interview guide is built to match: rather than a fixed structure, the protocol follows the experience arc relevant to the specific program and population.
The first design decision is who to hear from. Depending on the program, this might be affected community members, program responders or field staff, community liaisons, or some combination. Each of these populations holds a different part of the picture, and the evaluation's design should reflect which perspectives are most relevant to the questions at hand. Sonderly's bring-your-own-audience model means the research team controls recruitment — through community organization partnerships, field contact lists, SMS outreach, direct mail, or QR codes distributed on the ground.
The second design decision is what experience arc to follow. A useful interview guide for program efficacy typically covers three areas: awareness and access (how respondents first learned about the program and whether there were barriers to reaching it), experience in practice (what the program was actually like versus what they expected), and outcomes and gaps (whether it achieved what they needed and where it fell short). The value of open-ended questions at each stage is that respondents can name what was actually true for them rather than rating a scale that was defined in advance.
For studies covering geographically distributed areas — disaster zones, rural service regions, multi-district rollouts — the asynchronous delivery model is particularly well suited. Respondents complete the interview when and where it is convenient for them, which removes the logistical and access constraints that make large-scale qualitative research difficult in field conditions.
What the analysis reveals in practice
To illustrate: imagine a state emergency management agency conducting a retrospective evaluation of its response to a flooding event. Administrative data shows strong coverage figures — a high percentage of registered affected households received outreach. But field coordinators suspect the response underserved certain communities, and they want to understand why before the next event.
A Sonderly study with 60 respondents drawn from affected areas — recruited through community organization networks, local government contacts, and direct outreach in underserved neighborhoods — might return findings like the following:
Theme: Digital alert channels did not reach older residents in rural areas — Across respondents over 60 in outlying areas, a recurring account described not receiving timely information through the agency's primary channels (emergency app notifications and social media alerts). Many described learning about available assistance through neighbors or local church networks rather than official sources. Several assumed they were ineligible because they had not received direct communication. Representative quote: "I didn't find out about the assistance until my neighbor told me. I figured if I was supposed to know, I would have heard."
Theme: Perceived eligibility was a barrier independent of actual eligibility — A consistent pattern across respondents who did not access assistance was confusion about who qualified. The eligibility criteria were defined by insurance status and property ownership type, but respondents described their understanding in much simpler terms — "I thought it was for people who lost everything" or "I assumed renters didn't qualify." The communication materials were technically accurate; they were not interpreted as intended.
Theme: Field staff experience varied significantly by deployment area — Respondents in areas with established community liaisons described coordinated, responsive assistance. Respondents in areas without existing community infrastructure described fragmented communication and repeated re-registration requirements. This variation did not appear in aggregate satisfaction data because the two populations were measured together.
The second finding reshapes the communications strategy for the next response — not by adding eligibility criteria, but by changing how they are communicated. The third finding points to a structural question about how field coordination is staffed and whether liaison relationships should be built before an event rather than during one.
(Note: The above is illustrative of the type of output Sonderly's analysis produces. It is not drawn from a specific published study.)
What Both Use Cases Share: Demographic Segmentation at Scale
The two use cases described above — campaign strategy and program efficacy — are methodologically distinct, but they share a capability that is particularly valuable for public sector research: the ability to surface how findings differ across population segments.
Sonderly's analysis can be cross-referenced by any intake attribute collected at the start of the interview — geography, age range, prior program contact, referral source, deployment area, or whatever dimensions the research team specifies. For a campaign team, this means understanding not just that a concern is prevalent but whether it is concentrated in specific demographics or distributed broadly — a distinction that determines whether it belongs in a general message or a targeted one. For a program evaluator, it means distinguishing between patterns that hold across the full respondent set and those specific to a particular community or geography — which determines whether an issue is structural or local, and whether the fix is universal or targeted.
This level of segmentation has historically required large quantitative samples or multiple separately recruited focus group cohorts — both expensive and slow. AI interviews conducted asynchronously at scale make segment-level qualitative analysis achievable within a single study, in a timeline that is useful for both campaign decision-making and program iteration cycles.
From Finding to Decision
The output Sonderly returns is not a transcript dump or a data export awaiting interpretation. The analysis surfaces themes, prevalence, representative quotes, and segment breakdowns — and goes a step further: suggested next steps based on what the findings point toward. In the disaster response scenario described above, a finding that awareness gaps were concentrated among older residents in areas with lower digital access does not end with a theme label. It surfaces with a direction: what the pattern implies for how communications should be designed differently before the next event.
This is the distinction between findings that require interpretation and findings that are ready to act on. The analysis does not replace judgment — it does not make decisions, and it should not be treated as if it does. What it does is substantially reduce the ambiguity that makes judgment difficult. A campaign strategist who knows not just that an issue is salient but exactly how three different demographic segments frame it is in a qualitatively different position from one who knows only that the issue polls well. A program evaluator who has the specific pattern of where an implementation gap opened up — which communities, what the barrier was, what residents described when they encountered it — can bring a targeted recommendation to a policy conversation rather than a general observation.
On the role of quotes in particular: a quote is not the action — it is the evidence that makes an action defensible. The judgment remains the team's. What a verbatim account from an affected resident does is anchor that judgment in something specific enough to survive scrutiny. When a program administrator recommends restructuring how eligibility information is communicated, and can support that recommendation with direct accounts of exactly where respondents' understanding broke down, the recommendation is harder to dismiss and easier to prioritize. The quote does not tell you what to decide. It tells you why the decision you are already leaning toward is grounded in something real.
Segment breakdowns work the same way: they do not make the call, but they clarify what kind of decision is being faced. A finding that is evenly distributed across respondents points toward a systemic response. One concentrated in a specific geography or community points toward something targeted — a pilot, a local partnership, a separate investigation before a broader commitment. Knowing which situation you are in before you act is the difference between a well-calibrated response and an overcorrection.
Speed amplifies all of this. A campaign team mid-cycle and a program team managing an active response share the same constraint: findings are only actionable if they arrive while decisions are still being made. The elimination of scheduling, transcription, and manual coding — which in traditional qualitative research extend timelines by weeks — means analysis reaches the people who need it when it still has somewhere to go.
Frequently Asked Questions
What is an AI community research tool?
An AI community research tool is a platform that conducts structured, open-ended qualitative interviews with constituents, community members, or program participants at scale — without requiring a live interviewer for each session. The AI follows a research guide designed by the team, probes based on what respondents say, and returns thematic analysis across all sessions. Sonderly is an AI community research platform built for public sector teams, campaign organizations, and NGOs who need qualitative constituent insight at a speed and scale that conventional methods cannot provide.
How is this different from a constituent satisfaction survey?
A satisfaction survey captures ratings and selections from people already in contact with a program or response effort — a population that has already navigated whatever access and awareness barriers exist. It does not capture the reasoning behind those ratings, variation in experience across communities or geographies, or themes the survey designer did not think to include as response options. AI constituent interviews collect open-ended accounts in respondents' own words across a broader and more deliberately recruited population, and return structured analysis across the full set. The two methods serve different research questions and are often more useful in combination than as substitutes.
Can Sonderly reach respondents who don't have reliable internet access?
Sonderly's interview delivery is web-based and operates over a standard browser connection. Teams working with populations facing significant digital access barriers should consider this when designing their recruitment approach. For studies where internet access is unevenly distributed across the target population, supplementing Sonderly interviews with additional recruitment mechanisms — community organization referrals, library access, or in-person tablet sessions — helps ensure the sample is representative. Reach out to the Sonderly team for guidance on study design for specific access contexts.
How do you recruit respondents for a public sector study?
Sonderly operates on a bring-your-own-audience model. Public sector teams typically recruit through voter file outreach, community organization partnerships, program intake records, canvassing lists, social media in targeted geographies, or direct mail. The interview link can be embedded in any outreach format — email, SMS, or printed QR code — and respondents complete the session asynchronously at their convenience. This flexibility is particularly useful for reaching respondents across geographic distances or varied schedules.
How does demographic segmentation work in Sonderly?
Demographic and intake attributes are collected through screening questions at the start of the interview. Researchers specify which fields to collect — geography, age, prior program contact, referral source, or any other dimension relevant to the study design. Sonderly's analysis can then surface how themes differ across those segments, showing not just what a respondent group collectively said but how the finding varies by the attributes the team specified.
How long does it take to get results?
A study can be live and collecting responses within an hour of setup. Analysis is generated as responses accumulate. The practical timeline for a study depends on respondent outreach and participation rates, which are determined by the team's recruitment approach. For teams working on campaign timelines or policy decision cycles, the absence of scheduling, transcription, and manual coding steps — which traditionally extend research timelines significantly — makes AI interviews useful in time windows where conventional methods would not be feasible.
Conclusion
The problem public sector teams face is not a lack of opinions in the communities they serve. It is a lack of mechanisms for hearing them at scale, in their own words, and across the full range of people who make up a constituency — not just the ones who attend the meeting or return the survey.
Sonderly's AI community research tool is built for both sides of this problem. For campaign strategy, it surfaces what communities actually care about and how they frame it — not just which issues poll well in a binary question. For program efficacy, it surfaces the gap between intended outcomes and actual experience on the ground — the implementation friction, the unintended consequences, and the variation across communities that aggregate metrics cannot show.
In both use cases, the output is not a score or a ranking. It is a set of themes, with prevalence, with quotes, with segment breakdowns, and with source transcripts accessible for verification. The kind of evidence that makes a decision feel grounded in what people actually said.
Run a constituent study with Sonderly →