← Back to home
A Szlamka Consulting project

Reading thousands of voices, honestly.

Prism turns a large volume of free-text responses into a structured, inspectable account of the positions people actually took — one where every figure traces back to named responses, and every claim traces back to the words it came from.

Why this is hard.

A public consultation that draws thousands of free-text responses is difficult to read honestly. Counting responses rewards whoever writes most. Summarising them by hand is slow, unrepeatable, and quietly discards the minority and the inconvenient.

The usual shortcuts — keyword tallies, sentiment scores, a sample read by an overworked team — each lose something a decision-maker needs: the actual structure of opinion, and the ability to stand behind the result when challenged.

From raw responses to a map of positions.

1

Distil each response into the positions it advances

A single comment is reduced to the discrete points it actually makes — each one a topic together with the stance the writer takes toward it. Filler and boilerplate are set aside as non-substantive: counted, but not treated as positions.

2

Group positions that mean the same thing

Positions that express the same idea are gathered together by resemblance of meaning, not shared keywords. The strands of opinion are not chosen from a predefined list — they emerge from what people actually wrote, and you can read them coarsely or finely as the question requires.

3

Organise each group by stance, with evidence

Within a group, positions are sorted by the stance they take — support, opposition, a proposed change — and each is shown with its strongest verbatim quotation. The result is not a single average but the internal structure of agreement and disagreement.

4

Keep everything traceable

Every position stays linked to the exact response it came from and the specific passage of original text behind it. Nothing is asserted that cannot be checked by reading the source.

See it read a real comment.

Two comments from a live US public consultation — the Federal Trade Commission's 2024 review of children's online-privacy rules — as Prism read them. Each colour marks a distinct topic the comment touched; greyed-out text is what Prism judged non-substantive and deliberately filed as nothing. Every highlight traces back to the exact words on the page.

Public comment · FTC COPPA consultation · 2024
One comment, five positions across four topics.
Support for COPPA Rule amendments to safeguard children's privacy Support for third-party data sharing restrictions for childrenminority EdTech School Authorization Data and Research Concerns School-authorized educational data collection authorization for ed tech
I appreciate the goals of COPPA and fully support any protections that can be made that can improve the safety and well-being of children in their day-to-day lives. I agree that use of personal data for profit and marketing to children should not be legal. However, at the same time we need to protect the future of children and the capacity of researchers, schools, teachers and administrators to take steps to improve our educational system. I fear that the l language of COPPA is too vague and would inadvertently stifle important research in AI, including 100s of millions of dollars invested by NSF, Dept of Ed, and DOD to improve K-12 education for the future of our nation. It is imperative that these regulations not stifle these critical efforts. This regulation can easily be fixed if the FTC: Extends the definition of “school-authorized educational purpose” to also include “external evaluation on educational effectiveness, and the use and sharing of data for the purpose of conducting basic research in education and learning.” and/or clarifying that the rule does not apply to research and evaluation conducted by not-for-profit institutions, even if the research uses data from platforms offered by for-profit entities. Please consider making these important adjustments.

71% of the text carried a position. The rest — the framing, the rhetoric, the closing ask — is greyed out: classified as nothing, rather than forced into a box.

Public comment · FTC COPPA consultation · 2024
A dense, citation-heavy argument — resolved cleanly.
Regulating children's privacy Supporting, opposing, and modifying verifiable parental consent methods School-authorized educational data collection authorization for ed tech
You can give as much parental consent as you want. However, it is still up to the parents to decide what they want to do with their child and their sense of privacy for the child. With the current social and technical climate we live in where many parents are often lenient towards their child’s privacy online, I believe modifying consent requirements where parents must provide separate verifiable parental consent is not necessary. Adding this is simply an extra step that is not needed for an operator to have consumers use their services. Parents are already willing to share and disclose information about their children and adding extra verification to confident parents is too much. Hoy and colleagues (2023) conducted a study where a survey was sent to parents with children 10 years old or younger and found that although they believe that their child’s information is sensitive parents, especially single parents and fathers, are more willing to share it. Despite having a sense of awareness that a child’s information is valuable, parents continue to freely share moments of their child’s life such as images of them going to their first day of school or home life. Parents do this because they want to share their child’s likeness to the world for a variety of reasons such as fame, sponsorships if fame is attained, or just how adorable they can be. Adding extra verification to parents who are already set in confidently sharing their child’s likeness will not do much. Furthermore, many parents are quick to click and demonstrate their understanding of privacy policies websites set without really reading into what it says. Privacy notices can be more concise and easier to comprehend these days. However, there is no guarantee that a parent will readily read what it says. Takshid (2022) cites an example where parents must consent to google to use their child’s information to utilize google classroom in school making the child susceptible to the major tech company using their information. A parent would most likely want to skim through and click “I understand” to allow their child to utilize the resource for educational purposes. Parents will do the same thing with extra verification defeating the purpose of informed consent from operators. Given the nature of how parents navigate online privacy for their children, there truly is no point in adding extra verification for parents who already are already set in disclosing children information no matter how valuable it can be. At this point, it is up to state legislatures and Congress to step in and create stricter child privacy laws. COPPA and the FTC can only do so much.

81% coverage. One dominant position (blue) threaded through two related topics; the academic citations are read as substance, the connective tissue is not.

That's the span-level view of single comments. Two worked examples go further — across many commenters:

What Prism is for.

Scale without a black box

It reads thousands of responses, but every conclusion stays checkable by a non-expert: follow any figure back to the highlighted source text and judge for yourself.

Emergent, not imposed

The strands of opinion and their labels come from the responses themselves, not from a codebook fixed in advance — and the level of detail is adjustable.

Faithful to disagreement

It preserves the structure of opinion rather than collapsing it to a score, and keeps a coherent minority view visible instead of letting volume bury it.

Breadth distinguished from amplification

It reports how many distinct people stand behind a position — not merely how many statements were filed — and makes coordinated or duplicated input visible and checkable.

Built for human judgment

It organises, measures, and makes traceable; it defers contested cases to a reviewer and records its uncertainty rather than resolving it silently. A defensible starting point for a decision — not a verdict that replaces one.

Where this fits.

Prism began as a proposal for reading public consultations — where getting the answer right, and being seen to, matters most. That origin fixed a set of commitments into how it works.

A lens, not a gatekeeper

Every response stays classifiable, traceable, and retrievable. The system surfaces the whole picture; it never silently decides what a decision-maker sees.

Nothing is “noise” until a human says so

Volume is data. A hundred messages about the same thing is a signal about severity, not a duplicate to discard.

Auditable by people who didn't build it

Open methodology, documented logic, checkable figures. “Trust us” is not a transparency mechanism.

People know it's there

No covert processing — what's collected, how it's handled, and how to opt out are all visible.

Close the feedback loop

Reading input is worthless without a visible response. The method has to connect what people said to what was done about it.

Bias is assumed, not disproven

It's measured and reported, not waved away — the goal is to keep bias visible and correctable, not to pretend there is none.

Augments judgment, never replaces it

The system organises; humans decide. That boundary is architectural, not just a promise.

None of this is new ground. Taiwan's vTaiwan has run citizen consultations at national scale since 2015; Decidim (Barcelona) and CONSUL (Madrid) power participatory democracy across more than a hundred cities. What changed is the input: those platforms rely on structured agree/disagree voting, while Prism reads unstructured free text — lowering the barrier to taking part, and keeping the result auditable and documented to the standard public bodies now answer to.

Want to see it on a real consultation?

Prism is one of the systems we've built. If you're working with large volumes of free-text responses and need an account you can stand behind, we'd be glad to walk you through it.

info@szlamkaconsulting.com