28,619 Persian-language voices become a navigable map of propaganda discourse — then guide deeper human conversations.
HarvestInterviewOrganizeValidate
Navigate with → / ← arrow keys or click the dots below.
MechanismsWho funds and amplifies a narrative?
تأثیراین روایتها چه اثری بر اعتماد مردم دارد؟
EvidenceEvery insight stays attached to people’s words.
راهحلچطور میتوان در برابر پروپاگاندا مقاومت کرد؟
Context 1/3
What is "sensemaking"?
When thousands of people talk about an issue, nobody can read it all. The usual shortcuts —
polls, hashtag counts, viral posts — either flatten opinion into multiple-choice, or amplify
whoever shouts loudest.
Sensemaking is a middle path: use AI to read everything, organize it
into themes and distinct opinions, keep people's actual words attached as evidence, and present
it so a human can explore what a whole community thinks — including the quiet, thoughtful voices.
The research community calls the collection side of this "broadlistening" —
the opposite of broadcasting: one institution listening to many people, instead of speaking to them.
The term comes out of the Plurality movement around Audrey Tang (Taiwan's former digital
minister) and Glen Weyl.
Context 2/3
Whose tools are we standing on?
Google Jigsaw — sensemaking-tools
Jigsaw open-sourced the toolkit behind our analysis engine: gathering responses, discovering topics and opinions, preserving supporting quotes, building reports, generating propositions, and preparing them for validation. We adapted it to run on Claude models.
AI Objectives Institute — Talk to the City
One of the AI interviewer approaches we evaluated. Its elicitation bot conducts open-ended conversations and asks follow-up questions to draw out participants’ reasoning.
Team Mirai — みらい議会 (mirai-gikai)
An AI depth-interview platform with adaptive follow-ups and a structured report for each participant. We localized its interviewer and interface fully into Persian.
Showra — ours
Our community platform now brings the pieces together: a native Persian AI interviewer for collecting depth conversations, plus a Polis-style voting engine for validating propositions with real participants.
Context 3/3
Project Prelude: the issue at hand
Prelude studies media propaganda targeting Iranians — how audiences talk about
the influence of outlets like Iran International, Manoto, BBC Persian, state TV (صداوسیما), and
social-media influencers, across 2016–2026 including the war period.
28,619
Farsi statements collected & labeled
6
analytical dimensions per statement
~$600
total collection + labeling cost
10 yrs
of discourse covered
The az-democracy hashtag study (#از_دموکراسی_بگو, 4,180 tweets) was our prototype:
we ran it through every stage first, end to end, so Prelude runs on a proven path.
The big picture
Two ways of listening, one analysis machine
⌁ Harvest the public record
Broad, cheap, unprompted. Reveals the terrain and what is over- or under-represented.
◌ Interview for depth
Adaptive, invited, reflective. Captures reasons, lived experience, and possible solutions.
⇢
Merge
One evidence-preserving dataset
→
⌘
Organize
Topics → opinions → quotes
→
▤
Explore
Bilingual public report
→
✓
Validate
Simulated shortlist → real votes
The map from harvesting makes the interviews smarter. Interview answers reveal what the public record missed. Any stage can send us back to listen again.
Outlet names, propaganda vocabulary, media criticism, six analytical dimensions, and 40 seed discourses.
Collect
32,775 posts
→
Label
Role + dimensions
→
Export
28,619 statements
One clean row per statement, with the verbatim Persian text preserved.
What the collection contains
Discourse
16,316
Mechanisms
14,888
Facts
11,126
Impact
7,769
Solutions
2,366
Who’s affected
1,402
The thin green bars tell us where deeper interviews are most valuable.
Stage 1b · Select the interview experience
One conversation, tested across three AI interviewers
AI Objectives Institute Talk to the City elicitation botTeam Mirai Mirai Gikai depth interviewerShowra · ours Our integrated Persian interviewer
The screenshots show the same test conversation, allowing us to compare probing behavior, readability, right-to-left support, safety framing, and the participant experience.
Initial sensemaking passJigsaw’s tools reveal themes, competing claims, and gaps in the harvested Twitter conversation.
Interview briefWe translate those findings into a neutral persona, safety rules, opening language, and follow-up strategy.
Configured on ShowraA 15-minute anonymous Persian interview, ready to publish across web and messaging channels.
After interviews, their deeper answers merge with the harvested posts and the combined dataset runs through the full analysis again.
Stage 1a → 1b · Build the question plan six dimensions, six paths
Every analytical dimension becomes a probing path
1–2Mechanisms & tools · facts & claims3–4Impact on people · recurring narratives5–6Affected groups · ways to respond
Each question carries optional follow-up guidance derived from the discourse itself—examples to probe, distinctions to clarify, and gaps worth exploring—without steering participants toward a preferred answer.
Stage 1b · Collect: interview Showra AI interviewer
From consent to depth — in one conversation
1Anonymous consent and clear safety boundaries
→
2Six-question plan derived from the harvest
→
3Adaptive probes surface reasons and lived experience
~15 minutes · any question can be skipped · no identity or location requested · transcript and structured report merge back into the analysis dataset · full bilingual interview guide
Stage 2 · Organize Jigsaw: categorization_runner
From 28k posts + interview responses to a map of opinion
TopicCredibility and trust in media
→
Distinct opinionTransparent funding is necessary for audiences to evaluate a source.
→
Verbatim evidence«وقتی منبع مالی رسانه روشن نباشد، چطور میتوان به روایتش اعتماد کرد؟»
Every statement
Harvest + interview text
→
AI organizes
Taxonomy, positions, evidence
→
Humans audit
Labels stay traceable to quotes
The analytical claim is never detached from its receipt. That traceability is the core of the method.
Stage 2½ · Score Jigsaw: get_bridging_scores
Finding the constructive voices
Purpose
Not all quotes are equal. This stage scores each one for constructive quality —
does it give reasons, acknowledge other views, invite dialogue?
Input
The organized topic/opinion/quote table.
Tool
A cheap, fast model (Claude Haiku) scoring tens of thousands of quotes.
Output
The same table plus a quality score per quote.
This is what lets the final report surface thoughtful contributions instead of the loudest or
most repeated ones — the quiet middle of a discourse becomes visible.
Themes, summaries, opposing positions, and real Persian quotes in one static site — with a language switcher and right-to-left layout.
Stage 4 · Test agreement Jigsaw: propositions + simulated jury
From "what people said" to "what they'd endorse"
Purpose
Turn the opinion map into concrete, votable statements and predict which could command broad agreement.
Input
The opinion map from stage 2.
Tool
Jigsaw's deliberative track: it distills opinions into short "propositions", then
runs a simulated jury — the AI role-plays real participants (each persona built from that
person's own posts) voting Agree/Disagree — and aggregates the votes with real election math,
including a proportional method so minority viewpoints stay represented.
Output
Ranked statement slates per topic, with simulated approval rates.
The honest caveat we always attach: these are AI predictions of
votes, not votes. We treat the output as a draft questionnaire for a real poll — which is
exactly what the next stage does with Showra — never as "X% of people agree." The rankings and gaps
carry the signal, not the percentages.
Stage 4½ · Validate with real people Showra — our Polis implementation
From simulated votes to real ones
1 · DistillTurn the opinion map into clear, votable propositions.
3 · Vote on ShowraReal people choose Agree, Disagree, or Pass.
4 · Find bridgesClustering reveals groups and shared ground.
Two different jobs: simulated votes generate hypotheses; real Showra votes create findings. A gap between them is itself evidence that the harvested discourse missed part of the wider community.
See the larger media-propaganda analysis:media-propaganda-report.pages.dev —
the live bilingual report produced from the Prelude Twitter collection.
Every stage ran on Claude (the toolkits were built for other providers — the
adaptations are ours and reusable), and the interviewer has hosted real Persian test sessions.
Not yet exercised: Showra validation and interview-fed re-analysis — Prelude will
be the first project to run the full process.
Limitations
What this process can't tell us
Who's missing: harvested posts represent people who post — not lurkers, not the
offline public. Interviews reach only people we can invite safely.
AI bias flows downstream: topic discovery, scoring, translation, and jury
simulation all inherit the model's blind spots. We spot-check every stage against raw quotes.
Simulated agreement is a hypothesis: jury results shortlist statements worth
testing with real people; they are never findings on their own.
Participant safety comes first: for this population, anonymity, consent framing,
and never collecting identifying details are requirements baked into the interviewer — not options.
Where we are · what's next
Prelude status
✓28,619 posts collected
✓Topics and evidence extracted
1Review + fine-tune
2Test interviews
3Documentary release
1 · Review and refineReview the early analysis of harvested data and fine-tune the interview guide.
2 · Run a test populationPrepare the site and run the interview with a starting population for testing.
3 · Prepare the releasePrepare the project and its public-facing materials for the documentary release.
Questions → Hadjar. This deck: docs/prelude-sensemaking-deck.html in the sensemaking workspace.
