Bridging the Discussion Divide with GenAI

Ray had known Dana for 19 years, since their daughters learned to ride bikes in his cul-de-sac. On the last Saturday in June they stood in his driveway and agreed, without hesitation, on a single proposition: only American citizens should vote in American elections.

90 seconds later, Dana was walking to her car.

What happened in between was not a debate. Ray said the country needed proof, real proof. Dana said it had proof and needed a way for people to use it. Ray said something about what the other side wanted. Dana said something about what Ray’s side really meant. Neither said anything false. Neither answered the other.

She stopped at the driver’s door.

“My grandmother,” she said, “was born in an Ohio county hospital, because that is where the hospital was. Her family lived across the state line. She spent 91 years there and voted in 18 presidential elections, starting with Eisenhower. In March she moved here to be nearer to me. Ohio has never registered her. Her certificate spells her name Marguerete. 3 Es, not 2. Her passport expired in 2009.”

Ray opened his mouth and found nothing in it.

It was not that he had no answer.

It was that Marguerete had come to Dana’s for Sunday lunch for years, and often ended up on his patio with a plate for him. She would tell him what he was doing wrong with his tomatoes. “Roots, not leaves. The leaves aren’t thirsty.” He had heard the story about the hospital at least twice.

He had been arguing about a category for 3 years. The category had been bringing him lunch.

On the workbench behind him was the bill, 40-some pages, printed that morning. Ray had gotten to page 4.

Dana had not printed it at all.

---

Bridging the discussion divide begins with an uncomfortable admission: most public arguments are not really disagreements about facts, and supplying one more fact will not end them. RAND describes the broader pattern as “Truth Decay” — rising disagreement about facts, a blurring of fact and opinion, and declining trust in the institutions that once settled both.

The goal is not to eliminate disagreement or force every issue into an artificial middle. It is to help people see what they actually agree on, what each believes to be true, what evidence supports those beliefs, and which differences come from values rather than data. A large language model (LLM), used well, can take the third chair — separating what can be verified from what is merely asserted, and what people disagree about from what they are afraid of. It does not decide who is right. The people at the table do that, if anyone does.

Bridging the Discussion Divide Starts Before the Facts

Positions Are Built From More Than Evidence

A position on a contested subject is rarely built from evidence alone. It is also connected to personal experience, family history, professional responsibility, political identity, religious conviction, economic interest, and a deeply held sense of what is fair.

That does not make facts less important. It means another fact may not move the conversation.

When people sense that something important to them is being dismissed, they defend harder. They repeat the point, attack the source, change the subject, or assume bad faith. The visible argument concerns legislation or immigration or climate or school policy. The deeper concern is trust, safety, freedom, dignity, or control.

Before asking who is right, a more productive discussion asks:

• What does each person believe is at risk?
• Which value is each person trying to protect?
• What experience shaped each position?
• Which evidence would each person accept as credible?
• What information could cause either person to reconsider?

None of these questions require anyone to accept misinformation or extend undeserved credibility to a weak claim. They identify the actual structure of the disagreement. One person prioritizes security while another prioritizes access. One fears government overreach, the other fears insufficient enforcement. Both may support the same goal while disagreeing about the scale of the problem, the effectiveness of the remedy, or the tolerable risk of unintended consequences.

Emotion Explains Interpretation. It Does Not Prove Anything.

Understanding emotional stakes is not the same as treating emotion as evidence. Fear that a policy will cause harm does not prove the harm will occur. Trust in an institution does not prove the institution is correct, and distrust does not prove it is wrong.

Emotional stakes matter because they explain why evidence is being read a particular way — why two people can open the same report and come away holding different findings, different risks, and different uncertainties. A useful discussion keeps four things separate: what is demonstrably true, what is inferred or predicted, what each participant values, and what each participant fears or hopes will happen.

Once those are visible, participants may still disagree. But they are more likely to be disagreeing about the real issue than about a caricature of each other.

GenAI as a Third Participant, Not a Referee

GenAI is most constructive as a facilitator, not a judge: summarizing positions, separating claims from values, surfacing unanswered questions, and making the evidence easier to inspect. Research from Stanford’s Deliberative Democracy Lab suggests informed, moderated deliberation can reduce some forms of political polarization. The lesson is not that everyone eventually agrees. It is that structured discussion changes how participants understand the issue and each other.

Consider two people who disagree. Instead of asking GenAI which one is correct, they run a sequence: each explains their position in their own words; GenAI summarizes both without evaluating them; each confirms the summary is fair; GenAI separates factual claims from assumptions, predictions, values, and emotional stakes; the participants ask for a fact-check of the central claims; and then the people — not the system — weigh the remaining tradeoffs.

The opening prompt can be simple:

“We disagree about [topic]. Before evaluating our positions, help us understand the disagreement. Ask each of us what we believe, what we are concerned about, what values we are trying to protect, and what evidence we consider credible. Summarize our positions in terms we both agree are fair.”

The step that does the real work comes next, after both people sign off on the summaries:

“Now separate our statements into verifiable facts, assumptions, causal interpretations, forecasts, value judgments, emotional concerns, and policy preferences. Identify where we agree, where evidence can resolve the disagreement, and where the decision ultimately depends on different values or risk tolerances.”

This slows down the instinct to win. Each person gets to be understood before being challenged.

GenAI can also name the patterns that stall a conversation — a factual question answered with a slogan, criticism of a policy answered by attacking the character of its critics, a narrow question inflated into an ideological one — and it can do so without humiliating anyone. It can ask questions an opponent cannot ask without sounding hostile: What evidence would weaken your position? Which consequence worries you most? What is the strongest legitimate concern on the other side? Which outcome would tell you your preferred policy is not working?

Companion Toolkit Put better disagreement into practice

The ideas in this article are easier to understand than to use in the middle of a real disagreement. The Discussion Divide Playbook turns them into short exercises, structured discussion processes, and GenAI prompts that can be used at a kitchen table, in a boardroom, classroom, civic group, or community meeting.

The Playbook starts with a simple premise: the goal is not agreement. The goal is a better disagreement. GenAI can help summarize, classify, question, and compare evidence — but people remain responsible for judgment and decisions.

It includes Six Moves for working through disagreement: finding the shared goal, creating a fair summary of each position, using a real story to reveal what abstractions miss, separating facts from assumptions and values, testing claims against evidence, and weighing the actual tradeoffs.

For a fuller discussion, the Discussion Bridge combines those ideas into a structured process. A Two-Laptop variant is included for situations where participants keep characterizing each other’s motives instead of stating their own positions.

The Playbook also includes Rules of Engagement, reusable GenAI prompts, guidance for evaluating evidence, and the “Pluto” stop/reset rule — a reminder to ask whether people are actually disagreeing about reality or merely about the label being attached to it.

Just as important, the Playbook identifies when not to continue. It distinguishes ordinary disagreement from situations that should move to a human facilitator, mediator, counselor, attorney, crisis resource, or other appropriate professional.

Download The Discussion Divide Playbook — PDF

Free to use, teach, share, and adapt for noncommercial purposes with attribution under CC BY-NC 4.0.

Evidence-Weighted Does Not Mean Equal Space

Here is where most attempts at balance fail. A response is not unbiased because it gives every side the same number of paragraphs. Equal space manufactures false equivalence when one claim rests on substantial evidence and another rests on speculation or a mistaken premise. Procedural fairness means applying the same standard to every claim. It does not mean promising every claim the same outcome.

Ray and Dana’s argument makes the distinction concrete. The House passed an amendment to S.1383 on February 11, 2026, under the short title “SAVE America Act,” requiring documentary proof of citizenship to register for federal elections and photo identification to vote.

Start with what is not in dispute. The Bipartisan Policy Center states plainly that both parties agree voter registration should permit all eligible citizens — and only eligible citizens — to register and vote. Ray and Dana had already agreed on that in the driveway. It took them ninety seconds to forget it.

Now the measurable claims, which resolve unevenly. Detected noncitizen registration is rare: Utah reviewed more than two million registrations between April 2025 and May 2026 and confirmed 27 instances. Federal verification data shows roughly 0.04% of checked cases returning as noncitizens. On the other side of the ledger, BPC estimates about 12% of registered voters lack the documents the bill would require, and Kansas — which adopted a documentary proof requirement — blocked roughly 31,000 eligible citizens, about 12% of applicants, while its noncitizen registration rate stood near 0.002%.

Those numbers are strong enough to retire two talking points at once. “Noncitizen voting is swinging elections” does not survive them. Neither does “nobody will lose access over a piece of paper.”

What survives is the actual argument. Supporters reason that detected fraud understates real fraud, that verification integrity has value independent of the rate, and that public confidence in elections is itself worth a one-time documentation burden. Opponents reason that a remedy which stops thousands of eligible citizens to catch dozens of ineligible ones is a bad trade, and that back-end government verification achieves the same goal without shifting the burden onto voters. Both positions are internally coherent. They differ on how to weigh a false exclusion against a false inclusion — and that is a values question, not a data question. No amount of evidence dissolves it.

Even the polling illustrates the problem. Politico found 52% support for documentary proof of citizenship, with 18% opposed and 17% neutral; the White House cited polling at 71%. Both figures can be accurately reported and mean different things, because question wording differs. More revealing: in the same research, support ran about 58% among people who had heard a little about the bill and about 53% among those who had heard a lot. Familiarity moved the number down. That is worth noticing regardless of which side you are on.

Making the Fact Base Visible

An analysis is only as strong as the sources under it. A credible one does not simply declare a claim true or false — it shows where the information came from, why that source is relevant, and how the evidence connects to the conclusion. Ask for it directly:

“List the sources used in this analysis, explain why each is credible and relevant, and show how the evidence supports each major conclusion. Identify any claims that remain uncertain, disputed, outdated, or dependent on incomplete information.”

For policy discussions, the source hierarchy generally runs from legislation, court decisions, and government data; to peer-reviewed research and transparent institutional studies; to nonpartisan research organizations; to journalism that identifies its underlying evidence; to advocacy sources, useful for understanding positions but not as the sole authority on disputed facts. These are not interchangeable. A primary document establishes what a bill says. A dataset estimates the size of a problem. A study evaluates likely consequences. An advocacy group explains why a particular outcome matters to the people it represents.

GenAI accelerates all of this. It can also produce incorrect information, or cite a real source that does not support the claim being attached to it. The National Institute of Standards and Technology identifies this risk as confabulation: confidently presented content that is false or erroneous.

Human review therefore remains essential. Open the sources that matter most. Confirm they exist, that they are current, and that they say what the analysis claims. Give primary evidence more weight than summaries or commentary. This transparency changes the discussion itself — participants are no longer asked to accept a conclusion because it sounds persuasive. They can inspect the sources, challenge the logic, contribute better evidence, and regenerate the analysis.

That is Perpetual Innovation™ applied to public reasoning: assess the discussion, improve the questions, examine the evidence, act on human judgment, and regenerate as new information arrives.

Conclusion: Bridging the Discussion Divide

Bridging the discussion divide is not a prompt-engineering problem. It is a human problem involving trust, identity, evidence, values, fear, experience, and competing definitions of fairness.

What GenAI can do is slow the exchange down. It can separate facts from interpretations, surface the moral stakes, expose unsupported claims, and show where the disagreement actually lives — which is almost never where the argument started. It can help two people ask better questions before they start defending answers.

The goal is not artificial neutrality. It is a transparent process in which claims are weighed by evidence, participants are represented accurately, sources stay visible, and uncertainty is acknowledged rather than smoothed over. Human strategic judgment stays decisive: people must still determine which values matter most, which risks are acceptable, and what action follows.

Perpetual Innovation™ is not limited to new products, systems, or strategies. It includes continuously improving how we think, communicate, assess evidence, and decide together. In a divided environment, that may prove to be GenAI’s most valuable human-led application.

---

Dana printed the bill the following weekend. Ray finally got past page 4.

During the week she sat with her grandmother and worked out the sequence. Register in Ohio, in person, with proof. The bill does not let her mail it. The proof she has spells her name wrong. Fixing that means probate court; Ohio sends any record older than a year there. $30 to $85 to file. 3 to 4 months once it reaches Columbus. And the court wants documents from 1935 that spell her name the way she has spelled it since.

They sat at Ray’s kitchen table on Thursday with two laptops and one rule: neither could characterize the other’s position. Only the AI wrote the summaries, and only when both signed off. It took 4 tries.

What they found was not agreement. Ray still thinks the requirement is worth its cost. Dana still thinks it is not. They disagree now about a number, and how to weigh one kind of error against the other. That can be researched.

Somewhere around 11, Ray asked how to spell her grandmother’s name.

Not to win anything. He wanted to check whether the affidavit provision would cover her.

Dynamic Links

Internal resources:

• Pi-rdAI: Rapid Strategic Planning with Regenerative Dynamic AI — https://perpetualinnovation.org/pi-rdai/
• AI Agents Like Custom GPTs for Strategic, Ethical Decisions — [URL to be confirmed]
• External resources:
• RAND: Truth Decay and the Diminishing Role of Facts and Analysis in American Public Life — https://www.rand.org/research/projects/truth-decay.html
• Stanford Deliberative Democracy Lab: Can Deliberative Democracy Depolarize America? — https://deliberation.stanford.edu/
• NIST: Artificial Intelligence Risk Management Framework — Generative AI Profile (NIST AI 600-1) — https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-generative-artificial-intelligence
• Bipartisan Policy Center: Six Things to Know About the SAVE America Act — https://bipartisanpolicy.org/article/five-things-to-know-about-the-save-act/
• Congress.gov: S.1383 — 119th Congress, House amendment text — https://www.congress.gov/119/bills/s1383/BILLS-119s1383eah.pdf

Structured dialogue organizations:

• Braver Angels — https://braverangels.org/
• Living Room Conversations — https://livingroomconversations.org/
• National Issues Forums Institute — https://nifi.org/
• ListenFirst Coalition (Listen First Project) — https://www.listenfirstproject.org/listen-first-coalition

Suggested GenAI Prompts

1. I like using stories to convey complex or challenging topics. Tell a story that conveys both the fact and the counter-factual about this topic: [the SAVE America Act, or specify your own]. Make sure the narrative is grounded in verifiable data and be prepared to cite reliable sources for every factual claim embedded in the story.
2. What are the three or four things my family, congregation, or civic group most often argues about where we probably agree on the underlying goal and disagree only about the method? Help me test that theory with questions rather than assertions.
3. Diagnose [the topic or debate] by separating verifiable facts, causal interpretations, forecasts, value judgments, emotional concerns, and unsupported talking points — and tell me which parts evidence can settle and which parts it cannot.
4. Review the following exchange for talking points, deflections, false premises, unanswered questions, and claims carrying unequal evidentiary support. Reframe each one as a constructive question the participants could examine together.
5. I would rather do this in person than on a screen. Help me find structured dialogue programs active near [city or state] — Braver Angels workshops, Living Room Conversations guides, National Issues Forums, or other #ListenFirst Coalition members — and tell me what each format actually asks of me before I commit an evening to it.
6. Create a GenAI prompt related to [the topic or debate] that provides the objective facts of the issue along with fact-checked, evidence-weighted arguments for and against it — one I can share with someone who disagrees with me so they can reproduce and inspect the analysis themselves.

Sidebar The low-tech version came first

Nothing in this process requires a computer. The structured-dialogue field has been doing it with folding chairs and printed guides for years — and one of these organizations has been seating seven conservatives and seven progressives at the same table since 2016.

Braver Angels runs the format closest to what this article describes. Founded in December 2016 as Better Angels and renamed in 2021, the nonprofit seats seven conservative-leaning and seven progressive-leaning participants in a moderated Red/Blue Workshop. The stated goal is not consensus. It is accurate understanding and the dismantling of stereotypes — the same outcome the summarize-until-both-agree step is trying to reach.

Living Room Conversations is the self-service version. Founded in 2010, it publishes more than a hundred free conversation guides built for about six people and ninety minutes, structured around explicit conversation agreements and three rounds of questions. No facilitator required.

The National Issues Forums Institute, a nonpartisan operating foundation in Dayton, produces issue guides that frame a problem around three or four competing approaches and their tradeoffs, then supply a neutral moderator format. Libraries, schools, and civic groups use them regularly. NIFI’s framing discipline is essentially evidence-weighted analysis performed by editors rather than by a model.

The #ListenFirst Coalition, convened by Listen First Project, connects roughly 500 organizations working to reduce toxic polarization — the fastest way to find a practitioner nearby.

The honest comparison: a trained human moderator reading a room will outperform any current model at knowing when to press and when to let a silence sit. What GenAI adds is availability. It works at eleven at night, between two people who will never book a workshop, and it will restate a position twenty times without getting tired.

Those are real advantages. They are not the same as being better.

AI Disclosure and Attribution

This article was co-created with assistance from GPT-5.6 Thinking (2026, Jul) and Claude Opus 5 (2026, Jul) as part of the Pi-rdAI Rapid Strategic Planning ecosystem. Feature image and infographic is based on the article, topic and generated using ChatGPT image-generation model (2026, Aug.) under direct human curation. Content development and review by Dr. Elmer Hall — Strategic Business Planning Company (SBPlan.com) and PerpetualInnovation.org.

Copyright © 2026 Strategic Business Planning Company. All rights reserved.