Guidelines for information integrity and persuasion in industry
A 1-day workshop proposal
Ever more consumers use AI systems to learn about new categories of products and services, compare brands, and make purchase decisions. AI builds those answers from information provided — directly or indirectly — by the organizations themselves and by third parties. Because those answers carry enormous commercial value, organizations have a motive — a duty — to attempt to persuade AI to recommend them: either by maximizing their full verifiable information across AI sources (a long-term commitment) or by selectively creating, editing, structuring, and placing it (relatively short term).
Early in the workshop, we will define "Information Integrity" together in the context of persuasion. Persuasion is not a bad thing; commerce, lifestyle decisions, and much more depend on it. It can help a person — or machine — make a more-informed choice. But the fullness of evidence itself is not always enough for organizations under pressure to persuade.
While profit motive can drive persuasion, the brand value of information integrity pushes back. That tension, mostly unexamined inside organizations, is what this workshop is for.
One thing integrity is not: a purity test. Most sources hold incomplete information about an organization for ordinary reasons — lack of time, resources, or incentive. The line this workshop will draw is not between complete and incomplete. It is the point where an action meant to improve short-term persuasion puts information integrity — and brand value — at risk.
And integrity is not the enemy of emphasis. Selective emphasis is legitimate — an organization investing in a new direction, rallies around it -- on its site, in interviews, in the articles it pitches -- even while older business may still dwarfs it. Nothing there is false. That is positioning, and every organization does it. The question worth working through is what separates emphasis from misdirection.
This is also a safety question. The sources AI uses to respond to prompts about an organization's category are often the same sources it uses to respond to any number and types of prompts from others. If one organization's self-serving actions distort information in an AI source, will it potentially affect AI's responses to others whose prompts also draw on that source?
This is my work. I measure how AI systems represent organizations. I open and score the sources AI cites, checking what the pages actually say, to know the sources that define a category or marketplace for customers using AI.
Their incentive to act is powerful. An organization that is misclassified, poorly documented, or missing from the sources, and from AI responses that include its competition, needs to do something.
The question is not whether organizations will try to influence AI-mediated decisions. Of course, they will — either by a systemic maintenance of their verifiable truth or through the emphasis of marketing. The questions are where emphasis morphs into distortion, and who inside the organization can tell the difference and see it in AI sources.
What the workshop is for
This is a practical working session. Every section within it produces method for insuring the information an organization and its vendors produce for eventual inclusion in AI sources is safe and ethical – for employees, customers, and anyone depending on the accuracy of those sources.
The overall goal: give every attendee a way to surface and understand the potential pressure inside their own organization to shape AI's sources — pressures that mostly go unexamined because the effects are often indirect and uncertain — and to help define a clear line those efforts should not cross.
The shape of the day (to be refined w/collaborators)
Welcome: 10 min Speaker 1: 20 min Speaker 2: 20 min
Introduction of breakout sessions. Each runs 30 minutes; a participant presents their group's findings, followed by a 30-minute discussion with the whole room.
1. The definitions. The room builds and agrees a working definition of information integrity — what counts, and what doesn't. New is not the same as true; repeated is not the same as validated; incomplete is not the same as dishonest. Take-away: the agreed working definition, and the one-page test — "does our proposed update, interview, or registration add evidence anyone can check, or only the appearance of it?" — in language an attendee can use in their next marketing, communications, or vendor meeting.
2. Open the sources. Attendees query AI about a claim their own organization makes, then open some of the sources AI cites. Did we know about these sources describing us? Are they accurate, current, biased? Take-away: a simple method for auditing one claim your own organization makes, and the value of checking the sources.
3. What can information integrity look like? Organizations must attract and keep employees, customers, and investors — three audiences, persuaded continuously, in public. When might achieving information integrity in AI sources for one audience complicate the goal of persuading another? Take-away: a working picture of integrity in practice — examples matched to the three audiences an attendee's organization must persuade, ready to compare against its own current practices.
4. Where the pressure comes from. Crossing the line rarely starts with a decision to mislead. It starts with a revenue target, a launch date, an agency proposal, a vendor promising AI visibility. This session names the situations where crossing becomes tempting before anyone flags it. Take-away: a checklist (to complete) of the situations most likely to pull a team across the line, for attendees to review against their own pipeline, agencies, and vendor contracts.
5. Rules an organization can keep. The principles, pressure-tested against the morning's discussions: Truth (can it be substantiated?), Provenance (can anyone find out where it came from and who paid?), Independence (is corroboration real?), Completeness (was something important left out in a way that misleads?), Accountability (is a named person responsible?), and Stake (what do we have to lose — a brand, a reputation — that AI is unlikely to forget?). Take-away: a draft internal standard — a one-page policy sketch covering how employees, agencies, and vendors may and may not attempt to influence the organization's representation in AI, and who owns it — ready to adapt and table internally.
Close: 10 min — What we publish. The room's agreed definition, its revisions to the principles, and what deserves development next. Take-away: co-authorship of the published output. Everything the workshop produces will be published openly, with every collaborator named as a co-author."
Who I am looking for
Experts in their fields who are also effective leaders of impromptu discussions, collaborators interested in the role of industry in maintaining the integrity of information in AI sources for all.
The immediate objective of the workshop: agree on what information integrity means and why it matters -- and leave every attendee with something their organization can debate the next day.
The longer ambition is practical, published guidelines for the safe and ethical management of information produced by organizations that is intended for the sources AI systems read.
Thank you.
DISCLOSURE: I run a commercial practice, MichaelQuinn.ai, that measures how AI systems represent organizations and advises them on the accuracy and completeness of their information in AI sources — the subject of this workshop. I also publish the “Brand AI Report”, a newsletter on the same questions, and I am a former faculty workshop creator and facilitator for the Association of National Advertisers. I plan to incorporate the experience and output of this workshop into my professional work, including commercial settings, as I expect the workshop attendees and collaborators will. Everything the workshop produces will be published openly, with every collaborator named as a co-author. If you have concerns, questions, or suggestions, please contact me.