My name is Janette Roush. I am the Chief AI Officer for Brand USA, and you are here for our latest Agents of Change webinar: your virtual advisory board. How can you use AI to get feedback on something before the stakes are real? I used my own virtual advisory board to review this presentation after Claude wrote it for me, and I implemented those changes in the final deck. I’m going to walk you through every piece of what I’ve done.
I’ve been using this since about February of this year to review my work. Most recently I invoked my virtual advisory board when I was preparing my standard keynote for my fall conferences. The first of those was West Virginia, and one member of my advisory board came back and said, “I think anybody could have given this talk.” This section is generic. It does not feel like it came from you specifically. That can be scary feedback to get, but if it’s important feedback, you want it before you’re on stage, not from attendees typing it into their Cvent apps. You want to address it before it’s out in public, and that’s what we’re going to give you the tools to do today.
This was the slide Sally Hogshead was pointing at. Looking at my AI agenda, she said, any SVP could throw an AI agenda in the deck. What does this have to do with you specifically?
Why solicit hard feedback? If your role includes things like board meetings, that’s a place where you want to understand the objections up front, before you’re facing your board.
And AI isn’t going to do this by itself, because AI suffers from something called sycophancy. If you’ve used ChatGPT or Claude at all, you’ve seen it: “Good point. You are so smart. What a genius you are.” Maybe it makes you feel good. For me, it’s irritating, because that kind of feedback doesn’t push my thinking forward. I don’t need AI to tell me whether I’m smart. Somebody did an actual study on this. It’s a little old, from May 2025, and the models are certainly better now than when it was first tested. But it found that AI was 45 points more likely than a human reviewer to spare your delicate feelings. That’s not helpful when we’re working on work projects.
These are the members of my imaginary virtual advisory board, and these are real examples of feedback they’ve given me on presentations. Brené Brown asked, “Are you in the arena, or are you in the cheap seats?” Am I encouraging change and action, or just talking about it? On a presentation I did for MICEbook in New York City, Chip Heath said I’d jammed too many ideas together into one concept, an AI-ready events agency. Nancy Duarte is focused on making sure the audience is the hero, not you. Looking at my West Virginia presentation, she asked whose transformation I was promoting: Janette’s, Brand USA’s, or the businesses and DMOs across West Virginia, who are the real audience? And for a talk at the Virtuoso Travel Trends Summit, Sally Hogshead said I was borrowing somebody else’s framework when I could be using my own. All of that is feedback I find really helpful, because the line that stings is the one you need. You want that harsh feedback somewhere you can cry at your laptop by yourself, before it comes from external stakeholders.
So what is this? It’s a panel of AI reviewers, each built from an expert’s published work. I’ll also walk through how you could apply this to your own actual board, or to personas from a marketing or research perspective. You create one persona for each seat, whether it’s a board member, Brené Brown, or a marketing persona, and then you tell the AI how to have them all work in tandem to review your work.
Honestly, I read something about this on LinkedIn, went to Claude, probably gave it a screenshot, and said, “I want to do this. Tell me who to put on the advisory board based on the presentations I work on with you.” Some of those people I had never heard of outside my advisory board, but I was so happy with the feedback that I never went in to swap them for people I know better. Claude named the five people on the board, researched them, and wrote the files and the rules for how it all works. For this presentation, I went back afterward and asked Claude to tell me what it actually did and how it came to life.
Every seat on the board has four parts. The lens is what that reviewer looks for. The framework is how they make decisions or think; if someone has published a book, they have one. That won’t necessarily be true if you’re building this for a marketing persona, but the closer you can get, the easier it is to get this feedback. The score is what they rate the presentation on, from one to ten. And then there’s their own voice. That’s easy for thought leaders with lots of material online, because the AI can build a library of the real feedback they would give, so what you get feels authentic to that person.
These are the five people who sit on my board, and if you’re looking for feedback on presentations, it’s a great place to start. Nancy Duarte is a presentation designer and the author of Resonate, and she thinks from the audience’s perspective, which matters a lot to me. Carmine Gallo is a communication coach. There’s a great story about Bill Gates giving a TED Talk on diseases spread by mosquitoes, where he brought a jar of mosquitoes, opened the lid, and let them out to bite everybody, so he could talk about buying nets for people to sleep under. That’s a jaw-drop moment. Probably not one I’ll borrow anytime soon, but you don’t forget it. Sally Hogshead is a brand strategist and the author of Fascinate, and she asks what you’re doing that’s unique, or whether someone else could have done it. Chip Heath is a Stanford professor and co-author of Made to Stick, and he’s thinking about whether the things you say will be used later, whether people will quote you. And Brené Brown, who writes about vulnerability and wrote Dare to Lead, wants to see where you’re hiding behind words versus being truly vulnerable, because vulnerability helps your audience connect with you and take on board your advice. I really like this group, so feel free to steal it. There’s also no rule that says there can only be five. You can make different advisory groups for different types of reviews.
This is the first example of what these files look like. I’m not writing these myself. Have your AI tool, whether it’s Claude, Copilot, ChatGPT, or Gemini, put this information together for you. The files aren’t long, about 38 lines. Right at the top is the lens, then the framework Nancy works from: the presenter is the mentor, and the audience is the hero. Then come the three things she scores a speech on: the arc of the story, the clarity of the visuals, and whether the audience has a transformation. Then a few examples of her voice, what her feedback would sound like.
The file also shares the evaluation lens. She’s looking to see if the audience is positioned as the hero, and whether the ending gives a vision of what a transformed reality looks like. Here’s real feedback she gave me on the West Virginia speech. One of my examples felt too high level to her, and it wouldn’t necessarily make sense to a small business. She suggested I keep that example but add a second one, so everybody could see themselves in the story I was telling.
So how does it all fit together? Claude writes three types of files for us, and they come together in a loop. First, a file that captures your own taste, which I’ll share in a moment. Second, one file per seat, each with the lens, framework, scores, and that person’s voice. Third, the board file, which tells Claude how the five members come together and what they actually do.
The loop works like this. You brief Claude on what you’re saying and who’s in the room. If it’s a presentation, give it a copy. Then you tell Claude to run the prompt inside the board file, which rates your work according to what each of the five members is looking for. Then you and Claude read the dissent: what each board member said, and the synthesis Claude puts together. Then you revise and run it through the board again.
For this presentation I was fairly hands-off. If you’ve watched my webinars on building presentations, you know I no longer open PowerPoint and start writing words on slides. We have a skill at Brand USA where I can tell Claude, “Invoke the PowerPoint design skill and make a presentation out of all the work we’ve done.” For these monthly webinars I try to come up with really new material every month, so I don’t have an old deck to tweak. I have a conversation with Claude about what order to present the information in and what the outline should be, and then it shows me a draft. Because Claude has the file for the presentation and knows what I want to say on every slide, it can give all of that to my advisory board at once and tell me the results. Honestly, I often don’t even open the slides. I say, “That all makes sense. Implement it and run the board again.”
But that only helps for about three rounds at most. If you don’t like the result by then, you may want to start from scratch. With AI, you can only tweak something that’s not great so many times before you’re in an infinite loop where you can’t tell what’s good and what’s not. That happens to me more than I’d like to admit. When it does, go outside, touch grass, come back, open a clean chat, and give it the context it needs. You won’t really be starting from scratch, because you’ve already been thinking about this, but it’s hard to pull yourself out of that pit once you’re muddying around in something that isn’t working. Three rounds of revisions is the absolute most I’d do.
Where do these files live? The easiest way is a project in Claude or ChatGPT. In Claude it looks like a file folder, and I believe it’s the same in ChatGPT. If you have a board for your presentations, make a project called “presentation board.” Add all the files Claude wrote for you: the voice file for yourself, the five files for the five board members, and the file that says how the board judges something. Then every time you want something reviewed, start a new chat inside that project and upload the presentation.
This is what the board file looks like, and I’ll show you how to build it in a moment. In the skill itself it says not to review something more than three times. Each member gives their own score, and we take the average. We’re trying to reach an average of eight out of ten. If any score is below six, that reviewer needs to write a note sharing everything that needs to be updated.
This is asking for a lot of work, so when do you actually want to do it? For high-stakes things: a keynote or other talk, any high-stakes external presentation like an actual board meeting, or a high-stakes project where you’ve made a major revision and want more feedback afterward. When you don’t want it is for quick edits, where there’s only a small change from the last version the board reviewed, or for routine writing. This is time-consuming and energy-consuming, and there’s no need to spend this many tokens on something that isn’t high stakes.
It’s also a reminder, and I need it too, that your own opinion matters. You have to keep your own sense of taste in the work you do. I’m working really quickly these days with what I can create using AI, so a lot of my work is opening a presentation that Claude wrote two weeks ago and spending real time on Friday or over the weekend making sure it’s the presentation I want to give, not just the one Claude made look great in ten minutes without me doing anything. Keeping your finger on the pulse of your own work will only get more important.
Here are the seven steps, with the prompts for each, to build your own advisory board.
The first step is building you first. In theory this is the most important file, the one that starts to capture your own taste. I have my own version. I didn’t write it; Claude writes it and updates it for me. It’s a ghostwriting guide based on my sent email folder. As you can see, it’s Claude trying to determine my core voice fingerprint: what makes an email sound like Janette wrote it and not AI. I’ll clarify that for actual writing, I don’t use AI a lot. I use it for quickly shifting information. Email is often less “let me show you the poetry of my soul” (I’ll save that for LinkedIn) and more “what’s the deadline for this?” or “I need some information you have.” In that case, it doesn’t matter whether Claude or Janette writes the email. What matters is not holding up someone who’s waiting on that information.
For email, and I wouldn’t have guessed this about myself, I mirror the level of energy the other person sent. If they don’t use exclamation points, I don’t send exclamation points back. I like to be specific. Forward-leaning verbs, great. No corporate hedging. I’m not going to tell you to “kindly advise.” I don’t believe in formality theater. Call those quirks; they’re all in the file.
My Claude can see all of my sent emails, so I didn’t even have to paste in examples. If you don’t have a connector into your sent mail, pick ten emails longer than a couple of words and ask Claude to write a guide on how you write.
Then turn it into a feedback loop. Every time Claude writes an email for you, you presumably edit it a little before sending. Share that edited email with your AI tool and have it compare what you kept and what you cut. Do that enough times and the AI gets closer to what you’re looking for. There will be small, invisible things, like “ew, that feels a little AI-ish,” or “three exclamation points, I would never be that excited.” It will start to note what you change in future emails. Having the before and the after is one of the best ways to give feedback to an AI tool.
Step two is choosing who sits on your board. There are four types of boards. The first is the one I’m showing you, based on thought leaders. It’s arguably the easiest to build, because they’ve done long interviews, been on podcasts, and written books, and people have written about those books. Their point of view is very clear, and they’ve created frameworks to sell books around it.
The second is a panel based on your real board, which is more difficult. To be ethical about it, base it on the public record. If someone asks a lot of questions in board meetings and that’s on the public record, you can include those questions in the document, and that helps the AI give you those kinds of questions when reviewing future work. A safer way to replicate your board’s point of view is to assign roles rather than try to reinvent real people. Perhaps one seat is a fiscal hawk and another is the hotelier seat, and you consider what’s important to each. The fourth type is your segments. If you have audience research that breaks out visitor segments, you might use this process to review creative lines, with a board made up of your visitors.
This is how you could set up the initial prompt. I’m building a review board for my year-end board deck. Name five published experts whose frameworks would each catch a different way the work could fail. For each one, I want to know who they are, which book to reference for the framework, the one question they would ask about my presentation, and the three things they would score from one to ten. You can iterate on this endlessly until you’re happy with the people it chose.
If you’re building from the public record, that can include published bios, minutes from open meetings, questions they asked at the last meeting, and, if they’re a board member with an organization, that organization’s public stated priorities. Leave out private emails, hearsay, and personality tests. Ultimately you want a document you wouldn’t be embarrassed to show that person.
If there’s no public record, build the role instead of trying to recreate a specific member of your board. The fiscal hawk asks questions like, “What’s the ROI on that room tax? What happens if we cut that room tax by 10%?” The hotelier wonders which night of the week your program is going to help with. Then there’s the skeptical resident’s point of view and the TID chair’s point of view. Whatever the archetypes on your board are, you can recreate them here.
I also asked Claude to find a real example from a U.S. destination, and it selected Travel Oregon, which published its visitor profiles back in 2022. Those segments could become actual seats on your board if you’re reviewing creative copy or new ad campaigns. The more information you have on each segment, the better. You’d want more than the two or three sentences on this page, though you’ll probably need to keep it succinct rather than feeding Claude pages and pages. Give it that information, tell it what you want the end state to be, and it will help you put it together.
Step three is writing the member files. For a public figure, the prompt is: write a one-page reviewer file for Nancy Duarte using only her published books and long interviews. You want four sections: her framework, what she looks for in a presentation, three dimensions she would score from one to ten, and the phrases she’d actually use. Keep it to about 40 lines. Those are the sections you want for anybody. It’s easier for a public figure than for something like that fiscal hawk archetype.
For your actual board, write the same file for your board chair using only her published bio and the last three sets of public meeting minutes, so you know what this person cares about, what questions she asks, and what makes her say no. Write “unknown” where the record is silent, because a gap is where the AI will hallucinate and make things up. For a role, you do the same thing: write the same file for the county fiscal hawk.
Step four is the board file, the instructions to Claude on what to do with all of these files. You tell Claude to write a board file for these five reviewers that includes who the members are, a quorum of three, an average of eight out of ten across all reviewers to pass, no more than three rounds, and when to invoke the skill. I showed this earlier, and this is what it looks like for me.
Step five is briefing the board. Open a new prompt, or if you’re using a Claude project, start a new chat in that project. If it’s a presentation, upload it, and include the notes field if that’s where you write your notes. Otherwise, spell out what you’ll say on each slide, or ask Claude to ask you questions about what you’ll say if it isn’t clear from the slide. Share who the intended audience is. Let it know what might look like an accident but is deliberate, and I’ll give an example in a moment. And tell it how long until the presentation, so that if it’s in two hours, the board doesn’t suggest ripping everything apart and starting from scratch.
This is an example. Attached is my deck and the five reviewer files. Here’s what I’ll say on each slide, here’s who’s in the room, here’s what I won’t do, and here’s what looks like an accident but is deliberate. For instance, if you give it a PDF of a PowerPoint, a slide might look blank but actually have a video on it. That could read as dead air to the review panel when it’s really a video they can’t see yet. Then have Claude review it as each member in turn, in their own voice, scoring the three dimensions. Average the board’s scores. Anybody who scores it under six writes a note on how to fix it. And they should stay in character and not hedge. The whole point of going through this effort is not to get a sycophantic “I don’t know if you want to do this.” You want them to say what they think the issue is, so you can decide as quickly as possible whether it’s worth changing.
For the MICEbook summit I did in the spring, I initially got a score of 5.94. I swapped three slides and gave the board more information on my actual talking points, and the score went up to 8.2.
Step six is running the board and reading what it didn’t like. Looking at that MICEbook presentation, Brené Brown thought I was using too much of a framework, hiding behind it instead of presenting my own original thinking.
The board can also be wrong. On a different presentation, a big video showed up as a blank page to the board. In some cases it’s going to miss things if it doesn’t have the information. On another, I had a screenshot in a presentation about creating an AI policy for your organization. One of my resources is someone whose brand name is Dr. David Privacy. That’s not the name on his birth certificate, obviously, but he’s somebody with a doctorate who studies privacy law as it pertains to AI, and he offers courses under that name. The board saw it and asked why I was using these made-up names. I had to go back and say I don’t want to change it, because that’s the person’s brand online.
So remember that the board isn’t telling you what to do. We remain in the driver’s seat. They’re offering perspectives that we may or may not find useful.
The final step is folding in the feedback you find useful and running the board again, so the re-review tells you whether the fix landed. In one talk, I had text on a slide saying “storytelling is the moat.” Moat is a very AI term for the thing only you can do, the thing AI could never steal from you. Brené Brown said that’s the most human thing in the room, and I’d put defensive castle armor around it. I restated it from “storytelling is the moat” to “the story is yours,” and Brené was very happy with that.
When you run it a second time, say: here’s the revised deck, rescore it as the same five board members, use the same three dimensions, give me the average, and for every dissent from the first round, tell me whether the fix landed or where I have more work to do.
I use a version of this for the emails I have AI write. At the end of each week I run a skill I call Weekend, which is just instructions I gave Claude to do a number of steps in a row. There’s nothing magical about it. One of those steps is that for every email Claude drafted for me during the week, it finds the same email in my sent items, figures out the difference between its draft and what I sent, and implements anything it can learn in my voice document. When I first started this in August, I had to totally rewrite every email Claude wrote. Now, for lower-stakes emails that are just getting information from one place to another, I can often glance at it and hit send. I’m at 91%: I change an average of only 9% of the words.
This is the feedback I got on this exact deck. It started at 6.7 and ended at 7.8. I never quite broke the 8 mark, which is the goal in the board instructions, so this deck didn’t quite make it.
But at some point you can only tweak so many times before you have to go do the work. To start this week, pick one important deliverable due this month. Maybe it’s a board deck, another kind of presentation, a research deck, information on a new campaign you want to launch, or a response to an RFP from a meeting planner. Then use the prompts I’ve given you to ask who five people would be well suited to judging that type of work, because the people who should judge a conference keynote are not the same people who should vet a board report or your annual report. Claude can help you decide who. Share those prompts with Claude along with the draft you want feedback on. Just follow those steps this week to make this process work for you personally and for your own organization.
I want to thank you very much for your time.
