I ran my “Agentic CMO: Create your Content Agent” webinar in March.

1,200 signed up. 600 marketing leaders joined live. The chat was on fire and I didn’t get to half of it live. So I went back, read every message and answered the 30 questions that came up most.

What people said afterwards floored me:

“I’ve done a ton of AI-for-marketing webinars in the last 6 months. This has been absolutely the most insightful and helpful.”

“An incredibly clear and inspiring webinar. With the right amount of intrigue to make me immediately register interest in the 6-week builders training.”

“I love the analogy of treating it like your daily fitness and starting small. Really useful, thank you Victoria.”

Thank you for every word. It means the world to me. If you missed it, get your replay here.

Now, to the questions.

Content strategy

How does this approach to content production apply in B2C vs B2B?

Every company has a different content engine. Campaign-focused brands will have a messaging document at the heart that triggers repurposing. In other companies a landing page is the source of truth. In others it’s a whitepaper or a blog post.

The most important skill anyone can build in the agentic era is what I call workflow architecture. Can you look at your current process and map it as a system? If you’ve ever done process documentation or an SOP, you know the step-by-step approach I’m referring to.

If you can map your process, you can translate it into an agent. And when you do, you’ll find similar elements across every business:

• A central piece of content at the heart that becomes the foundation for everything else

• Research, consumer insight, creative ideation and reporting steps that inform your content engine

• Different touchpoints and channels that require different assets, each with their own vernacular and structure

If you learn how to think in systems and do this mapping, you can apply it to any content. B2B or B2C.

The workflow architecture is the skill. The industry is just the context.

Can you give some examples of copy swiping?

Copy swiping is a technique from conversion copywriting. Instead of using your words (or your copywriter’s words) to convey a message, you take impactful copy that was said or written by your customers and use their exact words.

For example: instead of me sitting down and thinking about how to explain the transformation the Agentic CMO Accelerator delivers, I use copy swiping to say “move from AI experimentation to agentic systems.” That’s exactly what CMOs are saying they want.

Copy swiping is used a lot on landing pages and in Meta ads. Your audience’s language is always more persuasive than yours.

You mentioned informational gain. How do you get that when using AI?

Information gain is a term used in Generative Engine Optimisation (GEO) and Answer Engine Optimisation (AEO). It means: if an LLM reads your content, what additional information does it gain versus every other source out there?

What generates information gain for a brand?

• Unique studies and original research

• Proprietary surveys and stats

• Quotes from genuine experts

• Detailed case studies with real results

• Podcasts with original conversations

• Scientific or advisory board output

All of these are unique sources of information that make your brand what it is. They bring information gain to LLMs but they also make your brand genuinely more interesting to humans. I think brands of the future will be built on the concept of information gain.

Agents and architecture

Do you create one agent that automatically creates sub-task agents based on the system prompt?

I’d start by creating sub-task agents individually, getting each one to deliver high quality, and then orchestrating them together with an Agent Manager. The risk of starting with the Agent Manager is that it delivers low quality and you won’t know which sub-agents need improvement.

The biggest challenge is quality. As soon as you have two agents each at 90% quality, your combined output drops to 81%. With three agents at 90%, you’re at 73%. Quality deteriorates with every additional agent in the chain.

For marketing, the majority of agents will be linear automations using n8n or Claude Code as linear processes. This is what’s easiest for marketing teams because it mirrors how teams currently work and keeps the human heavily in the loop for quality and ideation.

With Claude Code you can create more complex architecture using sub-agents or agent teams. But I don’t think architecture is the real challenge. Quality is.

Can agentic workflows created on one platform be migrated to another?

Absolutely. The platform isn’t the challenge, and with time launching agents on a new platform will only get easier. Claude Cowork made agents accessible for business people and the technology will continue to evolve.

The real challenge is how you translate your marketing or sales expertise (the knowledge you have from human-to-human collaboration) into AI. I call this workflow architecture. If you’ve architected a workflow that genuinely works, you can migrate it anywhere.

One caveat: every agent becomes more valuable with time because of memory. It learns from your context, your feedback, the reporting. When you’ve invested a lot of feedback, the agent becomes quite precious. But that feedback lives in a memory file that you can also migrate. So nothing is lost.

Where do I start to build an agent?

Start by mapping your real-life process and manually executing every step as one prompt in Claude Chat or Claude Cowork. If at the end you get a decent outcome, you can try to automate it with Cowork Skills.

The challenge isn’t building. The challenge is architecting an actually useful agent for your business that solves a real challenge and imitates or even improves a human process.

What tool are you using for agentic AI? Are you using just Claude?

As a business person I’m strong in Claude Cowork (which is really friendly for business people) and I’m getting better at Claude Code. I’m also OK-ish at n8n.

When I need to build complex agents for my clients I prefer to hire AI engineers who build them in n8n, Claude Code, Crew AI or OpenClaw. Tooling is normally guided by the client’s current stack and appetite for experimentation and risk.

What agent is a good place to start with for internal communication?

Claude Cowork can connect to your Gmail, Microsoft Outlook, Slack and Teams and automate internal comms. That’s a great starting point because it’s a repeatable workflow that most people do every single day.

Has anyone used the Marketing agent in Claude Cowork?

Claude Cowork has a library of Skills. Plus, there are lots of libraries of Skills that other people build (caveat: prompt injection is real). I prefer to use other people’s libraries as learning material with curiosity: how did they build it? And then build my own Skills that capture the uniqueness of the brand I work with.

Skills in Claude Cowork are the easiest way to start: take the ones from the official Claude marketing plugin, customise them for your business. You can look at Skills created by other marketers and get inspired, but you’ll always need to customise them to make them fully attuned to your business, your content and your needs.

The more advanced your marketing is, the more unique your brand becomes. The brand will be the only thing that differentiates you in the future.

Brand, privacy and governance

What do you mean by codified rules in the scope of brand guardrails?

Your brand is already codified with brand guidelines, tone-of-voice or style guide, messaging examples, do’s and don’ts. But in a lot of companies brand is also tacit. There’s a Creative Director who has taste and understanding of how the brand should evolve. There’s a VP of Brand who has a feel for how the brand will show up. There are people who guide the brand. How do you translate that to AI? Examples, mood boards and inspiration work well.

But here’s what I see a lot: companies have brand guidelines that are 3-5 years old and haven’t been updated. AI gets confused because the context you’re feeding it is outdated, so it can’t produce a good outcome.

By codifying rules for brand guardrails I mean creating lots of examples for all touchpoints. Even customer care is a brand touchpoint. What are the rules? How should people reply? What’s the tone of voice?

How do you share your brand soul as rules and examples with AI?

My client’s team is using AI but the collateral doesn’t conform to brand guidelines. What can she put in place?

Some thoughts to share:

1. Codify your brand guidelines thoroughly, with real examples and create Brand Guardrail Skill.

2. Put a human in the loop (Creative Director or Designer) who gives feedback and fine-tunes the output. This is non-negotiable, especially early on.

3. All feedback from the human in the loop needs to go back into the Brand Guardrail Skill and enhance it. This is how the agent gets better.

4. If your rules are very well codified, you can add a Quality Assurance Agent that checks the work of the first agent before it reaches a human. Two systems: one creates, one reviews.

A word on this: I see lots of companies hire junior copywriters or designer to manage AI and give feedback to its output. With all empathy and love, most junior marketers won’t have much to add to the agent’s output. You need a mega-senior copywriter or designer to give feedback to agentic output. The human in the loop needs to be someone with real taste and standards.

How do you ensure cyber safety for your AI agent? What stops an agent from going rogue?

First, agents only do what you tell them to do. They don’t have curiosity beyond your instructions and the definition of output. If your system prompt says “use this data to write blog posts about skincare,” the agent isn’t going to wander off and read your financial records. It follows the brief.

That said, build in practical guardrails. Scope your agents tightly. Give them access only to the files and data they need, nothing more. In Claude Cowork, you choose which folder to connect and the agent can’t see anything outside of it. Review outputs before they go anywhere public. Use separate projects for different clients so there’s no crossover.

The “rogue agent” fear is understandable but in practice it’s more like a very enthusiastic team member. It does exactly what you asked, sometimes too literally, but it doesn’t go exploring on its own.

For agencies and consultants, how do we manage data privacy and client IP?

I take this seriously because I work with client data too. Here’s what works:

Never allow AI to train on your data

“Help improve model” is a no-go. Always check your settings.

Personal vs business plans

• On a personal plan (Claude, ChatGPT, any of them) you need to go into settings and deactivate data sharing yourself.

• On business or team plans like Claude Teams or Claude Enterprise, this is deactivated at company level by default. That’s one of the real benefits of business plans. It doesn’t rely on every individual remembering to do it.

Use your client’s enterprise AI

For regulated clients, I don’t use my own ChatGPT or Claude. I ask the client to provide me an email address and access to their enterprise instance.

Keep client projects separated

Use distinct projects in Claude Cowork for different clients so there’s no bleed between data sets. Build Skills per client or per use case.

Be transparent

Tell your clients you’re using AI tools, explain how the data is handled and let them ask questions. Most clients are fine with it once they understand the safety and many will ask you to share the work you’re doing.

For larger enterprises

You can deploy models inside your own cloud environment (VPC) so the data never touches the provider’s infrastructure. Anthropic offers this through Amazon Bedrock and Google Cloud Vertex AI. OpenAI offers it through Azure OpenAI Service. This is the enterprise-grade solution.

Any concerns about privacy with voice transcription tools like Wispr Flow?

Valid concern. Wispr Flow specifically, it only listens when you activate it: you press the button, dictate and it stops. It’s not passively recording across your apps. It’s not like Siri, constantly waiting for a trigger word.

But the broader principle applies to every AI tool you use. Always check the settings and make sure the tool is not training on your data. On personal plans, deactivate data sharing yourself. On business plans it’s usually off by default.

Be intentional about what tools you use and where. It doesn’t need to be scary. It just needs to be conscious.

Enterprise access and Co-Pilot

Enterprise access is limited. My org only has free Copilot. Am I alone?

Definitely not alone. We have so many clients with the same challenge.

I did a role at Veygo, part of Admiral Group, and although Veygo operates as a scale-up, it follows all the compliance rules of Admiral. So I’ve lived this.

Here’s what’s important: if you’re in a large corporate, you’re much more restricted by the tools. Which means you need to be much better at agentic thinking. You need to know exactly what you want, how your agent will work, what the ROI will be. In a scale-up you can buy 5 tools, try them, assess them. In big companies with compliance layers, you don’t have that luxury.

But the agentic skill set is the real skill. How you think through your current process, map it out and move it to AI. That’s what matters. If you learn how to do it with Claude Cowork, you can afterwards do it with Copilot. The tool doesn’t matter. The thinking transfers.

The pressure in corporate is higher. You need to deliver ROI and something that’s genuinely embraced by the company. So the skill set required is actually higher. But that’s exactly what makes it valuable.

How is Co-Pilot agent?

Copilot is getting better with every update, and if it’s what your organisation provides, use it. The agentic thinking skills: how you design workflows, how you brief an AI, how you structure your prompts are transferable. The tool doesn’t matter. The process does.

How do you share tools like Claude with teams when there’s no enterprise licence?

An agent in Claude Cowork is a combination of a Project (memory) and task execution (Skills). You can share both with anyone. Caveat: as soon as you share a project it becomes less secure.

Claude Teams (Enteprise plans) is the simplest and safest route. It gives you a shared workspace where data isn’t used for training by default. Your projects are available only to your team members and Skills can be deployed at organisation level.

The ideal setup is to have AI Champions or Leads who build, test and manage agents at organisation level, and everyone has access to them.

Tools and tactics

What are the best tools for presentation decks, images and video?

For decks, Gamma is the crowd favourite and I’d agree it’s a solid starting point. Also with Claude Cowork and Code you can get a decent output.

For images, Midjourney and Higgsfield can get strong results depending on what you’re after. Plus Canva, which integrates with Claude.

For design, Gemini is getting very strong and tools like NanoBanana are worth exploring, as well as Pomelli for on-brand assets.

For video, Runway ML keeps coming up as the sweet spot between quality and usability.

💬 But here’s the thing I always say: use case first, tool second.

Don’t chase every new tool. Pick the use case, go deep and get genuinely good at it. The presentation that works for a weekly report won’t be the same as the one you need for a keynote to thousands of people.

What are people using for data and analytics?

For most marketing analytics workflows, I’d start with Claude. You can connect it to your data sources (Google Ads and GA4 MCPs are available) or upload CSVs and ask it to analyse patterns, create reports and surface insights.

For more automated reporting, n8n can pull data from multiple platforms and feed it into an LLM for analysis. The key is knowing what question you’re trying to answer before you pick the tool.

Any AI tool for employer branding?

Employer branding has so many applications. Think about the use case first, then pick the tool.

Are you trying to hire top candidates? Build your founder’s personal brand? Create social media assets your team can share? Or capture the authentic voice of your employees?

Each of these is a different workflow and AI can help with all of them. You can use Claude to build an Employer Branding Agent with Skills that codify your employer value proposition, tone of voice and candidate personas. Feed it your Glassdoor reviews, employee testimonials and internal comms as context. Use case first, tool second.

Any recommendations on AI 101?

Build a habit: block time to make it happen.

Start with one use case that’s part of your daily work. Not a big project. Something repeatable, like drafting an email, summarising meeting notes or creating a brief. Open Claude Chat, do it manually step by step and see what happens. That’s your AI 101.

AI is like fitness. You don’t start by running a marathon. You start by showing up every day and doing something small. Build the habit first, then build the complexity.

What AI tools to generate more B2B and B2C leads?

Apollo is strong for prospecting data, Clay is brilliant for enrichment and personalisation at scale, Lemlist handles the outreach side. Instantly for cold outreach and HeyReach for LinkedIn.

But the real unlock isn’t the tool. It’s how you design the workflow. Who are you targeting, what’s the message and how does the data inform the outreach? Get that right and any of these tools will serve you well. Messaging everyone with a general message won’t fly in 2026.

I’ve built agentic systems for social copy but still can’t automate creative asset creation. Any ideas?

It all depends on your brand maturity and visual identity. Are you Nike? You need a designer. NanoBanana can generate a prototype and first draft at most. Plus, you might need a Creative Designer to help you build and run an agent rather than doing it yourself.

The honest answer is that fully automated creative asset production at brand quality isn’t quite there yet for most use cases. The best approach right now is automating everything around the creative: the brief, the copy, the specs, the variations and keeping a human or a template in the loop for the visual execution.

Can you build a synthetic persona if you don’t have first-party data?

Yes, absolutely. Reddit is a goldmine for this. Find the subreddits where your ICP hangs out, copy the threads where they’re discussing their challenges and feed that into Claude as context. You’ll get a persona grounded in real language and real frustrations.

LinkedIn is an incredible source too. Reviews on G2, Trustpilot or Capterra are also brilliant. NPS data and customer service calls if you have access to them. Also your customer care emails and social media might have lots of gems.

My main question (with my marketing hat on): do your real customers hang out on Reddit and talk about you on LinkedIn? It depends on the product. For some products the answer is yes. For others, the people who talk about them are either enthusiastic promoters or unhappy detractors, both outliers and not your average customer. Just be careful to find your ICP rather than the extremes.

AEO/GEO and content production

What is AEO/GEO?

GEO stands for Generative Engine Optimisation and AEO is Answer Engine Optimisation. Both are about making sure your content shows up when people ask AI tools like ChatGPT, Perplexity or Google’s AI Mode a question. It’s SEO for the AI era, and it’s going to matter more and more as people shift from searching to asking.

My observation: the classical SEO toolkit correlates heavily with AI Overviews, while AI Mode, ChatGPT, Claude and Perplexity citations require a completely different approach.

What were the results with GEO content production for Oxford Online Pharmacy?

Appreciate the interest. I’ll be publishing a detailed GEO case study with real results when I have the client’s permission to share. Watch the newsletter for that or maybe a webinar.

Building a website. Any recommendations for optimising AEO/GEO from day one?

Create content with genuine information gain from day one. That means original data, unique perspectives, expert quotes and detailed how-to content that LLMs will want to cite. Structure your content clearly with headings, FAQs and schema markup so LLMs can easily parse it.

Also look at the technical foundation: LLMs love speedy, lightweight, minimal-JavaScript websites with lots of schema. The easier your website is to ingest, the more likely you are to be cited.

Are there tools to improve existing campaigns on Google and Meta?

You can feed your campaign data (export CSVs from Google Ads or Meta Ads Manager) into Claude Cowork and ask it to create reports, identify patterns, suggest optimisations and draft new ad variations.

To pull data automatically, you can use an intermediary tool. You can also connect to Google Ads and Meta Ads via MCP, but you might need to create a data dictionary to explain the structure of your data to the LLM.

For more automated approaches, Claude Code and n8n can connect to the APIs of both platforms and trigger analysis workflows.

Video and visual tools

How is Runway ML compared to HeyGen?

They solve different problems. HeyGen is specifically for AI avatar videos. It creates videos with a digital version of you speaking. Runway ML is more of a general video creation and editing tool. If you want “videos that look like you,” HeyGen is the more relevant one. If you want brand video content, motion graphics or footage editing, Runway is the better fit.

Has anyone used Relevance AI?

Relevance AI is a solid platform for building pre-defined and template AI agents. It uses different LLMs as the engine depending on the task and you can customise which LLM each step uses.

The question is whether you’re getting the output quality you need. If yes, keep going. If not, experiment with Claude Cowork for the design and thinking side, and keep Relevance AI for the automation execution.

Now go build something

The marketing leaders who’ll lead the next decade aren’t the ones reading more about AI. They’re the ones who pick one workflow this week, map it out and turn it into an agent.

If you want to do it with me and other CMOs, the Agentic CMO Accelerator is how. 6 weeks. Live builds. AI Roadmap and Team Adoption Plan. In a cohort of 20 CMOs and VPs of Marketing. Cohort 1 is full. Marketing leaders from AG1, ZOE, VEED, Essity, Ledgy, Cezanne HR, Verisk, ComplyAdvantage and PaySecure. What a room to be in.

Join our next cohorts in June-July and September-October (UK and US time zones). Applications open here.