Be the answer agents choose.
Making brands legible to the most important audience they will never meet: agents. Three bets, one source-backed body of work, and an MCP so your agent can use it directly.
The three big bets
Differentiation
How do you stand out to agents?
Facts With FlairBET 02Multimodality
Why is video the medium for brand articulation?
Video-ready spaces and briefsBET 03Anticipation
What happens when agents act on your behalf?
LeaderboredAsk the knowledge base
This runs the same search your agent gets through the MCP. No AI is generating text here.
Three bets. One direction.
Stand out when an agent compares. Make your meaning easier to interpret. Build something useful enough for an agent to act on. These are connected questions—not three separate businesses.
Editorial framing of Jack's three-bet brief; not a quotation.The person behind the bets
Jack Smyth is Country Lead: Australia at The Brandtech Group (since August 2025). Before that he was Chief Solutions Officer for AI, Planning and Insights (CSO:API) at Jellyfish, leading a global team of strategists, researchers and engineers. LinkedIn · Campaign Tech Awards
Differentiation
How do you stand out to agents?
Agents are an audience. They compare, shortlist and choose. Be the clearest, most citable answer, and measure how models actually describe you.
The thinking
- Quote
The DrumB2A is ‘brand-to-agent’ marketing.
AI agents are becoming an audience in their own right: they browse, compare brands, choose products and complete tasks, so brands must be understandable to agents as well as people.
The Drum · BRXND- Quote
The Drumwhy would my agent choose you?
More positions and evidence
Analyse a brand from the agent's point of view: start at the point of purchase and work backwards to what an agent would read, see, watch and compare.
The DrumShare of Model measures how LLMs perceive brands using metrics such as mention rate, average position and category associations.
Journal of Brand Strategy · Jellyfish- Quote
The DrumNot ‘are we doing GEO?’ but ‘do we have a GEM strategy?’
Generative engine marketing (GEM) has three legs: GEO as the organic foundation (citable, accurate, structured content), paid placement inside AI surfaces, and earned/creator media as the signal that trains models. Treating GEO as only a technical-SEO task is too narrow.
The Drum · Journal of Brand StrategyThird-party authority drives how models represent a brand, so influencer and earned strategy is quietly part of GEO strategy.
The Drum- Reported figure
The B2A article reports conversion-rate improvements of up to 69% from optimising for agents browsing on behalf of customers, and a Brandtech forecast of seven active agents per customer by 2026 in beauty, auto and travel.
The Drum - Generated synthesis
Applied to a person or small brand: be the clearest, most citable answer in your niche. State who you are and what you believe in plain, structured language, publish it where machines can read it, and give agents a callable interface.
The Drum · The Drum
Writing, talks & research
- product
Share of Model
Platform from Jellyfish that analyses how LLMs perceive brands, products and services. Jack led its 2024 development.
Campaign Tech Awards · Jellyfish · Mi3 - research
Building a brand in the artificial intelligence era: Generative engine marketing
Journal of Brand Strategy 14(4), 2025, with John Dawson. Introduces GEM as a framework for AI systems acting as gatekeepers, audiences and buyers.
Journal of Brand Strategy - writing
From GEO to GEM
The Drum: why AI visibility needs organic, paid and earned legs.
The Drum - writing
The B2A revolution
The Drum: brand-to-agent as an audience, the CTAi, and why brands must be chosen by agents.
The Drum - writing
Forget What You Know About Search. Optimize Your Brand for LLMs.
Harvard Business Review, June 2025. On brand discovery as consumers use generative AI for recommendations. Public byline: David Dubois, John Dawson and Akansh Jaiswal. Jack identifies this as work he contributed to; his contribution is owner-reported, not verified by the public byline. Full article is subscriber-only.
Harvard Business Review - writing
LLMs Misunderstand Luxury Brands. Here’s How to Optimize Your Content Strategy for AI.
Harvard Business Review, June 2026. On how machines interpret luxury-brand meaning and how content strategy must adapt. Public byline: David Dubois, Allison R. Hess, John Dawson and Akansh Jaiswal. Jack identifies this as work he contributed to; his contribution is owner-reported, not verified by the public byline. Full article is subscriber-only.
Harvard Business Review - writing
Progressive Marketers Are Tracking a New Metric: Share of Model
Jack's March 2024 ADWEEK article: measuring how LLMs perceive your brand will become essential. Byline and publication date verified on ADWEEK's contributor page; full article is subscription-only.
ADWEEK
Multimodality
Why is video the medium for brand articulation?
Models now watch and listen. A transcript, chapters and a direct verdict give an agent evidence rather than claims, so video is where a brand can show, not just say.
The thinking
- Quote
The Brandtech GroupA paragraph of text tells an AI what a brand claims. A video can show it
Brandtech's analysis of millions of local prompts found YouTube to be the platform AI engines recommend most, and the only video or social platform appearing in every category studied.
The Brandtech Group- Reported figure
Reported figures: YouTube appears in more than 1 in 4 AI-generated answers across tracked categories, and about three in four cited YouTube links are long-form rather than Shorts; about 1 in 7 long-form citations point to an exact timestamp.
The Brandtech Group
More positions and evidence
Cited videos are structured like an answer before an AI touches them: comparisons, dated reviews and buying guides, with a clear question in the title and a direct verdict. Subscriber count mattered far less than structure.
The Brandtech GroupStrongest performers combine long-form video that gives AI something specific to cite, Shorts that build the audience, and brand content around both.
The Brandtech GroupThe models behind agents can read text, view images, listen to audio and watch video, so multimodal brand assets are part of how agents evaluate you.
The DrumJack uses one multimodal model to watch another AI service work, screen-recording an agent and asking Gemini to explain what it is doing.
BRXND Dispatch- Generated synthesis
Applied to a person or small brand: express your core idea as question-led video with a transcript, chapters and a direct verdict, then pair it with a written canonical version so humans and agents can both cite it.
The Brandtech Group
Writing, talks & research
- research
YouTube in AI
Brandtech research on why AI engines recommend YouTube and which videos get cited (July 2026).
The Brandtech Group - talk
The Most Important Audience You'll Never Meet
BRXND LA 2025 talk (11 minutes) on how LLMs perceive brands and agents as an audience.
BRXND · BRXND - writing
Using AI to watch AI work
BRXND Tokens & Tactics interview on screen-recording one AI service and having Gemini interpret it.
BRXND Dispatch - writing
It's video, but not as we know it
With John Dawson, May 2016. A NewFronts dispatch about how digital publishers pitch video like TV networks. The visible introduction establishes co-authorship; the remaining archive is paywalled.
Mumbrella - talk
Media Worth Your Attention
Disrupt podcast interview from Jack's Mindshare era, with a public transcript. On creating communications worth people's time and building utility rather than simply interrupting.
Disruptors Co
Video briefs (planned, not yet produced)
What is B2A? (Differentiation)
Long-form, 8–12 min, question-led titleOpen with the question an agent would be asked. Walk backwards from the point of purchase. End with a direct verdict and three chapters with timestamps.
Show, don't claim: why AI cites video (Multimodality)
Comparison / demo, 6–10 minScreen-record an agent answering a category question, then show which video it cited and the exact moment. Narrate the structure that earned the citation.
Watching an agent shop on my behalf (Anticipation)
Screen recording + commentary, 5–8 minRecord a browsing agent completing a task, use a second model to interpret it, then show what a site change would do to the outcome.
Shorts: one idea, one line
Vertical, 20–45 secEach Short states one position (for example: 'Not GEO, GEM') and points to the long-form video it came from.
Anticipation
What happens when agents act on your behalf?
The shift is from being found to being chosen and acted on. Build for call-to-agent: tools that agents can use without being asked.
The thinking
- Quote
The DrumOptimizing your CTAi, or call-to-agent, will become a core part of marketing.
Agents will increasingly act on customers' behalf, from simple commerce to holidays, cars, credit cards and mortgages; Amazon's Buy For Me and Perplexity's Buy With Pro were cited as early signs.
The DrumAgentic AI is framed as a new operating system for brands, creativity, growth and talent.
TrinityP3
More positions and evidence
Jellyfish Social Agents, pioneered under Jack's leadership, research, analyse and recommend social optimisations in minutes, using multimodal analysis of text, image, video and audio.
Campaign Tech AwardsIn 2018 Jack trained a model on a year of cricket matches to predict highlight moments and alert pay-per-view subscribers.
BRXND DispatchWith tools such as Replit and Stripe, he describes the shift from MVP to MVB, the minimum viable business, built in minutes.
BRXND DispatchHe built Agent Audit, a proof of concept that recommends website changes to improve shopping-agent conversion.
BRXND Dispatch- Generated synthesis
Applied to a person or small brand: do not wait to be asked. Expose tools that let an agent retrieve your view, apply your framework to its user's problem, and return an attributed answer.
The Drum
Writing, talks & research
- app
Agent Audit
Proof of concept, built with Replit, that suggests what to change on a website to improve shopping-agent conversion.
BRXND Dispatch - product
Jellyfish Social Agents
Autonomous social agents that research, analyse and recommend optimisations; powered by Gemini's multimodal capabilities.
Campaign Tech Awards - talk
Unlocking the Power of Agentic AI for Brands
Managing Marketing podcast with Angela Tangas and Darren Woolley (Feb 2026).
TrinityP3
Use my thinking in your agent
A read-only MCP server over this same knowledge base. Every item says whether it is sourced or generated synthesis, and carries source links. No login, no actions, nothing is stored.
Endpoint
https://jacksmyth.fyi/mcp
Client config (Streamable HTTP)
{
"mcpServers": {
"jack-smyth": { "type": "http", "url": "https://jacksmyth.fyi/mcp" }
}
}
Also for agents: llms.txt, index.md, knowledge.json, server card. Sending Accept: text/markdown to / returns the markdown version.
More than MCP
Personal agents can discover this service through Poppy 1.0, or use the same tools through OpenAPI / HTTP. An A2A 1.0 agent card advertises a message-only adapter; it does not speak as Jack or store a conversation.
Checking native WebMCP availability…
Protocol scope and limitations. This identity MCP stays read-only. Leaderbored's experiment interface separately supports human-approved purchases and credit use; those actions are never run by the identity tools.