Bodnar & Flanagan (HubSpot UNBOUND 2026) — AI broke the old marketing playbook
▶ Watch on YouTube · HubSpot Live · 28:52
The funnel isn’t flowing. With most searches ending without a click and buyer attention scattered across every channel, the old marketing playbook no longer works. Join HubSpot CMO Kipp Bodnar and HubSpot Media Host Kieran Flanagan live from UNBOUND’s Main Stage as they break down what replaces it – and how marketers can turn AI into their competitive advantage.
TL;DR
A 29-minute main-stage keynote at HubSpot’s UNBOUND 2026 conference, published on the HubSpot Live channel on 18 September 2026. The speakers are Kipp Bodnar, HubSpot’s CMO, and Kieran Flanagan, host of HubSpot’s Marketing Against the Grain show. It launches their book Loop, out the following Tuesday.
It is the corpus’s first source on how AI is changing a marketing function. It is also the first to report an AI-search effect on one firm’s own numbers: HubSpot lost 80% of its web traffic, 140 million visits in a year, starting in May 2024, while growing leads by 20%. Their reading is that marketing has moved “from a world where volume of visits mattered to a world where value of visits matter[s].” The funnel has become an hourglass: awareness is scattered across AI search, creators and platforms that keep users in-app; getting visits “has never been harder”; and AI personalisation makes conversion easier.
Their replacement is a four-stage loop, Express → Tailor → Amplify → Evolve, run per campaign so that results compound:
- Express your taste and judgment to the AI, rather than outsourcing them to it.
- Tailor content to audiences and channels.
- Amplify it through the new buying channels.
- Evolve by learning fast.
The claim with most reach beyond marketing is Flanagan’s: “When the cost of production is free, judgment is the only thing that has a price.”
Key claims
1. The traffic collapse, and why leads still rose
The opening chart, in Bodnar’s words: “We saw an 80% drop in our traffic. We lost 140 million visits in a year. But we managed to increase our leads, our demand generation by 20%.” The explanation is a shift in buyer behaviour: “It used to be a buyer would land on a blog post, read all your blog posts, download your ebook … Now, they’re having those conversations with ChatGPT, with Claude, and the entire buying motion has changed.” The traffic that disappeared was low-value; the traffic that remained converts.
Flanagan’s audience poll makes the same point: few hands stay up for year-on-year organic traffic growth in 2026. “It’s not that we’re doing something wrong. It’s really that the rules of marketing have changed.”
2. Express: taste as the scarce input
The sea of sameness: “everybody here has access to the same tools, the same prompts, the same models … It’s not that the output AI has given us is bad. It’s just that we see it everywhere.” Most of the room raised hands when asked whether they had recently recognised AI-written content within a sentence or two.
The economic argument: for a hundred years the hard part of marketing was making things, and “that’s just no longer the case. AI can generate an infinite amount of stuff about anything … Making is no longer the problem. Choosing is the problem.” Therefore: “When the cost of production is free, judgment is the only thing that has a price.”
Taste is defined by example: Red Bull keeping a flavour customers dislike because the drink is about self-expression; Liquid Death selling canned water to punk kids; Ogilvy’s eye-patched Hathaway shirt model. The counter-example is Jaguar’s car-less rebrand ad, which they say was followed by a 97% sales decline over six months. Their Jaguar causal claim is not substantiated in the talk.
The prescription: “AI does not have taste. AI will give you the median of everything it was trained on. And that’s where all of your competitors live … You never outsource your taste to AI. You instill your taste to the AI.” The instrument is a written taste profile with two parts:
- Customer taste: who the customer is, how they talk, what earns their trust, what makes them stop scrolling.
- Brand taste: brand, product, “why you” and “why now” stories.
The AI reads the profile before every generation. Unlike a brand style guide “that live[s] in folders that no one ever reads”, the profile is “a living breathing artifact” updated as the team learns. Their demonstration: a generic SaaS landing page next to one generated with the profile.
3. Tailor: the content deluge and the 2×2
Bodnar: “In the last 2 years, we’ve seen a 4x increase in AI assisted content,” without a matching rise in attention. Content half-lives have collapsed: under 24 hours for a LinkedIn post, half an email’s audience within two hours.
He sorts content output on a 2×2 of quality against volume:
| Low volume | High volume | |
|---|---|---|
| Remarkable | invisible excellence | remarkable and relentless (the target) |
| Unremarkable | dead zone | slop factory |
“It’s not enough to just create great content. It has to be great content at scale. That’s now possible because of AI.” “Remarkable” means unique data, customer stories and examples “that you couldn’t go and just make asking … ChatGPT or Claude.”
The worked case is Sabrina Ramonov, a solo creator with 3.4 million followers built in 28 months. She tests in two directions. Start long: one long-form video or newsletter cut into pieces across channels. Start short: test 20–30 quick ideas over a week or two, then build long-form around the few that take off. An AI system rewrites each winning piece per channel.
4. Amplify: creative is the new targeting
Flanagan’s paid-media argument: the ad used to be the expensive part, and AI makes it cheap. His example is PJ Accetturo’s AI-made ad for Kalshi during the June 2025 NBA Finals, made for “a couple of thousand dollars,” which he says outperformed big-budget brand ads.
The mechanism: platforms now target better than marketers do. HubSpot’s own tests found a 22% better ROAS with Meta’s automated campaigns over manual targeting, and 32% lower acquisition cost with Google’s Performance Max. So “creative is the new targeting”: supply enough creative variants and “allow the algorithms to pick who your audience is.”
The data point: same brand, budget and platforms, and going from 10 to 20 creatives a month raises ROAS by 65%. His “benchmark” is 20 creatives a month. Flanagan extrapolates this into an exponential curve (10 creatives returns $3 per dollar, 20 returns $6, 100 returns $30), which one data point does not support; see Scope and reliability.
The second tactic is to optimise for closed revenue, not form fills: connect the CRM and send first-party data to the ad platforms so they learn what a good customer looks like.
5. Evolve: sprints over quarterly campaigns
Bodnar’s parable is Samuel Langley’s $1.5-million, government-backed aircraft crashing twice into the Potomac, while the Wright brothers, on about $1,000 and cheap gliders, “iterated really quickly” and flew nine days after Langley’s second crash. The organisational move: “moving from quarterly campaigns and calendars to a sprint model.” HubSpot marketing now runs two- and six-week sprints. “If you run a two-week sprint, that gives you 26 opportunities to learn in one year. If you run a quarterly campaign, you have four.”
AI is what makes the cadence affordable, because it automates the data gathering, analysis and reporting deck that used to make each cycle expensive. They show their sprint template: photograph it, give it your campaign data, and an AI drafts the retrospective and proposes the next sprint.
6. The close
Roger Bannister’s sub-four-minute mile is the lesson that the barrier was mental, since a dozen others followed within a year. “Our traffic dropped 80%. We had no choice. I would strongly encourage you, if … those things haven’t happened to you yet, that you take the action … and you run your first loop.”
Neighbour sources
- The search funnel breaking. Ognibeni closed his May 2026 talk predicting that AI agents would kill search-driven e-commerce first. HubSpot reports that prediction arriving in its own marketing funnel, with numbers.
- Taste, from the same firm. In Anthropic’s HubSpot customer story, HubSpot said it picked Claude because “marketing’s all about taste.” Here taste becomes an operating procedure (the taste profile) rather than a procurement criterion.
- Taste as the durable skill. Mollick calls taste the most valuable skill of the AI era, and this talk is the marketing-practitioner version. Dell’Acqua et al. provide the experimental version: AI lifts the quality of what is generated, and human judgment keeps its value in choosing. That is Flanagan’s “making is no longer the problem, choosing is.”
What was actually ingested
The full auto-generated transcript, 223 segments, from the walk-on to the close. ASR cleanup covered the speakers’ names, UNBOUND, Marketing Against the Grain, Sabrina Ramonov, PJ Accetturo, Kalshi, Ogilvy, Liquid Death, ROAS, Performance Max, Advantage+, the Potomac, Kitty Hawk and Roger Bannister. YouTube’s auto-chapter labels had leaked into the caption text and were removed; they did not match the talk’s own structure. The slides are not visible, including the traffic chart, the hourglass funnel, the 2×2, the ROAS chart and the sprint template. Their content is known only from what the speakers say about them.
Dynamic-capabilities reading
contextual/external-triggers: the talk starts from a changing-consumer-behaviour trigger in the cell’s own sense. Buyers do their research in AI assistants rather than on vendor websites, and platforms keep users in-app, which cost HubSpot 80% of its traffic.digital-seizing/strategic-agility: the move from quarterly campaigns to two- and six-week sprints is the cell’s “rapidly reallocating resources” and “pacing strategic responses”, applied to a marketing function. Their rationale, 26 learning cycles against 4, is a pacing argument.digital-seizing/rapid-prototyping: the Wright-brothers principle applied to content and ads. Start short and test 20–30 ideas, double down on the winners, and ship 20+ ad creatives a month so the platform finds the audience. This is lean experimentation, with AI bringing down the cost of each experiment.- Roles override:
roles:replaces the cell defaults with[cmo, ceo, cso, product-manager]. The talk is addressed to marketers, and the inherited defaults for these cells (COO, CTO, innovation-lab lead, and others) do not include the CMO at all.
Linked entities and concepts
- Entities: Anthropic, OpenAI (as the destinations of buyer research: “conversations with ChatGPT, with Claude”)
- Concepts: durable-skills, generative-ai, enterprise-ai-adoption, agentic-web
- Dangling (single-source mention, deferred): HubSpot Live, Kipp Bodnar, Kieran Flanagan, HubSpot (as an organisation, now the subject of two sources; a candidate for an entity page).
Scope and reliability
A vendor keynote launching a book. HubSpot sells marketing software, the talk promotes Loop, and giveaways and subscription requests run through it. Discount it as advocacy.
The numbers vary in quality.
- Strongest: the traffic and lead figures (−80% visits, −140 million visits, +20% leads), which are HubSpot’s own reported outcomes.
- Unsourced: the 4× growth in AI-assisted content, the content half-life figures and the 65% ROAS uplift. They are presented as data without a source; the ROAS figure appears to come from platform or partner data that is not named.
- Unsupported: the exponential extrapolation (100 creatives returning $30 per dollar) does not follow from one comparison between 10 and 20 creatives. Treat it as rhetoric.
- Asserted: the Jaguar sales-decline story is asserted as causal and is not substantiated.
The +20% leads needs context the talk does not give. It is not said whether lead quality, the lead definition or paid spend changed over the same period. “Leads up while traffic fell” is consistent with the value-over-volume reading, but it does not establish it.