Day 1 was about product. Day 2 was about proof. Two sessions of hard data on AI search, a panel of marketing leaders from Zapier, Zoom, LinkedIn, and Sequel.io who have already reorganized around it, and Ethan Mollick to close. Very little of it was a Webflow feature. Most of it was about how buyers behave now, where budgets are moving, and what a marketing team looks like when part of it is software.

Here's what mattered on Day 2 and what it means for a marketing team.

‍

1. Buyers are 90% decided before they call

Oana Manolache (CEO, Sequel.io) put the number on the table: buyers are now 90% of the way through a decision before they talk to a salesperson. Not long ago the accepted figure was 70%.

Sydney Sloan (four-time CMO, now advising Scale Ventures) added the other half from G2's new buyer report:

  • More than 80% of buyers use an LLM as part of their research.
  • Five years ago they took seven to nine vendors into evaluation. Now it's three. Sometimes one.
  • If you're not in the answer when the shortlist forms, you're not in the deal.

Webflow's own benchmark from 2,000 US brand sites: the median company shows up in 16% of the AI answers it would want to be in, and gets cited with a link 6% of the time.

What it means for the site. Manolache described two jobs. First, inform the LLMs so you appear when the question gets asked, which means the site has to be structured so machines can parse it. Second, when the human finally arrives, they've already done the research, so the site has to close the credibility gap fast, not walk them through 50 product pages. Sloan said G2 rebuilt part of its site as a prompt for that reason: "They're not going to come in and navigate after they just got an amazing answer."

2. Webinars are back

The channel conversation was the most surprising part of the leaders panel, because it went backwards.

Zoom pulled "a pretty substantial chunk" of media out of traditional search advertising and moved it into two places: things that influence AI citations, and community. In-person events, virtual events, creator partnerships. Whitney Magnuson (head of brand and media) was clear this wasn't a test: "The results we were seeing on the page told us we had to make that change. We couldn't wait for the next algorithm shift."

LinkedIn is shifting spend toward creators, because real people talking about real experiences are outperforming traditional ads, and toward in-person and online events, which Heather Freeland (Chief Brand Officer) said are pulling higher engagement than they've seen before. Dave Steer's summary from the Webflow side: "Webinars are hotter than ever."

On attribution, Magnuson said the thing most marketing leaders won't: building a single model for everything sets you up to fail. First-touch, last-touch, and multi-touch each have a job, and the team needs to know when to use which. Ethan Smith (Graphite) had the data point that shows why. Last-touch attribution said 0.9% of conversions came from LLMs. A self-reported "how did you hear about us" question said 9%. Ten times the difference, because most AI answers are zero-click and the buyer arrives through a branded search.

What to do. Add "how did you hear about us" after the conversion, and if the answer is AI, ask which one and what they typed. And look hard at whether your event and webinar budget is sized for a world where people are craving the human version of everything.

3. Trust is the metric

Manolache's other point landed with the whole room. She was on the phone with a payroll vendor recently and, mid-conversation, asked the person whether they were human. "The person felt a little offended. But it is the reality we live in."

Buyers can't tell human from bot in an email, a video, or a support chat. So the biggest thing a brand can do right now is build trust, and she named three ways a website does it:

  • How the brand shows up. Consistency, craft, the details that signal a real company.
  • Original content. Things that couldn't have been generated. James Cadwallader (Profound) called it "non-commodity content" and said the AI visitor doesn't respond to the shiny picture or the flowery language either; it responds to facts it can't find anywhere else.
  • Customer proof. Real relationships with customers and partners, made visible.

LinkedIn is measuring the same instinct from the other side. Its "Seems like AI slop" button, launched in August, has been clicked over a million times. AI-content signals in the feed are down about 40% since April. Freeland's read: the next generation "is craving authenticity."

Mollick put the risk in one line: "Raise your hand if you're sick of the LinkedIn post that all starts the same way." AI is a good writer and not a diverse one. If every brand uses the same tools the same way, everything evolves to the mean.

4. Fix fundamentals

The AEO panel's premise: a year ago everyone had an opinion about AI search. Now there's data. Guy Yalif moderated Mike King (iPullRank), Ethan Smith (Graphite), and Sloan, and the data mostly killed the loud takes.

"Everything changed." The 2,000-site study says the opposite. What's holding brands back is old:

  • 62% of sites have 10% or more broken internal links
  • 60% are missing SEO metadata on 20% or more of pages
  • 73% lack widespread plain-text brand mentions
  • 57% aren't answering the questions prospects actually ask

Yalif: "How are we in 2026 and both of those are true? If you've got broken links, don't bother with anything else."

"SEO is dead." King: "SEO is deprecated." The tactics still overlap today. The future needs engineering skills, because agents will ingest, remix, and serve your content more than people will visit your pages. Don't fire an SEO team that's evolving.

"Go all in on AEO." LLM referrals are mid-single-digit percent of traffic for most brands, converting well. Yalif's rule: treat 5% like it's 15 or 20%. Not 100%.

What actually moved numbers, per the panel: schema (a recipe-schema test lifted traffic 30%), internal links (25 to 50%), and page speed, because ChatGPT fetches pages in real time and drops slow ones from consideration. King fixed that for a large storage company and AI visibility rose about 40%.

On content, Smith's advice was the sharpest: stop writing "what is X" glossary pages, because AI Overviews ate that traffic. Expand comparison, use-case, product, help-center, and integration content, which is what branded and tie-breaker prompts pull from. And his data says lightly edited AI-generated content works in AI search today, doesn't work in Google, and will stop working when the models fingerprint it. King disagreed but didn't get the floor.

Freshness came up from a different angle in Cadwallader's session: half of everything AI cites is less than 13 weeks old. Profound now tracks citation decay as a metric. The five-year-old blog post that carried your rankings isn't carrying you anymore.

5. Human moments, agentic momentum

Every leader on the panel had reorganized the team in the past year. The approaches differed. The line they drew between human and agent work didn't.

Sequel.io mapped the problems before the tools, gave the team near-unlimited budget to experiment, and only added guardrails once something proved out. The internal theme is "human moments, agentic momentum." Manolache's non-negotiable: strategy and the original idea stay human. "AI will give you advice based on what's been done in the past." Everything after the idea is fair game for agents.

LinkedIn's first year was "a thousand flowers blooming": everyone told to use AI, lots of good work, no scale, no governance, no measurement. This spring Freeland called a timeout and created six AI-transformation workstreams by marketing function, each with a regular readout. Her image: from a wildflower field to the tulip rows in the Netherlands.

Zapier went the other direction: a "red alert" in Q1, three weeks for every marketer to learn Cursor or Claude Code with coaches available, then a two-week buildathon. Everyone shipped something. Dan Slagen (CMO and Chief AI Transformation Officer) reviewed 60-plus builds by hand. The team sorted itself into three tiers, and leadership now assigns real problems by tier.

Zoom built an AEO task force whose agents review every piece of content before it publishes, and a brand agent anyone can ask, "Am I speaking like Zoom right now?"

Mollick's framing for why the human part matters more, not less: the O-ring model. Work isn't additive. A few things have to be right, and those are the craft. "As the AI does some stuff better, that leaves holes where human intervention or human diversity matters a lot."

The cost nobody budgeted. Magnuson: tokens. During the experiment phase the bill sat with the CTO. "The bill's coming due," and marketing will have to justify which workflows earn their cost. Slagen's prediction for next year was the same from the other side: CFOs saying no to more spend without results. "Not just spending so we don't fall behind."

The new org chart. Manolache: experienced managers with no human reports, managing a "ton of agents." Entry-level marketers learning management by directing them.

6. The agent handoff

The one how-to session that belongs in this list. Emma Genesen and Adam Lehman walked through the Figma-to-Webflow workflow with an agent in the middle:

  • Connect the Figma MCP and the Webflow MCP to an agent in Claude or Cursor.
  • Point the agent at the Figma file. It reads the design, writes a design-system doc (colors, type, spacing, components), and saves it as Agent Instructions so it's reusable.
  • Ask for a page. The agent builds it in Webflow using your components and design system. About 20 minutes of setup the first time, about 15 minutes per page after that. An about page took three minutes at the end.

The designer-to-developer handoff, the one that adds a week to every landing page, is becoming a designer-to-agent handoff. It works today, on the current Webflow, not on Source. Two footnotes: Webflow's MCP is free and scoped to your plan; some other vendors charge for agent access, so budget for it.

‍

‍

‍