We got to attend Zero Click SF and we have to be honest – we weren’t fully prepared for how good it was. Profound put together something special in San Francisco. Sharp speakers, a genuinely engaged room, and conversations that kept going long after the sessions ended. We were meeting people, trading notes, realizing how much AI and the search landscape has already changed and how few teams are actually keeping up with it.
We wanted to share the inside scoop. Here’s what we learned.
The framing that cut through
Profound’s CEO James Cadwallader opened the day with a question that genuinely stopped us: how do you tell the thing that knows everything something it doesn’t know?
That’s the actual problem. AI systems were trained on the entire internet and your brand is a rounding error in that dataset. Your six-month-old blog post isn’t new information. Your keyword-stuffed FAQ isn’t interesting. If you’re not saying something that AI can’t reconstruct from existing training data, you’re invisible.
Granular specificity, current content, original quality. Not a revolutionary answer, but an honest one. Most AEO conversations get stuck on tactics – schema markup, FAQ sections, citation bait. This cut deeper.
The agent angle here is worth noting. Cadwallader’s whole argument about needing current, specific, original content becomes even more urgent when you factor in that agents, not humans, are increasingly the ones doing the retrieval. An agent pulling information to answer a user’s question has no patience for a stale page. It just moves on.
The mechanic worth understanding: query fan-out
Mike King from iPullRank gave the most technically dense talk of the day and it was one of our favorites. His core mechanic, query fan-out, is something we hadn’t heard in depth before.
When someone asks an AI a question, the system doesn’t search for that question directly. It fans out into a series of sub-queries, pulls document fragments for each, and stitches the answer together. Each sub-query has an expected content format. If that format is a video and you don’t have one, you’re not eligible for that slot.
He called it a raffle. More sub-queries you rank for, more tickets you have. That framing clicked.
He also made a point most people in that room probably didn’t want to hear: most SEO tools are running on a lexical model that’s over a decade out of date. Google went semantic in 2012. The platforms running today are hybrid. The tools most marketing teams use haven’t caught up.
The agent connection: query fan-out is essentially how agentic systems work too. An agent handling a task breaks it into sub-tasks, each requiring different information from different sources. If your content only answers the top-level question and not the follow-on ones, you drop out of the picture fast. This is something we’re watching closely.
The stat that stopped us
LinkedIn’s Dan Morrill came out with data that genuinely shifted how we’re thinking about traffic metrics. AI overviews are now appearing on 47% of LinkedIn’s keyword landscape – up from 7% a year ago. Traffic from LLMs has more than doubled. Traditional SEO traffic and MQLs are down.
But conversion rates are up nearly 2x. Deal sizes are growing.
The buyers showing up are showing up later, more informed, and more ready. They did their research inside AI. They came to LinkedIn when they were close to a decision. That’s a fundamentally different funnel – and if you’re only watching MQLs, you’re going to misread it as decay.
The playbook he laid out: post consistently (2-3x a week), publish original long-form content (60% of cited LinkedIn content is articles or newsletters), and activate real people with real expertise. Posts with 10+ comments are significantly more likely to drive AI citations. AI weights identity and real dialogue heavily.
What we’re watching: agents doing B2B research are already pulling from LinkedIn heavily. Real identity, real expertise, real conversation – those are signals that hold up well when an agent is trying to figure out who’s credible in a category. The brands with active, substantive LinkedIn presence are going to have an advantage as agentic research becomes the norm.
The talk we’re still thinking about
Daniel Shin Un Kang from Expedia had one of the most refreshing takes of the whole day. Instead of pretending anyone has this figured out, he told the room straight up: budget for a 90% failure rate and build like a venture fund anyway.
His point: you can’t predict exactly how AI commerce shakes out. But you can identify stable hypotheses – things almost certainly true regardless of which outcome hits. Two of his:
- Agents are becoming gatekeepers. They’re deciding what information is relevant, what products to surface, and in some cases completing transactions entirely. Humans own the intent. Agents own the execution.
- Agents think differently than humans. Low latency, structured data, no emotional impulse. They’re also biased in ways we don’t fully understand and make unpredictable decisions. Marketing to them is a separate discipline from marketing to people.
His no-regrets move: become legible to agents. Make sure they can find you, read you, and understand what you do. That’s table stakes before any of the more sophisticated bets matter.
This is the framework we’d actually recommend building around right now. You don’t need to know exactly where agents land – you need to make sure you show up clearly when they go looking.
One thing we’re still chewing on
Josh Blyskal’s session was the one we kept coming back to in conversations afterward. The historical through-line he drew – Jim Simons in finance, Maxime Beauchemin building Airflow at Airbnb, Varun Anand inventing the GTM engineer at Clay – lands. Every major function eventually gets an engineering layer, and marketing has been waiting for one.
The examples on stage were mostly competitive monitoring and automated publishing – real, useful things, but still a narrow slice of what the role could look like at scale. The part nobody quite answered was how you know what to build in the first place. Practitioners figuring it out in the field will define this faster than any curriculum will.
That said, the agent piece is where this gets interesting. The vision Blyskal was describing – systems that monitor, diagnose, act, and scale without human intervention – is essentially an agent-first marketing operation. We’re watching to see how teams actually staff and structure around this over the next 12 months.
What’s worth acting on
- Think in sub-queries. What are all the questions your category triggers, and are you showing up with the right format for each one?
- Publish original analysis. Things that are actually new. Information gain is one of the highest-correlation features for AI performance according to Mike King’s research.
- Make your brand legible to agents. Structured data, clear product descriptions, consistent entity presence. This is foundational.
- Watch conversion rates alongside traffic. Buyers arriving through AI channels are often further along. The funnel looks different, it’s not broken.
The session nobody expected: OpenAI is building an ad platform
This was the session we didn’t expect. OpenAI presenting at a marketing conference is one thing. Walking out knowing they’re actively building an ad platform is another.
Pilots launched in early February 2026. A self-serve Ads Manager is going live this month – KYB verification required at onboarding. Results so far, per the session, are moving “really, really fast.”
Campaigns are built through prompts – brand voice, audience, objectives – and matching is semantic. The system reads the intent behind a user’s message and connects it to advertiser goals contextually. Their example: someone planning a trip asks about camping, the platform surfaces hiking gear and family activity ads. No keyword match. Intent match.
The trust architecture they’re betting on:
- Model separation: the AI cannot see ads unless a user explicitly engages. The model isn’t being influenced by spend.
- Visual separation: clear UI distinction between organic and sponsored. It won’t be hidden.
- Data privacy: conversations never shared with advertisers. All matching is internal.
- Optimized for user value, not clicks or time spent. Ad-free option available on Pro Plus.
Three things they said drive campaign performance: deep audience understanding (intent and context), strong product-market fit (the platform can’t compensate for a weak value prop), and ads that integrate naturally into the conversation.
What to do before access opens: get specific about who your product actually matters to, define your brand voice clearly enough to put it in a prompt, and build your organic AI presence first. Paid placement alone won’t do it.
What we’re watching: ChatGPT has richer intent data than any search engine ever has. People are talking to it – planning trips in detail, asking for product recommendations, describing problems at length. OpenAI knows what that signal is worth. The self-serve manager launching this month is moving fast, and this is a space we’re going to keep a close eye on.
Profound: context is what separates useful agents from generic ones
Profound’s second session was more product than theory and we appreciated that. The core point – without context, an AI agent produces output that could apply to any company – sounds obvious when you say it out loud, but most teams aren’t building with that in mind.
What they built is a marketing context layer connecting agents to your brand guidelines, sales call recordings, CMS content, and AI search data. The deployment model runs in three stages: Diagnose (surface insights, citation drop analysis), Act (draft and publish directly to CMS), and Scale (hundreds of agents in parallel via Profound Sheets). Two features coming: Background Agents that run continuously, and Profound Fact Check, which flags inaccurate AI responses about your brand – wrong pricing, outdated descriptions.
Twenty thousand production-level agents built by beta customers in six weeks. That’s the number that stood out. Worth watching where this goes.
Parallel: agents don’t browse like humans, and your site isn’t ready
Parag Agrawal – ex-CEO of Twitter, now founder of Parallel – said something early in his talk that we wrote down immediately. AIs will use the web 1,000x more than humans. The old bargain – publish, humans discover, click, convert – is under real pressure.
Agents don’t browse like people. Humans use short keywords. Agents use verbose, precise queries. Humans click through. Agents want content delivered directly. Humans respond to flashy UX. Agents want verifiable facts and authoritative sources. Your UX still matters for human visitors – but for agent visibility, structured and authoritative content is what counts.
His historical framing: every major web shift created net new winners who leaned in early. Video, short-form, social – each time, the brands that moved first built advantages that lasted. Agent-native brands have that same window right now, and it won’t stay open forever.
Webflow: the Frankenstack is killing your AI strategy
Dave Steer from Webflow named something that got a lot of knowing nods in the room: the Frankenstack. Teams organized by channel with separate tools and separate KPIs. No shared context. Scaling with AI becomes difficult because agents have no coherent picture of the brand to work from.
Their content playbook numbers from their own team: shifting from annual content refreshes to ongoing agentic updates drove 5x more updates and 42% traffic growth. FAQ sections with schema markup added 24% more impressions and three new AI citations. Old webinar recordings turned into citable articles.
The org model he proposed has four roles: Strategist, Creator, Connector, and a new one – the Orchestrator, who builds the agent systems that multiply the whole team. That last role doesn’t exist on most teams yet. It probably should.
Reddit: you’re a guest, not a host
Rob Gaige from Reddit gave the talk nobody saw coming and honestly it might have been the most fun session of the day. Most brand Reddit strategies fail because brands show up like they own the place. Reddit is built around what you say, not who you are. A Kardashian starts at zero karma just like everyone else.
The dinner party framing: do your homework before showing up, bring something exclusive, make others feel smart, don’t fight critics, act like you’re lucky to be there. Practically: max 3 posts per week, target 30 to 50 subreddits – the long tail is where most relevant queries actually live – and advertise first. Redditors expect it.
The agent angle here is real. Reddit content is heavily indexed by AI systems precisely because it’s authentic, conversational, and specific. Agents pulling research on products or categories are landing on Reddit threads constantly. A brand with a genuine, earned presence in relevant subreddits is going to show up in that research. One without one won’t.
G2: the only AEO metric that connects to revenue
Tim Sanders from G2 opened with a slide that made the whole room go quiet. 50% of software buyers now start research in AI chat. 54% use chat to build their shortlist. 9 out of 10 buyers pick their winner from the day-one shortlist.
If you’re not in the first answer, you’re probably not in the deal.
G2’s own results: rewriting content using real review verbatims in first-person got 40% of their citations coming from Learn G2. Adding context summaries to product pages drove 44% more citations. FAQ sections built from natural review language improved SEO and AEO at the same time.
The metric hierarchy: Mentions, Citations, Branded Citations, and then Winning the Answer – being the recommendation in a shortlist prompt. That last one is the only metric with a direct line to revenue.
One more number worth carrying: Tier 2 publications get 97% of AI citations. Tier 1 outlets like the WSJ block AI crawlers. If top-tier press has been your AEO strategy, the math doesn’t work.
The agent connection is direct here. When an agent builds a shortlist for a B2B buyer, it’s doing exactly what Sanders described – pulling from trusted sources, surfacing credible options, and recommending. The brands that show up consistently in that retrieval process will win deals before a salesperson ever gets involved. That’s the shift worth preparing for.
Our Takeaway
We walked out of Zero Click SF with more tabs open, more notes scribbled, and more things to look into than we expected going in. That’s the best thing you can say about a conference. The people we met were all wrestling with the same questions, and just being in that room for a day made the whole AI search landscape feel a lot more concrete.
This space is moving fast. Profound is clearly building toward something real, and the brands and practitioners in that room are ahead of most. If they do another one, we’ll be there early.
