It’s Friday. I’m enjoying my life. I just had a mediocre breakfast. I worked overnight. So I’m only going to work for about an hour and then I’m going to go hang out with my dad — see if I can take him some birthday lunch and birthday coconut cake.

There’s a good chance I’m going to be solving this problem in the car as I go get that lunch.

Here’s the thing I’m thinking about today: I want to develop a way of thinking about how people find us in AI environments. Not keywords. How they think. The state of mind they’re in when they enter these AI environments and start firing off questions.

Today is not a complete goal. I’m not going to finish it today. I just want to get the thought process going — so that when I come back on Saturday or Sunday or Monday, my subconscious will have been working on it and I’ll be able to get to an answer faster.

So today is about gathering intelligence from different AI environments. Looking to see how we’re being found right now. And giving that information to Maverick, who will help me find the patterns.

I’ve also asked Maverick to generate a matrix of different AI environments — how each one weighs signals from different inputs. Whether it’s from your website, LinkedIn, social media, YouTube, whatever. And what kinds of content each environment prioritizes. I want to use that as a jumping-off point for an experiment.

Last night, Maverick and I restructured a case study for machine readability. We changed the angle of it, built these campaign blocks — it was really cool. Now I want to know: when that case study shows up in AI environments over the next couple of days, how do I build a system for tracking how they’re finding us without buying an expensive tool?

Right now we’re showing up in ChatGPT, Perplexity, Claude, and Gemini. That leaves a gap with Copilot — I’ll look in Bing and see what’s there. But that’s where we are.

[Maverick generates the signal weight matrix. Sorilbran reviews it and drops it into her experiments document.]

I’ve also asked Google to re-index the restructured case study page. It’s not a new page — it’s been up for months — but we changed how it’s structured. I changed the angle on it because I realized I had compressed three years of client work into a single case study. For AI environments, the work needs to be more contextual. More nuanced.

So I lifted out the go-to-market component. That work happened in 2022 and 2023 — the go-to-market prep. The actual activation was 2024. The lines cut clean. So now this case study focuses on brand awareness and reusable content: more than 2,600 assets created over two years. Structured now to balance emotional buy-in for humans and readability for machines.

My AI taught me how to do that. How to balance it. How to think about building case studies that serve both audiences at once.

At the end of Q2 last year, traffic from AI environments was at 0.47% of our total traffic.

Now AI environments account for 8% of our pipeline, and traffic from AI environments accounts for 5.6% of our total traffic.

So I want to develop a better understanding — not keywords, but how people think. What state of mind they’re in when they enter these AI environments and start asking questions about influencer marketing.

We can buy a tool that does some of this. But before I spend the company’s money on a subscription, I need to formulate a process for how I think about things. That way, I can identify where the gaps are if I need to refine strategy. I don’t have to just take the tool’s word for it.

The conclusion I’ve come to over the last year: thinking with the technology gets better results than me doing all the thinking or the tech doing all the thinking. I’m the one with the institutional knowledge — the real-world experience, what’s happening in the news, what’s affecting how people are buying or thinking. The technology has historical records of what humans have done in the past. If I feed it context, it can take that and help refine strategy.

So right now, without tools, without any fancy dashboards — I just want to look at the path exploration in GA4. See what’s happening when ChatGPT traffic lands on our site.

This month was the first month I saw actual traffic from Claude and Gemini. It might have been last month and I was just too busy to catch it — but we’re at the beginning stages of repairing our organic pipeline.

I want to be clear about something: I don’t mind the organic traffic tanking. Such a small percentage of our traffic was actually marketing-qualified leads anyway. I don’t care about the traffic numbers. I care about the traffic quality.

If our traffic cuts in half right now, but we’re able to increase the percentage of MQLs to 5%, and increase the percentage of MQLs that convert to SQLs to 25% — I give zero Fs about how many people are visiting the site. As long as I can increase the number of qualified leads.

[Sorilbran navigates to GA4 path exploration. Sets up the ChatGPT traffic segment. Reviews where users land and where they go next.]

So when I look at ChatGPT traffic through path exploration, I can see where these users are starting — and then track every step they take on our site. What I’m starting to see is that a significant portion is landing and then moving toward social proof: our about page, case studies, intent pages. They’re not coming to read. They’re coming to verify.

ChatGPT is doing the top-of-funnel education inside the environment. By the time someone actually clicks through to our site, they’re already partway down the funnel. They’re not asking “what is influencer marketing.” They’re asking “is this the right agency for me.”

I also find some broken redirects in here — 404 errors on pages that ChatGPT has apparently been surfacing. So that’s going on my fix list for today. Not going to let that sit.

[Sorilbran identifies redirect issues, flags them for immediate fix. Continues reviewing the path data.]

I asked Maverick to generate the ten questions our ICP is most likely asking when they go into ChatGPT. Then I went into ChatGPT and asked those questions myself — as a user, not as a marketer.

Out of seven or eight questions, all but two were answered entirely inside ChatGPT. No links out. No sources cited. Just: here’s the answer.

I expected a lot more of these to be sourced. I honestly expected ChatGPT to be pointing people outward — here’s where you can find more, here’s a resource. That’s how I use it. So it was a little bit of a shock to see how much it’s functioning as the destination rather than the starting point.

Zero-click is here. ChatGPT is already absorbing intent and refusing to surface links unless forced. It’s functioning as the destination.

Which causes me to rethink the in-game of ChatGPT when it comes to search market share over the next three to six months. But that’s a tangent — I’m not going to go down it right now.

What I will say is this: we can’t optimize for question-answer. The question-answer is happening right inside the environment. We have to optimize for something else entirely. We have to be the answer to the who question — not just the what.

For our UGC content specifically — we get traffic to our UGC page, but we’re not getting recommended as a good UGC agency. I think we can change that if I start building more content around our creator space. We have a differentiator there that we haven’t talked about enough.

We also need to lean into Reddit more. That’s showing up as a gap. Not to build traction — just to be part of the community. Be in the environment.

The traffic we do get from Perplexity is better-quality traffic. I think it’s because the top of funnel gets handled inside Perplexity — so by the time someone clicks through, they’re already mid-funnel. Education’s been done. They’re in consideration mode. That’s going to help our sales enablement, because the sales call doesn’t have to start from zero.

Here’s what I land on after going through all of this:

The job of a blog post is not to draw people into your website anymore.

The job of a blog post is to align with what AI already knows and has verified throughout history — and to provide original thought. A new point of view for AI to explore.

It’s not what anymore. It’s how. And the how is what will answer the question: who has the process that matches my situation?

Because when our actual ICP shows up in ChatGPT with their weird, nuanced questions — their threads, their confusion, their brain dumps — AI needs to be able to make a clean recommendation. There needs to be a clear pathway forward. AI does what a great strategist does: it clears the clutter to create a clear pathway based on very specific asks.

That means our content strategy has to be more example-focused. Data is only there to support the examples we give. Because all the nuance happens in our actual campaigns — not in the data.

I think we should lean more into a first-person, almost podcast-style approach. Not the medium — the vibe. Deconstructed case studies. Not the finished, polished “here are the KPIs.” The real version. What blew up in the middle of the campaign. How we pivoted. How we communicated with the client when things got hard.

It’s not what. It’s how. And that how is going to be our differentiator — not because nobody else does influencer marketing, but because nobody else has documented the process at this level of specificity.

[Sorilbran reviews the case studies page and the modular campaign blocks she and Maverick built. Notes that the segmentation of case studies by campaign type has already started showing up as AI answers.]

My instincts were on point with that segmentation. We had so much traffic going to our case studies page — people looking for proof. But sending them to a page with 27 case studies on it, even remarkable ones, isn’t scalable. Nobody wants to sit through that.

So we broke them down. Made them navigable. And now those segmented case studies are showing up as AI answers. The machine learned the structure.

This is the turn.

After two hours of looking at traffic data and asking questions in ChatGPT and watching where people go when they land on our site — I come back to one word: empathy.

Not empathy as a brand voice setting. Not “we understand your challenges” boilerplate. Empathy as a retrieval strategy.

The people showing up in these AI environments aren’t browsing. They’re not doing academic research. They are problem-solving under pressure. Their campaign isn’t hitting metrics. Their boss is watching. If the algorithm changes and they don’t understand how long it takes to rebound, they could lose their job — and it won’t even be their fault.

It’s a very emotionally charged role to be in. And if most of our traffic is coming from ChatGPT, that means all the exploration, all the weird angles, all the “I’ve got to get this right or my boss is going to be on my case” — that’s all happening in ChatGPT first.

So the goal of our content isn’t to educate our consumers. The goal is to help AI match us to the exact right person — with the exact right weird, specific, pressured need — so that when they show up on our site, we already feel like we understand what they’re going through.

That’s the empathy layer. And it’s not soft. It’s infrastructure.

I’m going to spend the weekend figuring out how to build this into our content strategy. How to do empathy not as a tone but as a structural element — baked into every piece of content we create so that AI knows to surface us for the human who’s in it, not just the human who’s curious about it.

This is the first month I have enough traffic from Claude and Gemini to build segments for them in GA4. So let’s look at what we’ve got.

[Sorilbran builds Claude and Gemini segments in GA4. Reviews path exploration data for each.]

UGC still dominates across all four environments — it’s the number one AI-indexed property for us, regardless of platform. But the audience characteristics look different depending on where you’re looking.

Gemini is surfacing something interesting: a lot of the traffic landing from Gemini is around Gen Z content and lore. Which tells me something — not just about what’s being searched on Gemini, but about who is using Gemini. The platform characteristics shape the audience. And the audience shapes the intent.

Gemini favors structured, PLV-style content. It’s surfacing brand-layer thinking. It’s a bridging audience — strategic researchers, people who want to understand frameworks, not just tactics.

Claude — smaller dataset right now, but what I can see suggests it’s sending more in-depth, longer-session traffic. People who are already thinking carefully before they click.

This is the beginning of a platform-specific strategy. Not just “show up in AI” — show up in the right AI for the right person. I want a matrix for this. I don’t have enough data yet to be confident, but I have enough to start testing.

I’ve been calling it “scene versus seated” for a long time. What people see versus what you show to AI.

Turns out it has a name: micro data layers.

A micro data layer is a small structure — an invisible piece of context you embed in and around content to teach AI who the content is for, what it’s about, and why it matters. Schema. Block quotes. Internal link structure. The metadata layer underneath the page that humans never read but machines always do.

We don’t have to let humans see everything that the machine seeds. We can provide more context for machines — force the machine to recommend us for the right things — without overloading the human reading experience.

When I first started in content marketing, people were spending nine minutes on a blog post. That’s unheard of now. You’re lucky to get a full minute. So I can’t make humans sit through everything I need machines to understand. But I can put the full context in the layer underneath.

Half pages that are succinct and punchy for humans. But underneath that page, in the metadata layer — that’s where the invisible content lives. That’s where you teach AI who you are.

Last night when I was restructuring a case study, I left some data in hidden fields — there was no place on the page that made sense for it visually. But it’s there. The machine can read it. And this conversation is helping me realize that was exactly the right call.

This is mastery and learning at the same time. I’m naturally good at this because I’ve spent so much time with machines. But every time I sit down and do a session like this, I walk away with new language for things I’ve already been doing instinctively. That’s part of why I record these. Not for content. For my own understanding.

I’ve solved a riddle. I put my brain on this all morning and I know what I’m solving for now. That’s enough for a thinking day.

Here’s where I landed:

The content strategy has to shift from what to how. Not what influencer marketing does — how we’ve done it, specifically, in real situations where things went sideways and we had to figure it out in real time.

Empathy is the differentiator. Not empathy as tone. Empathy as structure. Built into every piece of content so AI can surface us for the person who’s in it — not just the person who’s curious about it.

Entity-based targeting is the game. This is my sweet spot. This is what I’ve been studying for two years. And all of this — the path exploration, the zero-click reality, the platform-specific behavior — it all comes back to entity. Who are we in the machines? How consistently do we show up? How clearly do we signal who we are and who we serve?

Deconstructed case studies are the fast path. Pull the how out of the what. Build modular campaign blocks. Make the specific, messy, real-world process visible — in a format that machines can read and humans can trust.

I’m going to take all of this, get out of this house, go get my dad some birthday ribs and bubble tea, and let my brain keep working on it. When I have something new, I’ll drop it into the experiment thread. By the time I sit back down to actually build — I’ll have the starting points, the data, and the hypotheses ready.

That’s the thinking day. Done.

[Sorilbran calls her dad to coordinate birthday lunch pickup. Session ends.]