“Claude told me to sign up; I don’t know what your service does.”
Google is sending less traffic than it did a year ago. Your potential customers are finding and evaluating products inside ChatGPT, Perplexity, and Gemini. And most founders are still running the 2021 playbook. This week, we ran a working session on what’s actually replacing it.
On the panel: Justin Coons (full-stack growth at Twilio, DigitalOcean, and LaunchDarkly, now leading Venture5 Digital) and Andy Hattemer (Head of Marketing at Neon, acquired by Databricks last year, after five years scaling DigitalOcean’s developer knowledge base to 5M monthly users).
A few things stood out. Andy’s “shelf space” framework (borrowed from grocery stores of all places) explains LLM visibility better than any dashboard currently can. Justin walked through a cautionary tale of a company that lost 98% of its organic traffic overnight. Both panelists were uncomfortably direct about why the GEO measurement tools everyone is buying right now don’t actually measure anything useful. And a single data point from Neon, dropped casually in the middle of the conversation, recalibrates where growth traffic is actually coming from in 2026.
Introducing Venture5 Digital
Venture5 Digital provides conversion optimization, organic search, and AI search visibility for venture funds and the startups in their portfolios. Built by the team that grew Venture5’s web properties to over 1 million page views a year and onto the first page of Google.
The “shelf space” strategy still wins (and turns out to work for LLMs too)
Andy’s core framework for growth at Neon is borrowed from an analogy someone at MongoDB coined years ago:
Someone at MongoDB coined this, the shelf space strategy. Think of it like a grocery store.
The grocery-store mental model is straightforward. Manufacturers are always competing for prime shelf space, end caps, and eye-level placement. In a developer-tools world, your shelves are the adjacent places your buyer is already showing up. For Neon, that means application frameworks like Next.js and Django, app platforms like Vercel and Cloudflare, and the documentation, SDKs, drivers, and communities that live around them. The goal is maximum visibility at every point where a developer is making decisions that lead to a database.
The unexpected payoff is that the same strategy, built for SEO, also happens to work for LLM visibility:
Turns out that also works really well for LLMs, because they’re just operating off of tokens.
Andy’s theory is that LLMs don’t yet weight credibility the way Google’s PageRank did. They ingest a broad corpus and surface companies based on how often they see them mentioned alongside relevant topics. Neon got well-represented inside both OpenAI and Claude’s pre-training data, and Andy chalks a lot of that up to the distribution work the team was already doing for other reasons.
Don’t try to outsmart the engineers at Anthropic
Both panelists pushed back hard on the emerging cottage industry of AI optimization hacks. Andy framed the choice bluntly:
Are we really going to outwit the 500 cracked engineers at Anthropic and OpenAI with our little hacks, or should we actually focus on solving problems for users and helping them?
Justin came at the same point from a different angle. The temptation right now, especially for founders with a little Claude budget and no content team, is to generate thousands of SEO-optimized pages and see what ranks. His view on why that’s a losing move:
There’s nothing new that can come from an LLM output.
A startup’s job is to bring something novel to the market. Templated content produced by the same models your competitors are using does the opposite. It also runs real risk on the SEO side. Justin pointed to ClickUp as a cautionary tale, which lost roughly 98% of its organic traffic after leaning into “best X for Y” content plays that ran afoul of Google’s helpful content classifier.
The through-line here is familiar if you look at how DigitalOcean actually won. The community site wasn’t engineered to dominate search. It was engineered to help developers solve real problems, which Google rewarded anyway. Andy’s read on where we are now:
I feel like we’re back to the old hacks.
The temptation to cut corners is bigger than ever. The advice is the same as it was in 2015.
The measurement problem is real (so talk to users instead)
One of the more honest moments of the conversation was when Andy described how unreliable the measurement layer for GEO still is. Neon tried Profound, one of the more popular tools in the space, and hit the same wall everyone else hits. The tools scrape your site and hand you a list of prompts they think your audience is typing into ChatGPT, which you then track your visibility against.
They scrape your site, and then they suggest prompts that they think your audience is typing in, and that you should track your visibility on. And that’s a huge assumption.
Those 20 fixed prompts are not what your real users are actually doing. Andy’s team does use the tool for a shareable dashboard, but the real signal comes from talking to customers directly. Every one of the roughly 6,000 people who sign up for Neon each day gets a plain-text email from the head of product asking if they have feedback, good or bad. Many of those turn into calls. And what those calls consistently reveal is just how much of the buying process has moved inside the LLM:
So many today say, Claude told me to sign up. I don’t know what your service does.
That one data point captures a lot about where growth is headed. Roughly two-thirds of Neon’s traffic is now coming from LLMs. A meaningful share of new users arrive already decided, having never read a marketing page, because Claude told them to:
It absolutely makes sense to think of them as a main interface for your content and your business.
Justin’s team did a similar exercise at LaunchDarkly, running a user research project with a few dozen developers who use AI in their workflows. They found two distinct behaviors:
You could broadly divide it into two categories. One is information and problem solving. The other one is recommend a platform for me.
Those two behaviors require completely different content strategies, and no off-the-shelf tool will tell you which one your specific buyers are doing. You have to go ask.
A 90-day playbook for teams without a growth department
For founders or solo marketers trying to figure out where to actually put the hours, a few specific tactics came out of the conversation.
Start with the user, not the robot. Justin’s advice for anyone tempted to over-index on GEO tactics:
When you begin thinking about, I need to show up in AI, or I need to show up in search tools, do you then start optimizing your content and your experience around the robot rather than the person?
That inversion is the trap. Content built around real user problems is more insulated from algorithm changes than anything engineered for the current version of a model.
Put your best stuff in the first third of the page. Justin cited studies showing LLMs weight the top third of a page most heavily when deciding what to cite:
The most valuable content on any given web page is in the first third.
His rule of thumb for any page you want cited is to put the bottom line up front, keep the goodies above the fold, write simply, and favor clear questions and answers.
Rethink your docs for agents. Andy’s team has been iterating specifically on how Neon’s documentation gets consumed by LLMs and coding agents. Markdown gets served automatically to agents to cut token count. FAQs are becoming a bigger share of the content because LLM queries tend to be more question-shaped than Google queries ever were. New protocols like llms.txt get implemented as they emerge. Andy also admitted his prior skepticism of LLM-generated content is softening for this specific use case. If you’re starting with comprehensive first-party docs and applying tight guidance, the FAQs you generate can be as useful as anything written by hand.
Do the user research SEO tools can’t do for you. The pattern Justin ran at LaunchDarkly (interviewing a couple dozen developers about how they actually use LLMs) is replicable at any stage. It’s the kind of primary research that becomes more valuable, not less, as the measurement tools get noisier.
Venture5’s Take
If there’s one throughline from the conversation, it’s that the growth playbook for the AI era looks suspiciously like the growth playbook that has always worked, and the companies winning the next era of growth look a lot like the companies that won the last one. They have a deep understanding of the user, broad distribution across the places that user already spends time, and content that’s genuinely useful rather than engineered for algorithms.
What’s changed is the surface area. Two-thirds of Neon’s traffic didn’t come from LLMs two years ago, and many of those users now sign up before they’ve ever seen a marketing page. The work is less about rewriting the playbook and more about redistributing effort toward fundamentals that were always going to matter.
The full conversation goes deeper on documentation strategy, the specific tradeoffs with tools like Profound, how to think about agentic workflows as a new interface layer, and a live walkthrough of a real page that Justin optimized. Worth the 45 minutes if growth is on your plate.
