For our first four years, the growth engine described in Chapter 3 kept doing its job. We wrote genuinely useful content, earned authoritative backlinks, ranked for the questions our future users were asking, and let the whole thing compound. Then the ground started to move.
The Ground Shifted Under SEO
By 2025, a growing share of the people who used to type questions into Google were asking AI assistants instead. ChatGPT, Claude, Perplexity, and Google's own AI results were answering questions directly, in full sentences, often without the user ever clicking a link. The blue links we had spent years climbing toward were being summarized away.
For a company whose acquisition strategy leaned heavily on compounding search traffic, this was an existential question. It was also a familiar one. We had seen a platform shift up close before, in late 2022, when ChatGPT appeared and we shipped an extension within days (Chapter 4). The lesson from that experience wasn't "AI is coming." It was "when the interface changes, the people who adapt early win attention that latecomers have to fight for."
So instead of mourning the decline of the ten blue links, we asked the same question we asked in 2022: what does this shift make newly valuable, and how does it connect to our mission?
From Ranking to Being Cited
The answer we arrived at has a name: Answer Engine Optimization, or AEO.
In the search era, the goal was to rank: get your page into the top results and earn the click. In the answer era, the goal is to be cited: when an AI assistant composes an answer about highlighting, learning techniques, or research workflows, you want it drawing on your work and pointing readers back to you.
What struck us was how little the underlying principles changed. Answer engines, like search engines before them, reward sources that are genuinely useful, clearly structured, and consistently trustworthy. The fundamentals we had practiced since Chapter 3 (real value, clean structure, patient compounding) still applied. What changed was the reader. We were no longer writing only for humans who skim, but also for models that parse, weigh, and quote.
That reframing turned an existential threat into an execution problem. We knew how to solve execution problems.
Deep Dive: Betting Again on Long-Form
Our biggest AEO investment was content, and it looked almost old-fashioned: a library of long-form, evergreen guides we call Deep Dive.
We built out more than 100 in-depth articles covering the territory our users care about: AI tools and how to choose between them, learning science, note-taking and knowledge management, reading workflows, research methods. Each one is structured the same way: a clear table of contents, key takeaways up front, FAQ sections, and consistent formatting that both a human skimmer and a parsing model can navigate.
Then we applied the multiplier we discovered back in Chapter 2, deliberately this time. Every article is translated into 7 languages. Where community members once translated a single press article for us, we now run translation as a standard part of the publishing pipeline. One well-researched guide becomes seven entry points in seven markets.
The bet is the same compounding bet as before: every guide is an asset that keeps working, except now it works in two ways. It ranks in what remains of traditional search, and it gets cited by the answer engines that are replacing it.
Making Glasp Machine-Readable
Content was half the work. The other half was making Glasp itself legible to machines.
We added an llms.txt file to the site, a plain-language guide that tells AI crawlers what Glasp is, what lives where, and what matters most. We expanded structured data (JSON-LD) across the site, so that articles, books, quotes, and profiles describe themselves in a vocabulary machines understand without guessing.
Then we went a step further than describing ourselves to AI, and connected to it. We built a remote MCP (Model Context Protocol) connector, so that users can plug Glasp directly into their AI assistants. With permission, an assistant can search your highlights, recall what you've saved about a topic, and bring your own collected knowledge into a conversation.
This is worth pausing on, because it reframes what "distribution" means. In the search era, your product surface was your website and your extension. In the answer era, your product surface includes the AI assistants your users already talk to every day. Being present there isn't marketing. It's product.
And it ties back to the mission in a way we found genuinely exciting. We've always said the knowledge you collect should outlive the moment you collected it. An assistant that can draw on your highlights years later is exactly that promise, kept through a new interface.
Beyond Text
Answer engines don't only read articles, and neither do people. So we started turning our strongest Deep Dive guides into other formats: podcast-style audio conversations and video versions distributed on YouTube.
This was the "create once, publish everywhere" principle from our resource-efficiency playbook, pointed at a new goal. The same research that produced a written guide becomes something to listen to on a commute and something YouTube surfaces to learners who would never have found the article. Each format reinforces the others, and each is one more way to be the source an answer draws on.
The Proof: 500 to 19,000 Daily Sessions from ChatGPT
Strategy is cheap. So we measured.
At the start of 2026, ChatGPT was sending us 517 visitors a day. We made a deliberate bet: stop investing in direct SEO and run the AEO playbook as a series of experiments on our largest content surface, a corpus of more than 400,000 YouTube Q&A pages.
The first decision set the tone: measure from our own server logs instead of subscribing to tools that poll the models from outside. Cloudflare's AI crawler logs and Search Console told us, deterministically, which pages AI bots actually fetched and how often. That data turned guesses into a roadmap.
The experiments themselves were almost embarrassingly concrete. Pages that bots requested often had question-form titles matching how people phrase prompts, so we rewrote titles as questions. They had prose summaries up top, around 130 characters, that worked as standalone answers, while ignored pages carried 14-character fragments, so we rewrote our TL;DRs to hold the complete answer even if a model reads nothing else. We mined the 404 errors AI bots left behind, tens of thousands a week, as a literal list of pages users were already asking for, and built them. We deleted tens of thousands of dead pages with zero Google and zero bot interest, and indexation on everything that remained improved. And pages already earning Google clicks were locked against rewrites, so the new channel never cannibalized the old one.
Four months later, on May 5, ChatGPT referrals hit 19,129 daily sessions: 37x growth. The striking part is that AI bot crawl volume stayed flat the whole time. The same bots were visiting. They were simply finding more answers worth citing. We shared the full playbook in a guest post on Sean Ellis's newsletter, in the same spirit as everything else in this story: what we learn, we publish.
What We Learned
The AEO era is young, and we don't pretend to have it fully figured out. But a few lessons already feel solid.
First, AEO is not a replacement for everything we knew. It's SEO's principles aging into a new interface. Genuine value, clear structure, and earned trust still win. If you built your growth on tricks, the answer engines are bad news. If you built it on substance, they're an opportunity.
Second, being early matters again. The window we exploited when ChatGPT launched has a sequel: most companies are still treating AI search as a curiosity, which means the citations are still up for grabs. "Be first, even if imperfect" survived the platform shift intact.
Third, what compounds has changed shape. It used to be rankings and backlinks. Now it's being a citable, structured, trustworthy source, in text, in audio, in video, and through protocols like MCP that put you inside the conversation itself.
The deepest lesson, though, was about identity. When the way people find information changed, we didn't have to change what we are. A platform built on capturing and sharing knowledge openly turns out to be well positioned in a world where machines are constantly looking for knowledge worth repeating. The mission aged well.
That confidence in substance over tactics led us somewhere we never expected a two-person startup to go: publishing original research. That's the next chapter.