Glasp’s new case study will be shared widely, because the headline is irresistible, 37x ChatGPT traffic. The more useful number is smaller, and less flattering, about 2x after controls. That is the point. It is the difference between growth theater and measurement.
For founders at DA 0, the temptation is to copy the wrong layer. You do not need a 400,000-page corpus to learn from this work. You do need a way to see what AI crawlers are doing, what they cannot reach, and which pages already match real demand.
This piece closes the gap Glasp leaves open, what transfers to a small site, what does not, and what to set up this week without buying an AEO dashboard.
The numbers first
Glasp’s post is strong because it separates raw lift from attributable lift. They report a 37x increase in ChatGPT traffic, but after control comparisons, the more conservative reading is closer to 2x. Their own framing matters more than the vanity number. It signals discipline.
That caution is rare in AEO content. Many posts stop at the largest number they can find, then infer causation from correlation. Glasp does not do that. They compare treated and untreated pages, and they report an interrupted time series estimate, not just before-and-after totals. In other words, they try to answer the question most case studies avoid, what changed because of the intervention, and what was already happening anyway.
Raw lift is easy to publish. Attributable lift is harder, and usually smaller.
If you are an early-stage founder, that distinction should shape how you read every vendor deck, every AEO dashboard, and every growth thread about “AI visibility.”
What they did
Glasp’s core move was not mystical. They looked at server logs, found patterns in AI bot activity, and used that evidence to decide which pages to improve, create, or retire. The method is operational, not rhetorical.
Three parts matter most.
- They read first-party logs rather than relying on third-party estimates.
- They treated bot behavior, including 404s and 403s, as signals.
- They created or revised pages based on observed demand, not a generic keyword list.
Cloudflare matters
One practical detail is worth isolating. Cloudflare now offers AI Crawl Control, which can give you a free view into AI crawler activity. For a small site, the value is not the policy itself, it is the visibility.
If you have a tiny team, you do not need a complex platform to start. You need a daily or weekly snapshot of what is being crawled, what is being blocked, and where the bot traffic is going. A simple cron job that exports or records that window is enough to create a trend line.
Cloudflare documents the product and its bot controls here, AI Crawl Control, and its broader bot management model here, bot traffic concepts.
404s are signals
One of the most useful ideas in the Glasp post is also one of the simplest. If ChatGPT, or any AI assistant, sends users to 404s, that is not noise. It is a content roadmap.
Ahrefs has cited research suggesting ChatGPT sends users to 404 pages roughly three times as often as Google. Whether your own ratio matches that figure matters less than the direction of the signal. AI systems are already surfacing intent that your site has not yet satisfied.
For a small site, the practical question is not, how do we stop 404s. It is, which 404s tell us what to build next?
Titles still matter
Glasp also emphasizes page structure. Titles and summaries function as the citation interface. AI systems need something readable, scannable, and unambiguous.
That does not mean writing for robots. It means making the answer legible to both humans and retrieval systems. Short titles, clear subheads, and direct TLDRs help. So do pages that answer one job, not five.
For founders, this is the old editorial lesson in new packaging, clarity compounds.
SEO Guard
They also describe an internal safeguard they call SEO Guard, a check against trading certain value for speculative value. That is an important idea, especially for teams under pressure to “optimize” everything for AI citation.
If a change improves impressions but weakens the product, obscures the message, or bloats the site, it is not optimization. It is drift.
Cold start translation
Now the part most readers actually need, what survives when you have no authority, no large corpus, and no budget for tooling.
The answer is, more than you think, but less than the headline implies.
Instrument first
Set up three free inputs before you touch content strategy.
- Cloudflare, or your equivalent edge log view, to see AI crawler behavior.
- Bing Webmaster Tools, to review grounding queries and see what questions Bing is associating with your pages.
- Google Search Console, to track the queries and pages already earning impressions.
These tools will not tell you what to publish in perfect order. They will tell you where actual demand already exists.
That is the difference between AEO as guesswork and AEO as evidence.
Build from demand
Glasp’s own numbers show why this matters. Their hit rate on Google impressions moved from 0.3 to 0.8 percent to 13.4 percent when they created pages from observed demand rather than from bulk generation. That is the right lesson for a small site.
Do not start with a list of 200 keywords. Start with questions your current visitors, crawlers, or customers are already revealing.
- Which pages attract impressions but no clicks?
- Which queries appear in Search Console but have no dedicated page?
- Which AI-cited answers map to a 404 or thin page?
Those are your first candidates.
Audit the firewall
Glasp found that a quarter of ChatGPT-User requests were being 403’d at one point. That is the kind of problem a founder can miss for weeks, sometimes months, because the site still “looks fine” from a browser.
If you are early, check access before you chase optimization. Confirm that the routes you want AI systems to see are reachable. Confirm that useful pages are not blocked by accident. Confirm that your robots, WAF, and bot settings are not fighting each other.
This is the cheapest gain available to small teams, because it prevents lost discovery before you start improving relevance.
What not to copy
The hardest part of reading a large-site case study is restraint. Glasp’s environment is not your environment.
Do not copy the parts that depend on scale.
- A 12,000-page rewrite queue is not a strategy for a small site.
- Tombstoning half a corpus is not a useful default if you only have 40 important pages.
- Assuming discovery is already solved is a luxury you do not have.
At DA 0, the bottleneck is usually not citation efficiency. It is discoverability, indexability, and topical focus.
That means the right move is narrower. Create a few strong pages from observed demand, make them crawlable, and measure what happens. Then repeat.
What we still know
We still do not know how much of this generalizes to the cold start case. Glasp says so directly, and they are right to. A large, trusted domain can exploit signal faster than a new one can earn it.
We also do not yet have enough public evidence on which page types are most likely to be cited by AI systems across categories, or how much of the effect is driven by technical accessibility versus content shape versus brand familiarity.
That is why the honest response is not to declare the playbook finished. It is to run a small experiment with better measurement.
A simple test
If you want to learn from this case without overcommitting, run a four-week experiment.
- Export AI bot hits, 404s, and 403s once a week.
- Group the URLs by intent, not by section.
- Pick five pages to improve or create, based on observed demand.
- Track impressions, AI citations if available, and actual referral traffic.
Then compare those pages against a similar set you did not touch. If you cannot do that, at least write down what changed and when. Small sites do not fail because they lack ideas. They fail because they cannot tell which ideas worked.
Why this matters
The value of the Glasp case is not the largest number. It is the method, paired with honesty.
Server logs are better than dashboards because they are closer to reality. 404s and 403s are useful because they expose mismatch. Demand-driven pages outperform bulk content because they start from evidence. Controls matter because AI visibility claims collapse without them.
That is a useful operating system for founders, especially those with little authority and little room for waste.
Takeaway
If you are building with zero domain authority, ignore the spectacle and keep the method.
- Use free instrumentation first.
- Read your own logs.
- Treat 404s and 403s as signals.
- Create pages from observed demand.
- Measure against a control, or assume you are fooling yourself.
Glasp’s post is valuable because it is unusually honest. The 37x headline gets attention, but the 2x control-adjusted claim is the lesson. In AEO, as in everything else, attribution is the work.
For a small site, that means one practical step this week, stop asking which dashboard to buy, and start asking which pages your own data already told you to build.