1
Understand your customers
Voice memos, recorded sales calls and a 125-question customer register became a governed Brand Kit with 53 product records, seven audiences and a binding compliance record.
Define your OfferingCase Study: Lyceum Technology
Winning non-branded searches
77×more often found for non-branded search queries
8 months ago Lyceum was only the answer when someone looked directly for them. Today they are found for the transactional questions their customers ask when they don’t know the provider yet.
- Non-branded searches8% → 74%Searches that don’t contain the company’s name
- Mentioned first30%Searches where the company is named first
- Reach77×More often found for search queries that don’t contain the company’s name
Building the answers, then the machinery to keep them right
Lyceum Technology sells GPU compute and inference from European data centres. Before this engagement its buyers found the company almost exclusively by name: 92% of its search impressions came from queries containing the word Lyceum.
Over the following months we built a technical magazine covering GPU memory maths, model hosting, inference economics and EU data residency, grounded in a fact base with per-claim source retrieval and bound by a compliance record that defines exactly what Lyceum may and may not claim.
Today roughly three quarters of the domain’s search visibility comes from non-brand questions, and on the buying questions that decide a deal an assistant that names Lyceum names it first three times in ten, ahead of providers many times its size.
“Our buyers arrive with a technical question long before they know which providers exist. We are now the answer to those questions, and every claim behind it is one we can stand behind.”
Magnus Grünewald
Co-Founder & CEO, Lyceum Technology
Findable by name. Invisible to the question.
Lyceum sells to enterprises that need GDPR-compliant, EU-sovereign AI infrastructure, to AI implementation consultancies specifying a stack for a client, and to AI-native startups that can cut inference costs by up to 97% by moving from commercial APIs to open-source models on Lyceum.
None of them starts by searching for a provider. They start with the problem, and Lyceum’s answers lived in a docs site and in its product team’s heads, where no engine could retrieve them.
Lyceum also has to be careful about what it claims and what it stands for. In this category a loose sentence about data residency is not a marketing risk, it is a compliance one.
- 92% of labelled search impressions came from queries containing the word Lyceum
- 148 distinct non-brand questions earned impressions, against 31 non-brand clicks
- No AI-visibility measurement in place
Traceability built in, not checked at the end
2
Find the questions customers ask
Demand harvested from Search Console, SERP pulls, People-Also-Ask trees and developer communities, reduced to 20 clusters and a 62-prompt panel in continuous measurement.
Find Customer Intent3
Publish content that sells
A technical corpus grounded in a 633-fact base with per-claim source retrieval, so every statement is traceable to where it came from.
Publish Content4
Update to stay relevant
The compliance record is read at generation time, and corrections run through a patch pipeline against it rather than by hand, so the corpus stays compliance-ready as it grows.
Refresh Content5
Report and improve
Weekly reporting on data the client owns, after nine months of analytics were recovered from behind two instrumentation faults.
Report and ImproveThree results, each against its own baseline
Search figures compare two equal 75-day windows. The AI figures are a 28-day window of the answers themselves, counted answer by answer.
First in three of every ten AI answers
Being mentioned is not the same as being the answer. Counted answer by answer, an assistant that names Lyceum names it first 30% of the time, and puts it in the top three in more than half of those answers.
Those wins sit on the questions closest to a decision: EU hosting, data residency, cost per token, running a fine-tuned model in production.
On these buying questions, Lyceum is the first provider named
Share of the answers naming Lyceum in which it is named first, 28 days to 17 August 2026
| Tracked buying question | Named first, as a share of the answers naming Lyceum |
|---|---|
| Private LLM hosting, scale-to-zero | 75% |
| Kimi K3 through a hosted API in Europe | 75% |
| EU-hosted API, open-source image models | 67% |
| Chinese open-weight models, EU data centres | 60% |
| Pay-per-token inference, EU data residency | 57% |
| Fine-tuned model as a production API in the EU | 50% |
How visible is your brand in AI search? Find out with a free check.
Non-branded searches went from 8% to 74%
This is the number that changed what the domain is. Eight months ago eight percent of Lyceum’s labelled search impressions came from questions that did not contain the company name. Today it is roughly three quarters.
The brand did not suffer for it: clicks on Lyceum’s own name nearly tripled over the same period.
Non-brand questions went from a rounding error to three quarters of visibility
Share of labelled Google Search impressions from queries that do not contain the company name
9%77%
- Dec
- Jan
- Feb
- Mar
- Apr
- May
- Jun
- Jul
- Aug
| Month | Non-brand share of impressions |
|---|---|
| Dec | 9% |
| Jan | 8% |
| Feb | 26% |
| Mar | 71% |
| Apr | 72% |
| May | 79% |
| Jun | 75% |
| Jul | 70% |
| Aug | 77% |
77× more often found for non-branded search queries
Eight months ago the domain earned impressions against 148 distinct questions that did not contain the company name. In the equivalent window today it earns them against 11,401.
That is the difference between a domain people find when they already know the name, and one they find while working out what to buy.
The domain is found for 77 times more buying questions
Distinct non-brand questions earning search impressions, two equal 75-day windows
| Window | Distinct non-brand questions earning impressions |
|---|---|
| Now, 2 Jun to 15 Aug 2026 | 11,401 |
| Baseline, 9 Dec 2025 to 21 Feb 2026 | 148 |
Figures re-derived 17 August 2026 from Google Search Console and the scaile platform’s AI-visibility record. Continuous AI measurement began 9 July 2026, so no pre-engagement AI baseline exists and no AI-visibility change is claimed.
SimonCo-Founder and CEO
Want to be the answer in your category?
Frequently asked questions
Need more detail? Book a Call
What exactly was measured here?
Two things, separately. The Google figures come from Search Console for lyceum.technology and compare two equal 75-day windows: 9 December 2025 to 21 February 2026 against 2 June to 15 August 2026. The AI figures come from the scaile platform’s answer record: 67 tracked buying questions across 8 engines, over the 28 days to 17 August 2026.
Why is there no before-and-after for AI visibility?
Because continuous AI measurement only began on 9 July 2026. There is no pre-engagement AI baseline to compare against, so nothing on this page claims an AI-visibility change. What it claims is the position Lyceum holds today, counted answer by answer.
What does “named first” mean?
In every answer that mentions Lyceum at all, the order the providers appear in is recorded. “Named first” is the share of those answers in which Lyceum is the first provider named. It is not a share of all answers, and it is not the same thing as being mentioned.
Is a rising non-brand share not just a side effect of publishing more?
Publishing is what moved it, but the point is what it moved to. Clicks on the brand name nearly tripled over the same period, so the non-brand share did not rise because the brand fell. The domain simply started earning impressions on questions it was previously absent from.
How long did this take?
The content retainer started in December 2025 and the first articles indexed on 25 February 2026. Non-brand share crossed 70% in March, the month after those articles landed, and has stayed there since.