Foundation & roadmap
- Buying questions and search intent
- Competitors and AI mentions
- Keyword and intent mapping
- Technical SEO and crawl audit
- Prioritised SEO and AI search roadmap
By month end
A clear roadmap with firm priorities.
How we work: the BELEG method.
Demand, engines, gaps, evidence, counter-check. Five steps in a fixed order that we run for every engagement. AI does not name the best provider, it names the one it can evidence.
Our clients
Niccos — Shopify & e-commerce
Touchpoint Amplify — Google Ads agency
NEONA — Designer lighting & e-commerce
Ad Specialist — Performance marketing agency
Insel Gruppe — Bern University Hospital
Bluerails — Agentic booking for hotels
voids.ai — Supply-chain SaaS
SPIELMANN Steuerberatung GmbH — Tax consulting
PUNKT PR — PR agency
myAutoAbo — Car subscription comparisonWhy the order decides.
Most visibility projects do not fail during delivery. They fail on the order. Pages get written before anyone checks whether the question is asked. Things get optimised before anyone measures who is named today. And in the end the comparison is missing that would show whether the work had an effect.
BELEG turns that around. First it is settled which question stays unanswered and which source answers it today. Only then does it become clear whether the gap is closed by a page, an article, a product data set or an external source.
That substance beats style is well researched. The Princeton study on Generative Engine Optimization (KDD 2024) measures a 115% higher citation probability for source citations, +41% for expert quotes with attribution and +40% for concrete statistics. Pure language polishing yields roughly +3%.
The fifth step is the one almost everybody leaves out. Without a comparison group every change looks like success. So we measure treated pages against untouched pages from the same period.
- +115%
- citation probability from source citations
- Source: Princeton GEO Study, KDD 2024
- +41%
- from expert quotes with attribution
- Source: Princeton GEO Study, KDD 2024
- +3%
- from pure language polishing
- Source: Princeton GEO Study, KDD 2024
- 76% → 38%
- overlap of AIO citations with the Google top 10
- Source: Ahrefs AIO Study, Jan 2026
What happens in which step.
Demand: What does your market actually ask?
We start with the questions where AI systems recommend providers, products or service companies, not with a keyword list. Search volume comes exclusively from the Google Ads API — never from tool estimates and never from a number a language model produced. If the check shows nobody searches for a service, we say so, even when that makes the engagement smaller.
Done when: You know which questions your market asks — and which service is not worth a page.
- Collect buying questions instead of keywords
- Check search volume through the Google Ads API only
- Match offer against demand before any page is written
Engines: Who gets named today, and where does the answer come from?
ChatGPT, Perplexity, Claude and Google AI are measured separately, because they select differently and cite different sources. Perplexity searches live on every query, Google AI Mode splits one question into eight to twelve sub-questions, Claude favours specialist publications. Looking at one system alone turns noise into a trend.
Done when: You see per system who gets recommended and which sources carry the answer.
- Measure every AI system on its own, not as an average
- Record named providers and cited sources per question
- Track competitor mentions as a benchmark
Gaps: Where exactly are you missing, and what weighs most?
Demand and measurement produce a gap list, not an overall score. Every gap gets a type: is a page missing, is evidence missing on an existing page, or is an external source missing? The order comes from your data — revenue, margin and effort decide, not a standard package.
Done when: It is settled what gets built first and why — traceable, not by gut feel.
- Separate gaps by type: page, evidence or source
- Prioritise by revenue and margin impact
- Find and resolve cannibalisation between your own pages
Evidence: What actually gets built?
Now the missing pieces get built: service and product pages, articles with evidenced claims, comparisons, product data, schema, press placements and entries on the sources that shape your category. What gets cited is well researched — the Princeton study (KDD 2024) measures +115% citation probability for source citations, +41% for expert quotes with attribution and +40% for concrete statistics. Pure language polishing yields roughly +3%.
Done when: Every claim on your pages is evidenced — and every gap has a visible answer.
- Build evidence instead of polishing copy: figures, sources, quotes, cases
- Answer first, question headings, verifiable claims
- Build external sources, not only your own website
Counter-check: Did the work move anything — or was it the market?
We measure the same questions again and compare treated pages against untouched ones. Without that comparison an improvement cannot be separated from market movement. Presence and citation are reported separately: being named and being cited as a source are two different outcomes with two different levers.
Done when: You know which measure worked — and which did not, which matters just as much.
- Re-measure per system, monthly instead of once
- Compare against untouched pages as a control
- Report presence and citation separately
On which layers evidence is built.
AI needs clear information on your website and trustworthy sources outside it. If one of the three layers is missing, the recommendation does not hold.
Website
Service pages, product pages, internal links, schema and the evidence behind a decision.
Owned content
Articles, comparisons, guides, FAQs and cases for concrete buying questions.
External sources
Listings, partner sites, trade media, reviews and the sources AI already uses.
We build visibility as a system.
SEO is the core. AI search, content, technology, authority and conversion work around it. This is what the first 90 days with us look like.
Content & authority
- Content and entity audit
- E-E-A-T review
- Money pages and content briefs
- AI search content recommendations
- Source, PR and link acquisition
By month end
The first new evidence is live.
Delivery & scale
- Information architecture and search journey
- On-page and entity optimisation
- Content production
- Technical implementation
- New measurement and second report
By month end
A measurable before-and-after baseline.
Four rules we do not break.
We check Google, AI answers and sources together. That produces clear tasks which are measured again afterwards.
Buying questions instead of a general topic list
Every AI system is checked on its own
Every gap gets a clear priority
We implement. Your team does not have to operate a dashboard
What the method produced for clients.
Figure, period and the work behind it. Each project ran over several months.
Niccos - Shopify & e-commerce+761 %
more impressions in 3 months
How a Shopify relaunch led to more organic demand.
NEONA - Designer lighting & e-commerce+315 %
more conversions from AI assistants in 3 months
How a lighting store used Sphen to reach more buyers through ChatGPT, Gemini and Perplexity.
AGS IT-Service - Healthcare IT+115 %
more impressions in 4 months
How a clear medical IT site structure doubled visibility in a regulated niche.
We did not invent BELEG to have an acronym. It is the order we work in anyway — and the last step is the one almost everybody leaves out. Without a comparison against untouched pages you cannot tell whether your work moved something or whether the market moved. That is the only question anyone pays us for.
Felix Ament, Founder, getSichtbar GmbH
What our clients say.
Quotes from founders and managing directors we work with. Where a case exists, the number sits next to it.
“No fuckin way, Felix Ament is the GOAT. Within weeks, NICCOS was #1 on ChatGPT and Claude. Our organic inbound leads have exploded thanks to getSichtbar.”
+761 %
more impressions in 3 months
“In the first month, the first request came in with the line: 'ChatGPT recommended you to me.' That is exactly why we use getSichtbar.”
+980 %
more organic conversions in 3 months
How we measure AI visibility.
The AI visibility score in our cases and reports is a measurement, not an impression. Across all clients we put thousands of buyer questions to AI systems every month, spread over the whole month instead of a one-off sample.
Buyer questions, not brand questions
For every client we collect the questions their buyers ask: recommendations, comparisons, cost, trust and concrete problems. Brand questions, meaning questions asked directly about you, do not count towards the score. Anyone asking for you already knows you.
Every AI system on its own
ChatGPT, Claude, Perplexity and Gemini, usually with live web search. We also query Gemini in a second variant with Google Search, as an approximation of Google's AI answers. Each system is evaluated separately because each picks different sources.
Continuous, not one-off
Measurement runs several times a week. The score combines the latest runs so that a single outlier does not bend the curve. Runs that are not reliable are never shown.
Mentioned and cited, separately
Mentioned means the brand appears in the answer. Cited means the company's own website appears among the answer's sources. We track both separately because they are two different levers.
What goes into the score
How prominently the brand is mentioned, in which position, how often the company's website appears as a source, the share of all mentions compared with competitors, and sentiment. This gives a value from 0 to 100 per system, weighted into an overall score.
What the score is not
AI answers vary from one request to the next. The score shows a trend over weeks, not a guarantee for any single answer. In our cases we show the change between the first and the last month of the period, in points.
Common questions about the method.
How do you measure AI visibility?
Across all clients we put thousands of buyer questions to ChatGPT, Claude, Perplexity and Gemini every month, spread continuously over the whole month. Brand questions, meaning questions asked directly about you, do not count. We track separately whether a brand is mentioned and whether its website is cited as a source, and combine that per system into a score from 0 to 100.
What does BELEG stand for?
For Bedarf, Engines, Lücken, Evidenz and Gegenprobe — demand, engines, gaps, evidence and counter-check, the five steps we work in. The German word also names the principle: AI systems do not name the best provider, they name the one they can evidence.
Is this not just SEO with new words?
The technical basis is the same, the selection is not. According to Ahrefs (January 2026) only 38% of the pages cited in AI Overviews still appeared in Google's top 10 — a year earlier it was 76%. We treat both channels separately and explain rankings with ranking factors and citations with citation factors.
Why the counter-check?
Because without a comparison group every change looks explainable. We compare treated pages against untouched pages from the same period. That is not a laboratory experiment, but it separates effect from market movement far better than a plain before-and-after view.
How long does one pass take?
Demand and engines take two to three weeks. The evidence work then runs continuously, gap by gap. We usually see first measurable movement after four to eight weeks; Perplexity reacts fastest because it searches live on every query.
Can you guarantee mentions?
No, and nobody can do that credibly. AI answers vary by query and model version — according to Ahrefs the move to Gemini 3 replaced 42% of the cited domains. We show where you are missing today, build the evidence and measure again afterwards.
What if there is no demand?
Then we say so and do not build the page. Some offers are not searched for but recommended or actively sold — in that case press work, AI answers or sales are the better channels than a service page. That honesty saves the most expensive SEO of all: SEO for pages nobody searches.
Does this require a website relaunch?
Rarely. Most of the leverage sits on existing URLs: structure, evidence, schema, internal linking. A relaunch without a prior demand check is the most expensive way to carry the same problem into a new design.
What does BELEG not solve?
An unclear offer, missing internal decisions and everything that happens outside your domain. Mentions on third-party sites can be prepared but not forced — editors decide themselves. And visibility does not replace sales.
Talk to us about your visibility.
15 minutes, straight from the calendar. Before the call we check which providers AI names in your category and bring the result.
Felix Ament
You talk to me directly.
What happens in the call
- We show which providers ChatGPT, Gemini, Perplexity and Google name for buying questions in your category.
- We show which sources support those answers and where competitors stand.
- We name the evidence your brand is missing today.
- You get the next step with the biggest effect. No obligation.