AI visibility
In short: Sikte AI built a staging mode into its own measurement tool to track B2S Research's new website's AI visibility while it was still sitting in a closed test environment. The result: the overall AI visibility score climbed from 66.5 ("moderate") on the old live site to 79.8 ("strong") in staging — before anything went live. Citability nearly quadrupled, from 17 to 63 points.
What Is AI Visibility - And Why Isn't Good Content Enough?
More and more buyers are finding their suppliers through ChatGPT, Perplexity, and Google's AI answers instead of a classic Google search. That means it's no longer enough for a website to be well-written and polished — it has to be something a language model actually understands, trusts, and chooses to cite. That's what we call AI visibility, and it's exactly where a lot of strong B2B content falls short: the writing is good, but it's pitched too broadly and isn't specific enough to get cited by AI.
The Problem: You Can't Measure What Isn't Published Yet
B2S Research AS, a Norwegian research firm with 25 years of experience in employee and customer surveys, was rebuilding its website from scratch. Sikte AI took on the job of restructuring and optimizing the content for AI visibility as the rebuild progressed.
This runs straight into a classic problem: an AI visibility tool expects a live, published site — a proper robots.txt, sitemap, llms.txt, and pages without a "noindex" tag. A site under construction gets all of this "wrong" on purpose, because it isn't meant to be found yet. Run a standard scan against a staging environment and the tool penalizes you for things that will be fixed at launch anyway, leaving you with a misleading picture. In practice, that's why most teams wait until everything is live to measure, by which point it's too late to change course without costly rework.
The Fix: A Staging Mode Built Into Our Own Measurement Tool
We adapted our own AI visibility tool with a dedicated staging mode. It recognizes that a site is a pre-launch build and excludes the launch-dependent signals; robots.txt, sitemap, llms.txt, and noindex, from the score. Those get flagged for a fresh check once the site goes live instead.
What's left in the calculation is exactly what content work actually determines right now:
how clearly the company reads as a distinct entity
how well-structured the underlying data is
how readable the structure is
and, above all, how citable the content is
That let us track improvements in real time while the work was underway, directly in the test environment, and make adjustments before anything was published.
The Results: +13.3 Points and a New Level of Citability
The baseline, the live b2s.no site, measured 05.08.2026, scored an overall AI visibility rating of 66.5 out of 100 ("Moderate"). The rebuilt version, measured in staging on 07.08.2026, already scored 79.8 ("Strong"), and the work is still ongoing.
Dimension | Before (live) | Now (staging) |
|---|---|---|
Overall score | 66.5 – Moderate | 79.8 – Strong |
Citability | 17 | 63 |
Entity clarity | 55 | 100 |
Crawlability | 80 | 95.2 |
The biggest jump is in citability, from 17 to 63, nearly a fourfold increase. Where the old site touted "25 years of experience" without a single source to back it up, the new version gives language models something concrete to anchor to and reproduce.
Entity clarity tells a similarly striking story, jumping from 55 to a perfect 100. The old site had no registration number, no external verification sources, and no named experts. The new one carries a full company registration number, address, phone, email, and sameAs links to the Norwegian business register, Proff, and LinkedIn, plus three named team members with their roles and client testimonials attached. For a Norwegian B2B company competing in a thin data landscape, that's precisely the difference between being confused with a competitor and being the one that gets chosen.
What We Actually Changed
Entity data: registration number, address, phone, email, and sameAs links to the Norwegian business register, Proff, and LinkedIn, added as structured data.
Named experts: three team members were given names, roles, and client testimonials, replacing anonymous claims.
Sourced claims: statements like "25 years of experience" were backed with concrete sources and figures instead of standing unverified.
Structure: content reorganized so each section answers one specific question a buyer would actually ask an AI.
The Takeaway: Measure While You Build, Not After
The headline here isn't really the number — it's the timing. Because we could measure inside the test environment, AI visibility became part of the rebuild itself, not a post-launch audit. Every structural change and every piece of content was validated before launch, and there's still more to extract before the site goes live.
Get a Free AI Visibility Analysis
Rebuilding a website, or wondering how visible you really are when customers ask an AI instead of googling? Contact us here to get a free lite AI visibility analysis from Sikte AI.
The figures in this article are from Sikte AI's AI visibility analyses of b2s.no (live, measured 05.08.2026) and the rebuilt staging version (measured 07.08.2026).
Frequently Asked Questions
What is an AI visibility score?
An AI visibility score measures how well a language model can understand, trust, and cite the content on a website, based on factors like entity clarity, citability, and crawlability.
Can you measure AI visibility before a website has launched?
Yes. With a staging mode that excludes launch-dependent signals (robots.txt, sitemap, llms.txt, noindex) from the score, you can measure content quality and structure in a closed test environment and adjust course before launch.
What's the difference between SEO and AI visibility (GEO)?
SEO optimizes a website to rank highly in classic search results. AI visibility — also called GEO, or Generative Engine Optimization — optimizes a website to be understood, trusted, and cited by language models like ChatGPT, Perplexity, and Google's AI answers.
Why does citability matter more than writing quality alone?
Because a language model needs something concrete to anchor to and reproduce — figures, sources, named people, and verifiable facts — not just well-written, general claims.
Related articles:
Torben Brenden Pedersen
Founder / GEO Specialist
Torben works with GEO and technical optimization, turning expertise into machine-readable authority across search engines and answer engines.