Success Case: BFE – A Full AI Layer Beside a Relaunched Website, and Google Didn't Blink

The setup nobody recommends, on purpose

A website replatform is when SEO damage shows fastest. URLs change, redirects strain, rankings wobble; every SEO playbook says: change nothing else while Google resettles. BFE Institut für Energie und Umwelt, one of Germany's leading energy-efficiency and sustainability consultancies, 16,000+ projects for around 800 active customers including STIHL, Josera and Diehl Defence, and like econ solutions a subsidiary of MVV Energie AG, went the other way. Their relaunch and their FAIND Knowledge Graph went live in the same month.

That makes BFE the strictest stress test we've published. Every theoretical objection to an AI layer, duplicate content, cannibalization, "Google will flag it", shows up first and worst in exactly this scenario. Four months later, we can audit each one against data instead of theory.

What happened in AI answers

Across the 180 monitored prompts of BFE's field (20 questions × 3 variants × 3 providers, identical every run), BFE went from barely existing to a fixture:

• Mention rate: 0.6% → 19.4%, a 34.7× lift. In April, BFE was outside the top 10 in share of voice for its field. By June: #2 of 10 tracked players, at 12.0% share of voice (from 1.1%).
• Recommendation rate: 0% → 43.6%. Before, AI systems never actively suggested BFE. Now they do in more than four of ten relevant answers, with zero negative-tone responses observed, and 46% of all brand answers actively recommending.
• Claude is BFE's strongest channel by far at 74% of AI recommendations, ahead of Google AI Overviews (23%) and ChatGPT (3%), a provider mix worth an article of its own, because it repeats across our fleet.

Where AI reads, and where the credit goes

Two numbers explain how the layer actually works, and they point in opposite directions on purpose.

Where AI reads: 97.8%. Of 16,502 AI-crawler requests to BFE's web presence between February and July, 97.8% went to the FAIND Knowledge Graph layer, not the human website. When a machine wants to know something about BFE, it overwhelmingly reads the machine-optimized version.

Where AI citations point: 100% to bfe-institut.com. We checked 2,989 AI citations of BFE in June alone. Citations pointing to the FAIND layer: zero. Every single one pointed to BFE's own domain. The layer does the grounding work invisibly; the original website collects every reference, every click, every ounce of credit.

That inversion, read almost exclusively, cited exactly never, is not a side effect. It's the design. (We see the same zero-citation pattern across every deployment in our fleet; the research note on that goes deeper.)

What happened in Google

The replatform settled cleanly. Google impressions and clicks to bfe-institut.com grew month over month through the entire observation window, with the impressions trend up around 18%, a new search footprint built with the FAIND layer live from week one. Brand searches for BFE sit at position 1 with roughly a 50% click-through rate. Indexing stayed clean throughout. In four months, the layer did not outrank a single BFE page: where BFE's site has a page, BFE's page wins by design; the layer only ranks where BFE had nothing. Zero conflicts observed.

The five SEO worries, audited

Every enterprise conversation about an AI layer surfaces the same five objections. BFE's four months let us answer each with data rather than assurances:

"Duplicate content will hurt us." The layer lives on its own subdomain, and every overlapping page carries a canonical tag pointing to bfe-institut.com. The main site remains the single source of truth in Google's eyes. Result: clean indexing, growth every month.

"It'll outrank our own pages." Structurally impossible: where the site has a page, the canonical hands the ranking to it. The layer only competes where the site was absent. Four months, zero conflicts.

"Isn't this cloaking?" No. The layer is public and byte-identical for every visitor and crawler. Nothing varies by user-agent, which is the definition cloaking violates.

"Google flags AI-built pages." Google's policies target pages built to manipulate rankings. The layer doesn't compete for rankings, see above. Four months of clean indexing while impressions grew is the empirical answer.

"AI will cite the copy, not us." 2,989 citations checked in June: zero to the layer, 100% to bfe-institut.com.

One more property that mattered in BFE's evaluation: the whole construction is reversible. The layer runs on a CNAME subdomain, switching it off leaves the site untouched, and the on-site script sets no cookies and stores nothing on the device. The full prompt set and raw AI responses behind every number above are available on request.

In BFE's words

"What convinced me is where the AI answers send people: to bfe-institut.com. FAIND makes us visible in AI systems without taking anything away from our own website."

"Visible in AI systems, without taking anything away from our own website. That's what mattered to us going in, and it's what the data shows."

, Monja Klefenz, Marketing Executive, BFE Institut für Energie und Umwelt

What this case settles

The econ solutions case showed AI visibility and Google performance rising together under normal conditions. BFE shows it under the worst conditions available: a full replatform with the layer live from day one. If duplicate-content damage were going to appear anywhere, it was here. It didn't.

BFE went from barely mentioned to the #2 name in AI answers for its field, during a replatform, with no rebuild required, live in days. If your team is holding the same five objections, the free AI Visibility Audit will show you your own before-picture, and this case is the after.

Sources: FAIND AI-Visibility Engine, observed Feb-Jun 2026; Google Cloud Console (Search data); BFE Institut. Full case study PDF available on request via hello@getfaind.com.