AI Visibility Case Hub: 7 Measured Case Studies With Real Numbers

We tested the playbook on the hardest target first — ourselves. No incumbent advantage, no friendly journalist, no warm client list. Just the loop, on our own brand, for three months.

Baseline: 2 citations across all tracked engines. Result at 3 months: 1,000+ AI citations. 819 measured AI mentions. 15.4% share-of-model in our category. Workspace data: 391 tracked queries, 984 brand citations, 20% coverage.

The point of this case is not that we are good at this. The point is that the loop works when the starting position is zero and the only input is structured, measurable work.

Full case: Humanswith.AI case study

Birdview had strong AI visibility inside the narrow PSA (Professional Services Automation) cluster — Perplexity 40%, Claude 41%. But the broader PMS (Project Management Software) cluster is where the larger audience lives. There, ChatGPT cited them in only 0.9% of answers.

What we did: Mapped 100 LLM queries in the PMS space, weighted Decision-stage prompts 2×, and shipped 8 articles in 8 weeks (one on the client's site, one on Medium per topic). Every article passed an LLM-citability check before publish.

Result: ChatGPT mention rate 0.9% → 21.5% (23× lift). Unique queries with Birdview present: 8 → 103. Gemini: 27.8%. 95 of 100 LLM queries overlap with Birdview's SEO keywords — one asset drives both channels.

Workspace data (latest scan, 13 Aug 2026): 589 prompts tracked, 413 citations, 5 engines, 14% citation share across the full panel.

Full case: Birdview PSA case study

GAC was AI-blind in St. Petersburg auto retail — one mention across nine AI engines, while competitors got recommended daily.

What we did: Picked nine formats AI systems extract most readily — calculations, competitor comparisons, "should I buy?" guides. Structured every article for chunk extraction. Published one topic per week for six weeks. Measured weekly across nine engines.

Result: 9 articles published, 6 of them cited by AI systems. 9,042 total reads, 7,250 from the top 4 articles alone. Now appearing as the default recommendation for the city's auto-buying queries across all 9 platforms.

Full case: GAC case study

Whitewill is a strong Dubai-real-estate brand offline. For 121 high-intent investment queries on AI engines, the brand had zero citations at baseline. Meanwhile the mentions sat elsewhere: Medium ~97, Engelvoelkers ~92, LinkedIn 44, Property Finder 21.

Our read: The answer layer is already occupied by trusted third-party surfaces, not brand sites. The fix is publication-surface strategy, not more site content.

Plan: 12-week cadence on Medium, LinkedIn, and 2 trusted property-investment publications. Owned-site work scoped to schema and chunk-ready FAQ blocks for the top 30 commercial queries.

Honest framing: This case is published as a baseline audit and execution plan, not as a finished growth outcome. The diagnostic is the most useful artifact.

Full case: Whitewill Dubai case study

Nonton had strong brand-search results — 127 mentions when AI engines were asked about the brand by name. But across 160 non-branded category queries, the brand appeared in exactly one answer.

Our read: Brand recognition was not translating into category-level answer visibility. The fix was a 200-query map split into branded and non-branded demand, two-circuit publishing, and chunk-ready owned-site pages built for the category prompts that actually decide a purchase.

Result: 200 queries mapped (40 branded + 160 non-branded). Two-circuit publishing live.

Full case: Nonton case study

Gorbilet's tourism niche behaves differently from B2B SaaS or real estate. The total AI citation surface for the river-tours topic produced 8,861 mentions. Gorbilet already held 289 of them, with 100 branded mentions and a 35% branded share.

Our read: 94.6% of citations in this niche came from operator websites themselves, not from blogs, media, or video platforms. The strategy shifted toward deep operator-site content as the primary channel.

Full case: Gorbilet case study

LS ELECTRIC's regional site had 66 AI citations on the topic, so the entity was recognised. But 64 of the 66 landed on the homepage; only two reached the support section. The global LS ELECTRIC site held 170 citations on the same topic.

Our read: AI systems knew the entity existed, but the regional site lacked the product, distributor, comparison, and catalogue pages needed to answer real buying questions. The fix is commercial-intent content depth, not more brand awareness.

Full case: LS ELECTRIC case study

CodHob is a B2B fintech white-label credit platform for the Kenyan market. The baseline was 35 citations across five AI engines in April 2026. After 7 weeks of structured publishing and measurement, the count reached 259 — a 7.4× lift. CodHob became the second most-cited brand in its niche, behind LinkedIn.

Workspace data (snapshot 88, 10 Aug 2026): 158 total queries, 259 brand citations, 30% coverage, 1% mention share.

Full case (RU): CodHob case study

Reading eight cases back to back, the pattern is boring on purpose. None of them hinge on a clever trick. All of them hinge on doing four unglamorous things in order, every week.

The engine does not care about effort. It cares about signals. The Princeton/Georgia Tech GEO study showed that citations, quotations, and statistics move generative-engine visibility by up to 40% — structure beats volume ^[1]. Every case above is an application of that finding to a specific query panel, nothing more.

The second pattern is where the lift comes from. Whitewill, Nonton, and Gorbilet all showed the same shape. The brand was known, but the answer surface was occupied by someone else — a marketplace, an operator site, a third-party publication. Awareness does not convert to citation. Structure and placement do. That is why a baseline audit is sometimes the whole answer: you cannot fix a placement problem with more owned-site blog posts.

"A case study without a baseline is a testimonial. A baseline with a weekly re-measure is evidence. We publish the second kind." — Gregory Shevchenko, founder, Humanswith.AI

1. A fixed query panel. Every case starts by defining the exact queries buyers ask, then freezing that panel for weekly measurement. No panel, no measurement.
2. A baseline. Every case has a "before" number. Without it, you cannot prove anything moved.
3. Weekly re-measurement. Single screenshots are noise. Trends on the same locked panel are signal.
4. Honest framing. Some cases are finished growth stories (Humanswith.AI, Birdview, GAC). Some are baseline audits with a plan (Whitewill, Nonton, LS ELECTRIC). Both are published because both are evidence of measurement.

[1] Aggarwal et al. — "GEO: Generative Engine Optimization" (Princeton / Georgia Tech / IIT Delhi, 2023, arXiv:2311.09735) — citations, quotations, and statistics improve generative-engine visibility by up to 40% — https://arxiv.org/abs/2311.09735