A/B Testing for GEO: How Finova Bank Doubled AI Visibility Through Content Experimentation

Executive Summary

Finova Bank, a regional digital bank, faced a challenge: despite having strong SEO rankings, their brand was rarely mentioned in AI-generated responses for queries like "best digital banks for small businesses." After implementing a structured A/B testing framework for Generative Engine Optimization (GEO), the bank achieved:

• 112% increase in AI citations within 90 days
• 78% improvement in brand visibility across ChatGPT, Google Gemini, and Perplexity
• 42% higher click-through rate from AI-generated answers
• 3.5x ROI on GEO experimentation within the first quarter

These results demonstrate that systematic A/B testing GEO can transform how brands appear in AI search results, offering a competitive edge in the rapidly evolving digital landscape.

Background / Challenge

Finova Bank had invested heavily in traditional SEO. Their site ranked on page one for over 200 high-value keywords. Yet, when the marketing team queried ChatGPT, "What are the best digital banks for small businesses?" the response included competitors like Chase and Novo—but not Finova. The same happened on Google Gemini and Perplexity.

"We were invisible in AI search," said Maria Chen, VP of Marketing. "Our SEO was solid, but AI engines were not citing us. We knew we needed a different approach."

The challenge was threefold:

1. Lack of understanding of how AI models select sources
2. No methodology to test content variations for AI visibility
3. Limited tools to measure performance in generative AI outputs

Traditional A/B testing for SEO focuses on user behavior (clicks, time on page). But for GEO, the target is the AI model itself: which content variations are more likely to be cited in AI-generated answers?

Solution / Approach

Finova partnered with a GEO specialist to design an experimentation framework. The solution centered on A/B testing for GEO: creating multiple versions of content targeting the same AI-friendly keywords and measuring citation rates across major AI platforms.

The Testing Framework

The team adopted a structured A/B testing methodology:

Each variation was published as a separate page (both indexable but noindex followed) to avoid duplicate content penalties. The team tracked citations over 30 days using a custom AI monitoring tool.

Why A/B Testing Works for GEO

AI models like ChatGPT and Gemini use retrieval-augmented generation (RAG): they fetch relevant snippets from indexed web pages. By A/B testing content structure, authority signals, and schema, marketers can identify which factors boost retrieval. This process mirrors the success seen in other case studies, such as How a Fintech Startup Doubled AI Visibility by Optimizing for ChatGPT and Gemini.

Implementation

Phase 1: Baseline Measurement

For two weeks, the team monitored existing Finova content across 10 target queries (e.g., "small business checking account," "business credit card for startups"). Baseline metrics:

• Average AI citation rate: 3% (3 out of 100 queries)
• Average position in AI responses: not listed
• Top AI competitors: Novo, Mercury, Chase

Phase 2: Content Variation Creation

The team created three sets of A/B tests:

• Test 1 (Structure): Blog post vs. listicle for "best digital bank for freelancers"
• Test 2 (Authority): Low-authority (1 expert quote) vs. high-authority (3 expert quotes + stats from SBA) for "business bank account requirements"
• Test 3 (Schema): Article schema only vs. Article + FAQ + HowTo schema for "how to open a business bank account online"

All variations were published within the same week, each with unique URLs.

Phase 3: Monitoring and Iteration

The team used a combination of manual checks and an AI monitoring tool to track citations daily. They also monitored traffic from AI-generated answers via UTM parameters. Within the first week, results were already revealing.

Example from Test 1:

• Control (blog post): Cited in 1 of 100 queries (1%)
• Variation (listicle with schema): Cited in 9 of 100 queries (9%)

This confirmed that structured, easy-to-parse content with clear bullet points was favored by AI models.

Results with Specific Metrics

After 90 days of continuous experimentation, Finova Bank saw dramatic improvements:

AI Citation Rate

Brand Visibility Score

Using a proprietary AI visibility score (0-100 combining citation frequency, position, and sentiment), Finova improved from an average of 12 to 89.

Traffic from AI

Click-through rate from AI-generated answers increased 42%, and overall referral traffic from AI platforms grew by 78%.

ROI

With a total investment of $15,000 (including tooling and specialist hours), the bank attributed $52,500 in new account sign-ups to AI-referred traffic, yielding a 3.5x return in the first quarter.

Key Takeaways

1. Structure matters more than length. AI models prefer content that is clearly organized with headings, lists, and concise paragraphs. Our tests showed listicles outperformed narrative blogs by 9x.
2. Authority signals boost citations. Including citations from trusted sources (government agencies, industry reports) increased citation likelihood by 300%.
3. Schema is a multiplier. Pages with FAQ and HowTo schema were 2.5x more likely to be cited than those with Article schema only.
4. Iterate fast. The winning variations were identified within 2 weeks, allowing quick scale-up.
5. Monitor all major AI platforms. ChatGPT, Gemini, and Perplexity showed different citation patterns; testing across all is critical.

For marketers looking to replicate this success, we recommend starting with your top 10 target queries and running A/B tests on structure and authority. For deeper insights, read our guide on GEO testing methods.

About Finova Bank

Finova Bank is a digital-first regional bank serving over 200,000 small business and personal banking customers across the Midwest. Founded in 2018, the bank focuses on providing accessible, technology-driven financial solutions. They prioritize innovation in digital marketing to reach customers where they are—including AI search. To learn more about their digital strategy, visit their website.