How we engineered visibility in AI search

Why traditional SEO wasn't enough

We'd built our growth on SEO. We understood how to rank. Position 1 on Google used to mean guaranteed visibility.

But large language models don't work like search engines. They don't have a "top 10." They synthesise answers from hundreds of sources, and only 12% of their citations come from Google's traditional top rankings. Domain authority doesn't guarantee visibility. Being "ranked" doesn't mean being "seen", users get their answer inside the AI interface and never click through.

An Ahref study showed that clicks on Google's #1 position had nearly halved for informational keywords. AI Overviews now appeared in 47% of search queries. The playbook was changing, fast.

Our objective

We needed to answer a critical question: Can you engineer visibility in AI search? And more importantly, could it generate real pipeline?

Not just "Can we get mentioned occasionally?" But: Can we systematically increase our presence in AI-generated answers, track that visibility, and measure its impact on revenue?

As a B2B demand generation agency, we live by a simple principle: if we can't make something work for ourselves, why would you trust us to make it work for you?

We don't sell theories. We don't package up best practices we've read about. We test, validate, and prove strategies on our own business first, then bring those results to our clients.

So, we decided to become our own test case. We needed to prove AEO could generate measurable pipeline before we'd position it as a service.

We wanted to validate whether we could improve our visibility and presence in AI-driven search engines by combining structured technical enhancements—schema, content hierarchy, entity clarity—with high-volume, intent-specific content generation.

Specifically, we wanted to observe increased citations in LLMs (tracked via Scrunch), traffic from AI referral sources like ChatGPT and Perplexity, early signals of user attribution through self-reported discovery, and most importantly: real pipeline.

This wasn't about vanity metrics. We needed to prove AEO could generate measurable, high-intent pipeline. Not just visibility. Revenue.

What we did

We broke our approach into seven key phases, each designed to systematically improve our AI visibility while maintaining the ability to track and measure impact. The goal wasn't just to appear in AI answers, it was to build a repeatable, scalable framework that could consistently drive high-intent leads and pipeline.

"This is a repeatable model that we found here. We produce the content, we get visibility, we see the lagging indicator of high-intent leads to our website. It's systematic and it works."

Phil Vallender, Co-Founder at Blend

The method

1. Built our ICP and persona foundations

Before creating any content, we needed to ensure the AI models we used had the right context about our audience. We compiled our persona documentation and customer insight library, real data on how our ICPs think, search, and evaluate agencies. This ensured that when we generated content through AI, it would be grounded in authentic audience behaviour, not generic language.

2. Created an AI-powered prompt simulation engine

Using those insights, we built a Claude project for persona synthesis, an intelligent prompt engine capable of thinking like our ICPs. This system allowed us to input a topic or industry and generate realistic, human-like search prompts that reflected how our target buyers would query AI engines. It effectively acted as a bridge between traditional keyword research and modern AI prompt simulation.

3. Set up tracking to measure AI visibility

Once our personas and prompt models were in place, we configured our AEO tool, Scrunch, to track the prompts that mattered most. We created a formulaic tracking framework with consistent prompt structures across our key industries. This gave us a scalable, repeatable way to monitor which prompts we appeared for and measure progress as we deployed more content.

4. Reverse-engineered how AI retrieves and cites content

Before producing content, we analyzed how AI engines source and prioritise information. From hundreds of test prompts, we identified clear patterns: AI prefers structured, answer-first content with clear lists, entities, and supporting context.

Using these insights, we created standardised wireframes for AEO content. The content outlines were designed for maximum AI retrievability, built so large language models could easily parse, cite, and reference Blend's content in answers.

5. Built an AI content engine for scale

Next, we developed dedicated Claude projects for content creation at scale. To ensure quality and consistency, we created an enhanced instruction guide, effectively our internal AEO playbook, which included the relevant wireframe structure, brand and tone guidelines, and a curated knowledge base of Blend's case studies, portfolio pieces, and proof points. This ensured the AI not only wrote in our voice but also embedded real data, links, and results, which are critical for citation and credibility in AI retrieval systems.

6. Produced content with embedded technical optimisation

With our framework in place, we began generating and publishing content. Tens of pages across key industries, each built with precise, persona-driven copy, inline case study references, and embedded schema markup. This further strengthened our site's structured data profile and helped AI systems better understand the relationships between Blend's services, industries, and results.

7. Monitored, measured, and iterated continuously

Once live, all content and schema were indexed for Scrunch tracking and cross-checked against AI results. We monitored which prompts began surfacing Blend content or citations, traffic from AI referral sources in HubSpot, canges in self-reported attribution mentioning AI tools, and pipeline generated from those leads. These signals provided concrete data on the direct link between AEO optimisation and measurable AI visibility.