They thought they’d been clear. AI was telling a different story.

The brief

A global brand was facing heavy press coverage of a sensitive issue. It thought it had been clear about its position on its site, but AI tools were giving people a different account. The team wanted to know what AI was saying and why, what it meant for the brand, and how the website needed to change.

The approach

1. Find out what AI is actually saying. Test key prompts across the major LLMs and capture how the issue was being told.
2. Build a shared understanding. Help the team learn how generative search works, how it differs from traditional search, and why this was happening.
3. Identify the gap. The brand’s content lacked the depth and structure AI systems need to treat a source as authoritative.
4. Close the gap. Design a content architecture that makes the brand an authority on its own topics without giving up its voice.
5. Fix the source of truth. Optimize the team’s key page, section by section, so it actually closes the gap.

What we found

The issues fell into four groups:

• AI answers that drew on everyone but the brand.
• Coverage too thin to be trusted, which is why AI pulled information from elsewhere.
• Content built for human impressions that machines couldn’t parse.
• A lack of depth, which required the right content architecture to fix.

This wasn’t a storytelling problem. The brand’s storytelling was strong. It was a depth and structure problem, in a landscape where machines are now a major audience.

It’s easy to assume AI simply knows. It doesn’t. It reads your content and derives meaning from what’s there. The brand’s stakeholders live and breathe the brand every day, so the gaps were hard for them to see. A machine visits, parses the page, and leaves with whatever it could understand.

What was working

• Exceptional brand storytelling: human-forward, distinctive and well crafted.
• A large archive of existing content that could be retrofitted rather than rebuilt.
• A starting point. The brand had already created a section to address the issue and state its position. It needed to be built out, not invented.

AI answers: other voices were telling the story

When people asked AI about the issue, the brand wasn’t the source.

• The LLMs answered differently, but none reflected the brand’s own position.
• The answers were synthesized largely from Reddit threads and press coverage.
• Where the brand’s own information was thin, other voices filled the gap.
• LLMs don’t defer to a brand because it’s the brand. They synthesize from the sources they judge most complete and consistent.

The effect: negative press extended into AI answers and was repeated back to people asking questions, including third parties the brand would never know about.

Depth: the coverage was too thin to be trusted

The brand’s position existed, but not at a depth an AI system would rely on.

• The section addressing the issue was brief and easy to miss.
• It read with a careful, corporate polish. It was comfortable for internal stakeholders, but it didn’t answer questions for the outside world.

Journalists were underserved too.

• A brand this size has an audience far beyond its customers: journalists, analysts, and now AI systems.
• My usability research with journalists showed they couldn’t find the facts they needed about the brand.
• The site was built around feature stories for a theoretical customer reader. Those customers weren’t the ones reading it.

Structure: content that came across as an exhibit

Content intended as a source of truth read more like an interactive exhibit than a structured source of information.

• Headlines didn’t name the topic. Read by a machine, the page gave no clear signal of what it was about.
• The writing was expressive and atmospheric. It circled the topic rather than stating it, which made it hard for humans to scan and hard for machines to parse.
• There was little of the structure LLMs rely on: descriptive headings, question-led sections, clear definitions, and short self-contained passages.
• Page titles and meta descriptions didn’t clarify the content.

The content was written like a magazine.

• Stories were long-form features, valued as complete pieces.
• Digital content needs to be atomized: one piece broken into clear segments that can stand on their own.
• Feature writing and structure aren’t at war. Done well, content hits the structural elements machines need and still carries the brand’s voice.

Architecture: stories without a cohesive structure

The existing architecture grouped stories, but didn’t tell one cohesive story about any topic.

• Content was mostly standalone feature articles, with a few landing pages pulling related pieces together.
• Each piece touched on part of a topic. None laid the topic out clearly and in depth.
• Factual reference information had been removed because it seemed too dry, yet that was exactly what journalists relied on. Machines need it too.

Discoverability: great content, few readers

The content wasn’t reaching the audience it was meant to influence.

• The brand didn’t depend on digital to drive business, so the site had no strong traffic or search strategy behind it.
• Stories weren’t actively promoted. Readers found them only by chance.
• Most customers engaged with the brand in its stores, not on its site.

This connected back to the client’s frustration that its stories weren’t shaping public perception. The site’s readership was small compared with the brand’s global customer footprint, so the people forming that perception largely weren’t reading it.

Why it happened

None of this came from a lack of care. The brand invested in telling its story well, and it showed.

What changed is the landscape. AI systems now read, synthesize and summarize a brand for people, and they reward depth, structure and clear statements of fact. That’s a different standard from the one most brand content was built for, and few teams have had to meet it yet.

A few common decisions added up:

• The content was designed for human impressions. Glossy features and a distinctive voice served the visitors who arrived, but gave machines little to work with.
• Factual reference content was cut because it seemed dry. That was the layer journalists and machines needed most.
• Sensitive topics were written for internal comfort. The language was careful enough to sign off on, but not complete enough to answer questions.
• Structure and creativity were seen as competing. The team worried that optimizing for search would make the brand sound robotic.
• PR and content worked separately. The messages that mattered most never made it into structured, evergreen content.

The optimization

The engagement delivered a GEO briefing for the team, the Authority Stack, a three-layer content architecture I developed, a section-by-section optimization of the team’s priority page, and a roadmap.