Understanding AI Behavior Is Becoming Part of Brand Strategy

At the Fast Company Innovation Festival, I got talking with a CMO who works closely on branding and storytelling. We were discussing how AI represents brands and where people with backgrounds like ours fit into that work.

I shared how my experience in organization development shapes what we're building at Envoyra. I'm interested in behavior: noticing patterns, understanding the context, and figuring out what we can learn from what we observe.

That led us to a question: how does a brand's story come across to a person, and how does it come across in an AI-generated answer?

She encouraged me to write about it. The question stayed with me because it matters whether you work in branding, marketing, sales, or run the business yourself:

If AI describes your company to a prospective customer, will that person understand why you might be the right choice?

You know what gets lost in a short description

If you work in sales, you have probably had a conversation with someone who arrived with an incomplete impression of your company.

They know what you sell, but not who it is best for. They have seen your price, but do not understand what is included. They think you are similar to a competitor, and you spend much of the conversation explaining the difference.

That difference may be the very thing your team has worked hardest to build.

Perhaps you make a complicated purchase easier. Perhaps your service is especially helpful for beginners. Perhaps customers stay because your team understands their circumstances and follows through.

A list of products and features may be accurate while missing much of the reason someone chooses you.

This is a useful starting point for thinking about AI and brand storytelling. When a prospective customer asks an assistant for advice, how much of that meaning appears in the answer?

What I mean by AI behavior

When I talk about AI behavior, I mean what we can observe in its answers: the claims it repeats, the sources it cites, the comparisons it makes, and the brands it recommends.

I do not mean that AI has human feelings or motivations.

With people, we can investigate what they understood, remembered, felt, and did. With AI systems, we can examine what an answer included, what it left out, and whether that pattern holds when we ask again.

The context matters in both cases. For an AI test, that includes the question, the system being used, and whether it can retrieve information from the web.

We also need to avoid an overly simple distinction between emotional humans and factual machines. People care about evidence. AI answers can contain emotional language. The useful comparison is how we study each response and what it tells us about the way a brand comes across.

For a business, the question is practical: does the answer help someone understand what makes us relevant to their situation?

AI can recognize a brand and still leave it out

One finding from our August pilot work at Envoyra made this distinction clear.

An AI system with web retrieval enabled recognized a brand in all five answers to questions that named it directly.

We then asked that same system 18 buyer-oriented questions that did not name the brand. It appeared in none of the answers.

Two days later, we repeated the questions. The result was unchanged.

5/5 vs 0/18 | Named-brand recognition vs unaided appearances — same brand, same AI system, unchanged when retested two days later

This was a small test of one brand under a particular configuration. It does not tell us how common this pattern is across the market. But it shows why asking AI about your company by name can give you an incomplete picture.

Your next customer may not know your name yet.

They might ask what suits their budget, which product would work in their circumstances, or where to find a particular kind of support. Being recognized when someone supplies your name is different from being brought into that conversation.

Across two pilot brands, we also found six answers that listed a brand-controlled page among their sources, left the brand itself unnamed, and named alternatives. Some were repeated observations of the same questions on different dates.

The citation list does not reveal exactly how the answer was produced. What we could see was that the source appeared while the brand did not.

That gives a team something specific to investigate. It does not, on its own, explain the cause.

Brand judgment belongs in this work

Imagine a fictional company that helps beginners take up a new activity.

Its story is about making the first attempt feel manageable. It supports that promise with patient instructions, approachable staff, and products designed for people who are still learning.

Now imagine an AI answer describing it as a premium specialist and listing its products and prices.

The individual facts might be correct. But someone reading that answer could reasonably decide, "This sounds too advanced for me."

A person who understands the brand and its customers would notice the problem.

They would also know which questions to ask next. Is the beginner-friendly promise supported by the actual experience? Is it clearly explained in the available information? Was it relevant to the question? What would a more useful description include?

This is why people in brand, marketing, and sales need to help define what we test.

They know the misunderstandings that come up in conversations. They hear the objections. They know which distinctions help a customer make a decision.

That knowledge can help us choose better questions and judge whether an answer is useful, accurate, and complete enough for the situation.

The technical side is part of the picture

At the festival, I also attended Design Bridge and Partners' SoHo Fast Track, "Branding for a Two-Brained World: Share of Mind and Share of AI Model." The session connected emotional storytelling with the factual evidence supporting AI citations.

As I remember it, one example involved comparing a webpage with JavaScript enabled and disabled. Some pages deliver important information immediately; others require code to run before that information appears.

Vercel and MERJ's December 2024 research found that several major AI crawlers they observed did not execute JavaScript. Google documents JavaScript rendering in its own search infrastructure, with limitations. The capabilities vary, so turning JavaScript off is a useful initial check rather than a universal view of what AI can access.

Our client scans did not include rendering tests or crawler logs, so we cannot attribute their results to JavaScript.

The broader point is that access and storytelling need to be examined together. A technical team can investigate whether information is accessible. People who understand the business can help judge whether the resulting answer communicates what matters.

What should a team do with a finding?

This is where my organization development background shapes the work.

A surprising answer is a starting point. The next step is to understand it well enough to decide what deserves attention.

Sales can bring questions and misunderstandings from customer conversations. Insights teams can investigate how widely those concerns are shared. Brand and product teams can verify the relevant promises and facts. Content, digital, and communications teams can examine how those claims appear across the company's website and external sources.

The company's own story needs scrutiny, too. Before deciding that AI got something wrong, we should check whether our claims are accurate, supported, and consistent.

Then someone needs to own the next step.

That might be correcting conflicting information, making an explanation clearer, or investigating an unexpected source. If a change is made, record it and arrange a comparable retest.

Our available pilot records do not yet establish that a documented change caused better recommendations. This is a proposed way to learn, not a guarantee that a particular edit will change an answer.

We have also learned to question the measurements themselves. In our review, some automated labels counted an answer as recognition because it repeated the brand's name while saying it could not verify the brand.

Reading the answer changed the classification.

That is why I want teams to be able to look beyond a score and understand the evidence behind it.

Your knowledge of the customer matters here

That conversation at the festival helped clarify what I want to contribute to this space.

Understanding AI behavior can give brand leaders another way to examine how their story reaches people. It can give sales teams a way to investigate the descriptions and comparisons a prospect might encounter. And it can give those teams a shared set of questions to work on with their technical colleagues.

The people closest to the brand and its customers have an essential role.

What should remain recognizable when the story is shortened? Which distinctions matter in a buying decision? What could mislead someone, even if the individual facts are correct?

At Envoyra, we examine how brands appear in answers to buyer questions and track those results over time. My organization development perspective shapes what I want that evidence to support: clearer conversations, better questions, and decisions with an owner.

A useful place to begin is with the questions your customers ask before they know to ask for you.

If you're wondering how AI describes your company to a potential customer, start with three buying questions your sales team hears regularly. Bring them to Envoyra, and we can discuss a diagnostic to examine the answers, where your brand appears, and what deserves attention.

[CTA: Bring us your buying questions | /contact]

---

Methodology and attribution

These observations come from Envoyra's review of 316 locally stored response records across two pilot brands and four scans conducted August 10–19, 2026. Examples included in this article were read by hand. The recognition-versus-discovery comparison uses one retrieval-enabled engine. The wider dataset includes different configurations, with some historical settings unconfirmed. These exploratory findings are not market estimates and do not establish effects on traffic or sales. Supporting methodology and anonymized findings are available on request.

The festival conversation and JavaScript discussion are the author's recollections, paraphrased rather than presented as direct quotations. The linked session listing verifies the event details.

Golara Serio is a co-founder of Envoyra and an organization development practitioner.

Sources: Fast Company Innovation Festival session listing, "Branding for a Two-Brained World": https://events.fastcompany.com/innovation-festival/session/4488422/branding-for-a-two-brained-world-share-of-mind-and-share-of-ai-model Vercel and MERJ, "The rise of the AI crawler" (December 2024): https://vercel.com/blog/the-rise-of-the-ai-crawler Google Search Central, JavaScript SEO basics: https://developers.google.com/search/docs/crawling-indexing/javascript/javascript-seo-basics