Your AI brand story is table stakes. Here’s what earns trust instead

AI positioning claims are cheap and easy to copy. The only thing that holds up under a skeptical buyer is proof. Michelle Huff, Chief Marketing Officer at Alteryx, lays out a practical framework for building that proof, one any B2B marketer can borrow regardless of company size.

Every B2B vendor now has an AI brand story. Scroll any category and the claims start to blur together: smarter, faster, more autonomous, built for the agentic era. When everyone is saying roughly the same thing, positioning around AI stops being a differentiator and becomes table stakes, and trust starts getting decided somewhere else entirely: in the gap between what a vendor claims and what a buyer can actually verify. That was the starting point for a recent conversation on the Trust and Influence in B2B podcast, where I sat down with Michelle Huff, Chief Marketing Officer at Alteryx, who has spent the past year trying to move her company’s messaging past the point where words alone can do the work.

Proof over brand promise

Huff’s answer to the sameness problem isn’t a sharper slogan. It’s evidence. “It’s not just what you say,” she told me, “because sometimes it’s just the words.” Her team’s focus has shifted from crafting claims to manufacturing proof: demos that show the mechanism rather than describe it, and customer stories that carry more weight than anything marketing writes itself, because. As she put it, “other leaders, other users, when they hear someone else who’s tackling a real world problem, there’s just a different level of credibility.”

The specific mechanism Alteryx leans on, a large existing user community that generates its own advocates, won’t be available to most marketers reading this, and it’s worth being clear-eyed about that: not every brand has a base that size to draw proof from. What travels regardless is the underlying discipline. If your positioning depends on a claim nobody else can independently verify, it isn’t positioning, it’s assertion. The test worth applying to your own AI story is simple. Strip out every adjective and ask what’s left that a skeptical buyer could actually go and check for themselves, without taking your word for it.

Run there fast, but don’t outrun what’s real

The harder balancing act, Huff argued, is holding two things at once: articulating where the market is heading while staying honest about what the product does today. She borrowed a line from her years at Salesforce: “We want to see where the future’s headed. We want to run there fast so we can turn around and greet them.” It’s a good description of the job. Vision earns attention, and buyers do want to know a vendor is thinking past the current quarter. But when the sales deck starts describing capability that doesn’t exist yet, the same vision becomes the thing that erodes trust rather than building it. As Huff put it, describing the discipline her own team tries to hold: “There’s vision, and then there’s actually what the capabilities are and how we solve for that.”

This is where the self promotional instinct that AI brand marketing rewards becomes a genuine liability, for any vendor, not just this one. It’s easy, when you’re excited about where a category is going, to let roadmap slide into pitch without quite meaning to. The discipline worth holding onto, more cleanly than any single conversation ever practises it, is to separate the “where we’re going” conversation from the “what you get on day one” conversation, explicitly, every time you write or say either one.

An AI framework worth stealing

The most useful thing to come out of the conversation wasn’t a company anecdote at all. It was a simple framework Huff’s team uses to decide what’s actually ready to be automated: visible, understandable, repeatable, auditable. Visible and understandable address the black box problem, the sense that a recommendation arrived from nowhere and can’t be interrogated. “It’s kind of speaking Spanish,” Huff said. “You kind of get it, but I don’t even know if I’d sign a contract in it.” Repeatable and auditable address something different: whether the same input reliably produces the same output, and whether anyone, months later, can reconstruct why a particular decision was made.

She illustrated the last two with an invoice automation example. Rather than trusting extracted numbers outright, the process scores its own confidence, flagging an invoice as roughly 90 percent reliable or 40 percent reliable. Anything below a set threshold gets routed to a human for review, and the threshold itself moves over time as confidence in the system grows. It’s a small piece of engineering, but it’s a genuinely transferable idea for any marketing team building its own AI workflows: don’t build a single on off switch for trust. Build a dial, and let evidence move it, rather than deciding once and walking away.

AI judgement, not just tasks

Huff’s clearest reframe was about what AI is actually for. Most of the public conversation is still about speed: doing the same task faster. She argues the more consequential shift is judgment: reasoning based decisions that used to require a human involved from end to end. Her way of thinking about deployment borrows from how you’d manage a new hire. “You probably wouldn’t shut them off and just give them full authority and tell them to do everything, and not inspect at all,” she said.

Her own team’s content pipeline is a reasonably honest example of what that looks like in practice, self promotion aside. A researcher agent pulls together background on a topic, a draft gets written automatically, and then a human reviews it before it goes anywhere near publication. Editing agents take another pass, but a person still checks the finished page before it goes live. It’s a workflow most marketing teams building their own AI processes would recognise, or should. Automate the assembly, keep a human at the checkpoints that matter, and only remove a checkpoint once the process has earned enough of a track record to justify it. Trust gets extended incrementally, and pulled back the moment something breaks the pattern.

It’s a useful discipline to sit alongside the career question Huff raised near the end of our conversation, about what keeps a marketer relevant rather than automatable. Her answer was to stop describing your job as a list of tasks (publishing the social calendar, generating the email campaigns) and start describing it as a set of problems you own. Task owners are exposed to automation by definition. Problem owners get to keep deciding how the problem gets solved, whichever tools end up doing the work. It’s the same logic, in the end, that governs whether a machine earns trust: not asserted all at once, but built through evidence, one decision at a time.

What to do with this

None of this requires a 25 year old brand or a community numbering in the hundreds of thousands. It asks for a smaller, more disciplined set of habits, most of which cost time rather than budget.

• Audit your own AI brand positioning for claims a buyer can’t independently verify, and replace at least one of them with a demo, a proof point, or a customer willing to say it in their own words instead of yours.
• Separate your vision language from your capability language in every piece of content you publish, so a reader can always tell which one they are reading. Never let one borrow credibility from the other.
• Borrow the visible, understandable, repeatable, auditable test for any process you are about to automate, and be honest with yourself about which of the four it currently fails.
• Build a confidence threshold rather than a single trust decision for at least one automated process. Exceptions get reviewed rather than silently shipped, and let that threshold move as the process earns a track record.
• Reframe your own role, or your team’s, around the problems you own rather than the tasks you perform. It is a better defence against automation than any tool you could buy, and it happens to be true regardless of what AI ends up being capable of.