AI: The Reputation Signal · Issue 8
The payment went through. Trust didn't
Elif Güvençer ·
Not in a sandbox. Visa and more than thirty European banks and issuers ran live purchases at real merchants including lastminute.com, Frasers, Cleverbridge and BrickDepot. An agent reviewed products, selected one and completed the payment, within limits the cardholder had set.
(If we wish to be literal about it, agents probably bought holidays before July 2 but it was announced that day).
Now agentic commerce is in early stages. The disclosed transactions number in the hundreds, not millions, and how it evolves is an open question. But the direction of travel is clear and investment is flowing that way. See other examples from: OpenAI, Google and Mastercard.
The comms industry read the moment through a different lens. Sarah Evans at Zen Media called it the first true attribution in PR history: your content gets cited in an AI answer, the answer sends traffic to your site, and now the same answer can end with an agent completing the purchase. "Cited, clicked, bought." Every link in the chain measurable, and the first link is comms work.
It is early days. The attribution is not always clean and might not be always easy to measure but theoretically the line holds and can really change things for communications. Imagine being able to bring into the management room not opportunity to see (OTS) or coverage volume but concrete proof that your communication shaped the conditions under which a purchase happened.
I have been talking about AI being a structural stress test that exposes whether the communications function's value is upstream, or whether it is a sophisticated production function that has mistaken output volume for strategic influence.
I have said the same technology disrupting communications also gives it the first real shot at its oldest weakness: measurement. The attribution infrastructure; causal proof that this piece of comms drove that commercial outcome doesn't really exist yet. But for the first time, it is buildable. That is where the focus should go. You can read more about this on the Two Clocks Framework™.
So…until this moment, the potential attribution trail of agentic commerce is good news.
I will pump the brakes, because there is a risk comms needs to consider before jumping in, and it is not only about agentic commerce, it is about agentic anything: the trust consequence.
According to Forrester, three quarters of online adults in the US, UK and Canada are uncomfortable letting an AI agent complete a purchase and pay on its own, even with spending limits they set themselves. They will let the agent recommend. They will let it compare. Some will even let it fill the cart. The payment is where they stop. And what trust exists is brittle: a YouGov survey in the UK says six in ten (60%) would stop using an AI agent after just one mistake.
AI has a reputation problem
I was part of the inaugural Baduel Brief hosted by Farzana Baduel and moderated by David Gallagher a few weeks ago. We tackled whether AI has a PR problem, or PR has an AI problem. My answer to the first has been about trust.
AI doesn't have a PR problem. It has a reputation problem.
AI has become enormously powerful without first building the stakeholder trust needed to sustain it. The narrative around the technology has done more to amplify fear than to demonstrate its transformative potential. At the same time, corporate messaging is often perceived as extracting value through AI rather than creating value with it.
The challenge is that AI's reputation has no single owner. It is a shared reputation, shaped collectively by governments, AI labs, corporations, academia and the media. When everyone contributes to the narrative, no one is fully accountable for it. As a result, the collective has spent more trust than it has built.
The RepTrak Company's normative research shows that three in five people (61%) believe AI needs more regulation and oversight, nearly half (48%) worry it will result in fewer jobs, and 46% are concerned about digital privacy.
At the same time, attitudes are not uniformly negative: a third (33%) believe AI makes businesses more effective, and one in three (33%) say it is improving their own lives.
That tells me AI's reputation isn't simply positive or negative—it's conflicted. People are beginning to recognise its value, but they remain uncertain about its broader societal impact.
So for anything agentic to work, especially where money changes hands, that trust has to be restored: first in AI itself, then in agents acting on behalf of humans.
There is a big role for communications in bridging that trust gap. And comms leaders, especially in for-profit organisations, will be increasingly pressured to build that trust, and fast. The commercial incentives as clear as day.
The first step in doing that is asking this question: I am expected to build trust in the agent, but did it build trust in me?
If you cannot honestly answer this question as yes, the consumers won’t either.
What agents are doing for us vs to us
Ravi Dhar and Jon Iwata published a piece in Fortune last month, drawing on Yale School of Management's interviews with more than 200 CEOs.
They argue trust in the agentic era through a reality that already exists: surveillance pricing, using a person's data profile to set their price. Their test is one question: is the company pricing the transaction, or pricing the person? Two riders waiting for a car on the same corner at the same moment, paying different fares, because an algorithm inferred one of them will accept it.
Then the piece turns to agents in a future scenario. An agent managing your household knows you're low on medication. It knows you shop when you're stressed. It knows you rarely compare prices. In their words, that profile "becomes a roadmap for either serving your interests" or "exploiting your patterns for maximum extraction."
On the signal architecture dimension of the Two Clocks Framework, I talk about how the total picture AI draws of an organisation is not owned by any single function. Investor relations owns the earnings narrative. Communications owns media and executive voice. Marketing governs brand. Legal reviews the disclosures. Each does its job, on its surface, to its standards. Nobody reads the whole. That was manageable before; AI is changing that. There is now an intermediary capable of reading the sum of all those parts and handing it over to anyone who asks.
Reading their piece made me think about the other half of that argument. The fragmentation that let inconsistency slide didn't only protect reputation. It protected the consumer too. Companies' consumer data was just as fragmented; no single company ever saw more than a slice of anyone's life. The agent removes that protection from both sides in the same pass. It reads the whole of your reputation shorthand with all the contradictions and gaps and as people delegate more of their lives to it, it reads the whole of how a person lives. What it does with that second reading, and in whose interest, is the question that will build trust in agents or remove it.
All of this reframes the three quarters who won't let an agent pay. They are not a friction problem awaiting better messaging. They are doing the second reading themselves. And that position will move by how the companies behave, not by what they say.
Communications is the function that can see the trust consequence of a decision that looks fine from every other angle. Plugging your brand into agentic checkout looks defensible from most corporate seats, and the functions operationalising it have many commercial reasons to say yes.
Yes can also be the right answer. But it will only be credible if the organisation has adopted AI responsibly and communicated transparently about it: governance, data, privacy, what the agent is allowed to do in the customer's name. What kind of accountability does it demonstrate when things go wrong?
Consumers seem to have already decided that one, by the way. Forrester found they expect the answer engine or the AI platform to carry the blame when an agent gets it wrong. Not the retailer. Not themselves.
(There is also the uncomfortable question of even when consumers are in control and setting parameters, how clear are they on what they are consenting to and the potential risks? I will park this for another post).
That credibility will also depend on how public trust or lack of it evolves around AI. Because that will be the ceiling for any organisation, even if they are doing everything right.
Either way, before asking consumers to put their trust in an agent, comms should ask the reverse question first: has the agent built trust in me? The three quarters who won’t allow AI to pay will move when it has.
Not before.