Your Customers Are Leaving Reviews. The Agent Can't Read Them.
Something I've noticed across several brands I've worked with over the past year is a specific and consistent gap between how much genuine positive customer sentiment exists and how much of it is actually visible to AI shopping agents.
The sentiment is real. Post-purchase surveys with strong NPS scores. Glowing DMs. Repeat customers who've bought every drop for two years. A genuine community of people who actively recommend the brand to others.
And almost none of it is in a format an agent can read.
It's sitting in internal survey tools, in Instagram comments, in WhatsApp groups, in email replies to post-purchase sequences. All of it inaccessible to the platforms that are increasingly deciding whose products to recommend before a customer ever visits the site.
This is a data infrastructure problem dressed up as a marketing problem. And most brands are solving it with the wrong frame.
Why the agent can't see your internal reviews
I've written about the agentic trust layer across this series in the context of GTIN ownership and Organisation schema. The same logic applies to review signals, and it's worth restating here because the mechanism is specific.
AI shopping agents are sceptical of self-reported signals. Anything a brand publishes on its own domain, which includes onsite reviews, sits in a category the agent treats with less confidence than signals that originate outside the brand's control. The agent can read onsite reviews if they're properly marked up with Review schema. But it weights them differently from signals on platforms it treats as structurally independent.
The historical parallel worth keeping in mind: this is the same logic as Google's PageRank. A backlink from an independent site carries weight because that site has its own credibility to protect. A review on Trustpilot or Google Reviews carries weight because those platforms have structural incentives to maintain signal integrity and are difficult to manipulate at scale. An internal NPS survey carries no weight to the agent at all, because the agent can't see it and wouldn't trust it if it could.
The brands that win in AI-mediated discovery are the ones who have built genuine review depth on platforms the agent already trusts. The brands that lose are the ones who have been collecting customer sentiment in places that don't exist to the agent.
How this applies to the fashion industry
Fashion brands have a structural disadvantage in review volume that other ecommerce categories don't face in the same way.
Customers buying from a premium streetwear brand are less likely to leave a review than customers buying a coffee machine or a mattress. The purchase is more personal, the customer relationship is more tribal, and the expectation of formal feedback is lower. You might have a deeply loyal customer base with a genuine affinity with the brand and a Trustpilot profile with fewer reviews. In my experience reviews for fashion brands is lower than many other industries I’ve worked within.
Drop model brands have an additional problem. Product-level reviews accumulate slowly on one-time SKUs that sell out in forty-eight hours. By the time you have enough product reviews to signal anything meaningful to an agent, the product is long gone. The review signal that matters for a drop-model brand isn't product-level. It's brand-level.
This is a point I made in the pre-drop preparation piece earlier in this series and it's worth returning to here. Trustpilot and Google Reviews are building your brand entity in the agent's understanding, not your individual products. The review volume and recency on those platforms is telling the agent: this is a real brand, real customers have transacted here, and enough of them had a positive enough experience to say so in a place they didn't have to.
That's a different job from a product review on your PDP. And most brands aren't doing both deliberately.
The response rate problem nobody talks about
One thing I've started checking on every brand audit I run is Trustpilot response rate. Not just review volume or average rating. How consistently is someone responding to reviews, positive and negative.
The reason this matters to the agent is the same reason it matters to a human. A brand that responds to reviews is demonstrably monitoring its customer experience in real time. The response itself is additional text on the platform, which increases the data density available to the agent when it's building its understanding of the brand. And a specific, considered response to a negative review is actually stronger trust signal than a generic five-star response, because it demonstrates the brand takes the feedback seriously enough to engage with it.
Most of the brands I work with have claimed their Trustpilot profile and left it largely unmanaged. Responses are occasional, generic when they happen, and there's no clear owner of the platform inside the team. It lives in the gap between customer service and marketing, so nobody is fully responsible for it and it doesn't get the attention it should.
That coordination problem is the same one I see with structured data work. Nobody feels responsible for it because it doesn't sit cleanly inside any one team's remit. So it doesn't get done, and the agent builds its picture of the brand from whatever signals happen to be there rather than from a deliberately managed presence.
How does an agent evaluate a brand?
I want to be specific about this because I think the abstract version of the argument doesn't land as clearly as the concrete one.
When an AI shopping agent is building a recommendation for a customer query like “what is the best premium streetwear hoodie in the UK" or "where to buy heavyweight cotton sweatpants," it's drawing on a combination of signals to decide which brands to include and what to say about them.
Structured product data tells it what you make and whether you're a legitimate entity. Organisation schema tells it who you are as a business. Review signals on independent platforms tell it what real customers think of transacting with you.
If the review signal is thin, the agent's characterisation of your brand will be thin. It may still surface you if the product data is strong. But the confidence attached to that recommendation will be lower, and lower confidence recommendations get displaced when a competitor with stronger independent signals is available.
I've tested this directly by querying brands in Perplexity, Gemini, Claude, and ChatGPT and asking about customer experience and product quality. The brands with active, managed review profiles on Trustpilot and Google Reviews get specific, confident characterisations. The brands with thin or unmanaged profiles get hedged, generic responses or no characterisation at all. The agent isn't making a judgement call. It's working with what's there.
What to actually do about it
The strategic shift is treating review generation as data infrastructure rather than customer service output. The question isn't "how do we get more five-star reviews." It's "how do we ensure the agent has enough independent signal to recommend us confidently."
For most brands that starts with three things.
Claim and actively manage your Trustpilot profile. Not just claim it. Set up a response workflow. Someone should be responding to every review within about forty-eight hours. Not a template. A specific response that acknowledges what the customer said. This is the work that builds data density over time and it compounds in a way that a burst of review requests doesn't.
Build review generation into your post-purchase sequence deliberately. Not as a generic "leave us a review" email buried in the shipping confirmation. A specific ask, timed well, that makes it easy for the customer to get to your Trustpilot or Google Reviews profile in one click. For drop-model brands this is especially important because the purchase enthusiasm is highest in the days immediately after the drop. That's the window to capture the signal.
For early access buyers specifically, there's an opportunity most drop brands aren't using. Someone who got early access to a drop and has already received their order is probably one of your most engaged customers. A targeted review ask to that group, timed about ten days after delivery, will generate a higher response rate and more considered feedback than a generic post-purchase sequence. That review goes onto Trustpilot before the next drop goes live and it's part of the signal the agent reads when the next wave of customers is researching the brand.
I'm still working out exactly how much weight review recency carries relative to volume in agent recommendation logic. My sense is that a steady cadence of recent reviews outperforms a large historical volume with a gap in the last few months, but I don't have enough data points to be definitive about that. The direction seems consistent across the agents I've tested.
Where this sits in the trust layer
The argument I've been making across this series is that the brands that will win in agentic commerce aren't necessarily the ones with the biggest budgets or the most sophisticated tech stack. They're the ones where the trust layer is coherent. Where the GTIN ownership is correct, the Organisation schema is verified, the authored content is in place, and the independent review signals are deep enough for an agent to make a confident recommendation.
Review signals are the layer most brands have the most existing material to work with and the least deliberate infrastructure around. The customer sentiment is often genuinely there. It's just not where the agent can find it.
Your customers are already vouching for you. The work is making sure that vouching happens somewhere the agent can read.