Public replies as brand voice.

The brand-voice document lives in a Notion page that nobody outside the marketing team reads. It contains the adjectives the brand uses to describe itself ("warm," "direct," "a little dry"). It contains the verbs the brand avoids ("revolutionize," "transform"). It contains three sample paragraphs in the approved register. It contains a section called "How we sound when something goes wrong" that, in most companies, was written once in 2023 and not touched since.

Then the brand publishes a review reply that says: "Thank you so much for your feedback! We're so glad you love the product. If you have any questions please reach out to support."

That sentence has nothing to do with the Notion page. It was not written by a person who has read the Notion page. It is the default reply that the platform suggested. It is signed by "Customer Care Team," which is not a person. It will be public on the product page for the lifetime of the product. It will be indexed by Google and crawled by GPTBot. It is, in citation terms, the brand's voice on that page.

The brand voice is the reply. The Notion page is a description of a voice that no buyer ever hears.

This essay is about why the review reply is the most under-used brand-voice surface in DTC commerce, and what it costs to ignore it.

Almost all brand-voice work is private. Style guides are internal. Voice training is internal. The team that knows what the brand "sounds like" is the small group that wrote the style guide. The team that produces public sentences in volume (customer support, fulfillment, occasionally the founder) often does not know the style guide exists.

This is true for ad copy. It is true for product descriptions. It is true for email. Each of those has at least an editor in the loop. The reply does not. The reply is written by a junior on a queue, optimized for speed, scored by reply rate, and shipped without editorial review.

The result is a public voice that does not match the private voice. Open any DTC product page with more than 200 reviews. Read the replies. Compare them to the brand's homepage copy. They are written, almost always, by different people in different registers. The homepage is in the brand voice. The replies are in the platform's default voice.

The companion piece to this essay walks through why the reply is, from a citation perspective, structurally privileged on a product page. See the review reply nobody indexed. The argument here is one rung over: even if the reply were not citation-shaped, it would still be the largest public voice surface the brand owns. The Notion page is 1,200 words. The brand's reply corpus on its top 50 SKUs is 80,000.

A brand voice in a reply has four properties that distinguish it from brand voice anywhere else. Each of them is structural, not stylistic.

The first is the level of warmth. A reply is being read by the customer who left the review and, simultaneously, by the future buyer reading the page. Pure warmth toward the existing customer ("So glad you love it! Hugs!") reads as performative to the future buyer. Pure transactional clarity ("Your order has been refunded, please allow 5-7 business days") reads as cold to the existing customer. The reply has to hold both audiences, which is a register that almost nobody is trained to write in. The closest cousin is the response a magazine editor writes to a published letter to the editor: warm but not familiar, specific but not procedural.

The second is the willingness to admit fault. Most brands refuse, in public, to admit that a product fell short. The brand-voice document, if it addresses negative reviews at all, says "acknowledge the concern" and "offer to make it right." This is a procedural framing. It is not a voice. A real voice in a reply admits, specifically, what the product did not do. "The new formula is less hydrating than the previous one. Some customers prefer the old. We kept both available for that reason." That sentence is more useful to the future buyer than any concession script. It is also more characteristic of a brand than any tagline.

The third is the use of specific product details. The bad reply talks about "the product." The good reply names it. The bad reply says "our team is here to help." The good reply says "the Vitamin C serum is at 15 percent L-ascorbic acid, which is the upper end of the clinical range; if you found it stingy, the 10 percent version (the Brightening Drops) is gentler." Specificity is voice, in a reply. The brands that have a voice are the brands that know their own product line at the SKU level. The brands that do not, sound the same.

The fourth is the avoidance of boilerplate. Every brand-voice document has a section on phrases to avoid. Almost none of them name the specific phrases that turn up in replies. "We appreciate your feedback." "We're sorry to hear about your experience." "Customer happiness is our top priority." "Please reach out to support so we can make it right." These are platform-default phrases. They are not voice. They are the absence of voice. A brand-voice document that does not explicitly ban them is a brand-voice document that does not cover the surface where the brand actually speaks.

A brief survey. Names used because their reply corpora are publicly indexable.

Glossier, in 2020 and 2021, replied to reviews with a recognizable register: lowercase first names of the reply writer, light specificity about ingredient changes, occasional admissions that a formula was reformulated and "some of you preferred the original." The voice was identifiable. By 2024, after Glossier's customer-care function had been outsourced, the replies on Sephora-syndicated reviews had reverted to the platform default. The voice had not moved to a new register. The voice had been replaced by no voice.

Bellroy, the leather-goods brand, has maintained a notable reply discipline. Replies are signed with a real first name. Replies cite specific wallet models when answering compatibility questions. Replies acknowledge product limitations specifically: "The Note Sleeve does not hold coins; the Coin Fold is the model for coin carriers." This is brand voice, applied at the SKU level, in public. It is also a substantial fraction of the reason Bellroy's product pages get cited by answer engines when buyers ask "which Bellroy wallet for coins."

Outdoor Voices, before its 2023 restructuring, had the strongest reply voice in its category: short, declarative, specific about fit ("the Hudson Shorts run a full size larger than the Move Free"). The replies were signed by people whose names appeared on the team page. After the restructuring, the reply corpus on archived pages reverted to "Thank you for your feedback." The reversion is observable in the timestamped reply pattern. It is also observable in the citations that ChatGPT now returns for "best running shorts for short torsos": Outdoor Voices is rarely cited, and when it is, the cited text is a customer review, not a reply.

The pattern across the three is the same. The brands that maintain a voice in the reply corpus get cited as having a voice. The brands that hand the reply queue to a generic process lose the citations and the voice in the same quarter.

The exercise that produces a citation-shaped reply voice is small and almost universally skipped. It takes four weeks, an hour a day, from someone in the brand-voice seat (not the customer-care seat).

Week one. Audit. Read the last 200 replies the brand has published. Read them in the order they appear on the product pages, not in the chronological order they were written. Mark each one as "in voice," "neutral," or "platform default." Tally. The result is usually 5 percent in voice, 15 percent neutral, 80 percent platform default. The 80 percent is the work.

Week two. Rewrite the platform-default replies. Not all of them. The top 50 product pages by traffic, the top 50 reviews per page. The replies should be specific (product name, customer name, the actual claim), signed (real first name, real role), and dated within the text where possible. This is a writing task. It is not a moderation task. The person doing it is an editor, not a customer-care agent.

Week three. Train the customer-care team on the new register. Three new rules. The reply must name the product. The reply must address the specific claim, not the general sentiment. The reply must be signed with a real first name. Drop the "Thank you for your feedback" preamble. Drop the "Customer Care Team" signature. The team will resist for two days and then adapt.

Week four. Audit again. The 5 / 15 / 80 should have moved to something closer to 30 / 40 / 30. The remaining 30 percent platform default is the hardest fraction; it will move to 10 percent over the next two months. By month three, the corpus is mostly in voice. By month six, the answer engines have re-crawled enough of the page to register the change.

We have argued elsewhere that the work in a review corpus is editorial, not procedural. See what an editor would do with your corpus. The reply voice exercise is the editorial work, applied to the surface where the brand speaks.

The default assumption is that customer care writes the replies. This is the wrong default for the same reason that customer support writing the help center is the wrong default. Reply writing is editorial work that intersects with customer-care knowledge. The intersection is not the same as the customer-care role.

The brands that have a public voice in their replies typically have one of three arrangements. The founder writes the hardest 5 percent (the complaints, the public refunds, the formula-change defenses). A senior editor (sometimes the head of brand, sometimes a dedicated content person) writes the next 30 percent, including the long-tail product questions where specificity matters. The customer-care team writes the routine 65 percent, but writes from a set of templates that have themselves been edited for voice, not from the platform's defaults.

The arrangement scales. The founder's 5 percent is fewer than a hundred replies a year for most brands. The editor's 30 percent is a few hundred. The team's 65 percent is the bulk, but it is bulk under a written voice guide that addresses replies specifically (not the generic Notion page). The total cost is one editor-day per week and a few hours of the founder's time per month. The compounding payoff is a reply corpus that is identifiable as the brand's voice for the lifetime of every product page.

The brands that do not have this arrangement spend zero hours per week on reply voice and ship 4,000 replies a year that are indistinguishable from every other brand's. They get one of the largest public content surfaces they own for free. They use it to publish "Thank you for your feedback."

None of the above requires a different platform. All of it would be easier with one.

The platform-level changes are small. Replace the one-click reply template with a forced draft that the operator must edit. Surface, in the reply composer, the product name and the customer's first name automatically. Require a signature field that is a real name, not a role. Drop the "reply rate" metric from the dashboard and replace it with "reply specificity," measured by the ratio of unique tokens in the reply to total tokens (a crude proxy, but a real one).

The deeper change is to stop treating the reply as moderation and start treating it as content. Moderation is a queue. Content is a craft. The platform that ships a content-grade reply composer ships a brand-voice tool that no competitor has thought to build. See reviews are language not inventory.

The brand voice is not the Notion page. The brand voice is the 80,000 words the brand has already published, in public, on its own product pages. The Notion page is a description of a voice that the brand has been refusing to use.

Every brand spends six months on the homepage. Every brand spends three weeks on the email welcome series. Every brand spends a year on the product description. Almost no brand spends a day on the reply corpus, which is, by volume, larger than all three combined and more durable than any of them.

The reply corpus compounds for as long as the product page exists. It is indexed. It is cited. It is the public voice of the brand on the page where buyers are deciding whether to buy. The platforms that treat the reply as moderation are leaving the brand's most-published voice surface in the hands of a junior on a queue. The brands that take the reply seriously are writing themselves into the citation graph for the next decade. The work is editorial. The surface is public. The voice is already there. Someone has to write it.