Journal
Essays on reviews, language, and the asset almost every store is sitting on without reading. Published from the studio, dated, and corrected in the open.
48 essaysIssue I to XLVIIIAug 2025 to May 2026
Start here
Three essays that carry the argument the rest of the run builds on.
The citation economy
AI answer engines do not return ten links. They return one paragraph and four citations. Commerce, quietly, has become a citation contest. The brand whose first-party language is most citation-shaped wins.
Why review widgets are a category error
Reviews are not inventory. They are language. The thing under the buy button is the wrong shape for what is actually inside it.
The dead text of e-commerce
A store collects, on average, twenty thousand sentences from its customers and reads almost none of them. The most useful writing in commerce is also the most ignored.
Latest
Why our Journal is dated
Every essay we publish carries a visible date in the byline. We do not refresh older essays. We do not silently re-date pieces. A short studio note on what that decision costs and what it earns.
Three properties of a citable sentence
The companion piece to first-person, dated, signed. Each property has edge cases that change the citation weight. What happens when a sentence is first-person plural, when a timestamp got refreshed, when the signature is an obvious pseudonym.
On editing customer language
A brand wants to lift a customer's sentence onto a product page. The FTC says no material alteration. The citation framing says keep the verb, the proper noun, the specifics. A working editorial rule, with three test cases.
How we read a corpus
A studio note on what happens when a single editor sits down with a small brand's entire review history and treats it as a manuscript. The exercise looks like literary criticism. The output is closer to a research brief than to a dashboard.
The contents
The whole run, filed by subject. Every essay appears once, under the question it is actually asking.
Six volumes
The run files into six subjects. Each one is a volume with its own contents, its own length, and its own argument.
On reviews as language
Why the writing under the buy button is the asset, not the stars above it.
18 essays2 hr 38 min of readingnewest first
Three properties of a citable sentence
The companion piece to first-person, dated, signed. Each property has edge cases that change the citation weight. What happens when a sentence is first-person plural, when a timestamp got refreshed, when the signature is an obvious pseudonym.
On editing customer language
A brand wants to lift a customer's sentence onto a product page. The FTC says no material alteration. The citation framing says keep the verb, the proper noun, the specifics. A working editorial rule, with three test cases.
How we read a corpus
A studio note on what happens when a single editor sits down with a small brand's entire review history and treats it as a manuscript. The exercise looks like literary criticism. The output is closer to a research brief than to a dashboard.
The discipline of writing back
A customer writes a thoughtful email and almost always receives a form. The companies that respond in kind, specific and signed and willing to be wrong, are doing both an ethical job and a strategic one. Every reply is now indexed.
What a magazine knows that a SaaS doesn't
A magazine treats the reader as an adult. A SaaS blog treats the reader as a conversion event. In 2026, the magazine register is no longer a stylistic choice. It is a citation primitive.
The hardware store and the AI answer
The clipboard behind the counter and the answer engine in 2026 do roughly the same work. They both keep track of who said what about which thing, and they both reward a brand that keeps specific things in specific places.
The product page is now a corpus
In 2018 a product page had one job. In 2026 it has three, and the new one is the largest. The page has to host the writing of the people who already bought, in a form the next buyer and the next crawler can both read.
Why Reddit eats your citation share, and what to copy from it
In Q1 2026, Reddit URLs appeared in ChatGPT citations at roughly four to seven times the rate of typical brand product pages. The shape of a Reddit thread is the reason. The fix is not to post on Reddit. It is to publish a page that is structurally a Reddit thread.
The long-tail keyword that already lives in a one-star review
SEO teams pay agencies to surface the questions buyers ask before they buy. The questions are already inside the corpus, written by the buyers themselves, mostly in the one-star reviews. Reading them is the work.
Q4 returns and the review your customer leaves in February
The most useful review a brand will collect this winter comes from the buyer who returned the wrong size in December and bought the right one in January. Most platforms suppress that buyer by default.
Reviews are language, not inventory
A sentence does not represent the other sentences. Treating a body of customer writing the way you treat a stockroom is the original mistake the category was built on.
The verified-buyer paragraph, and why the badge isn't enough
A green check next to a name is not a citation. The signal an answer engine reads is the paragraph behind the badge, and most badges are sitting on no paragraph at all.
Public replies as brand voice
Most brand-voice work happens in private. The review reply is the one place where brand voice goes public, dated, and indexed forever. It is the most under-used voice surface in DTC commerce.
The review reply nobody indexed
Every review platform offers the reply as a moderation feature. It is the only public, first-person, dated, signed sentence the brand is allowed to put on its own product page. The answer engines have noticed. The platforms have not.
The long tail begins inside the review
SEO teams pay for keyword research. Their customers wrote the keywords for free, in the reviews, and the team did not read them.
What an editor would do with your review corpus
A literary editor does not run sentiment scores on a manuscript. They read it. The same posture, applied to the writing your customers have already done for you, will produce work no dashboard can.
Why review widgets are a category error
Reviews are not inventory. They are language. The thing under the buy button is the wrong shape for what is actually inside it.
The dead text of e-commerce
A store collects, on average, twenty thousand sentences from its customers and reads almost none of them. The most useful writing in commerce is also the most ignored.
On citation and answer engines
How AI search engines now read commerce, and which sentences they quote.
14 essays2 hr 17 min of readingnewest first
Why our Journal is dated
Every essay we publish carries a visible date in the byline. We do not refresh older essays. We do not silently re-date pieces. A short studio note on what that decision costs and what it earns.
The robots.txt that quietly blocks your AI buyer
Shopify's default robots.txt lets Googlebot in. It says nothing about GPTBot, ClaudeBot, PerplexityBot, or Google-Extended. In 2026, the default is the wrong answer.
FAQPage on the product page, or a separate Q&A page. Both lose.
Google deprecated FAQPage rich results for almost every domain in August 2023. QAPage was never the right shape for product pages. Most stores are still shipping one of the two anyway.
sameAs, and the entity you forgot to declare
A two-line addition to a schema block, shipped once, that tells every AI search engine which company you are. Most stores have not shipped it.
First-person, dated, signed
Three properties that decide whether a sentence will be cited by an AI answer engine. Reviews can be all three. Marketing copy is none of them.
Bazaarvoice's Authentic Discovery API, read by someone who isn't selling it
The pitch is that the API exposes your UGC to AI agents. The skeptical read is that it solves Bazaarvoice's distribution problem, not the brand's domain problem. The product page is still empty.
The half-life of a product page
Two product pages launched on the same day. One decays. The other compounds, in the language of its own customers, for years.
The FTC Consumer Review Rule, read as a content brief
The same federal rule that bans fake reviews also describes, in plain language, the kind of writing AI answer engines now reward. Compliance and citation are the same brief, read from two ends.
What ChatGPT reaches for when it recommends a serum
A close reading of the sentences a major answer engine actually quotes when a buyer asks about a product. The cited line is almost never the one a copywriter wrote.
The citation economy
AI answer engines do not return ten links. They return one paragraph and four citations. Commerce, quietly, has become a citation contest. The brand whose first-party language is most citation-shaped wins.
Iframes don’t pass link equity. Review widgets are iframes. Do the math.
A backlink to a five-star review on the merchant’s product page is, in many configurations, a backlink to a different domain entirely. The link equity accrues elsewhere. The Cumulative Layout Shift accrues to the merchant.
The raw HTML your review widget never delivers
GPTBot does not run JavaScript. ClaudeBot does not run JavaScript. PerplexityBot does not run JavaScript. The reviews on the merchant’s product page are written in JavaScript. Do the substitution.
The engine the answer engine reads
The input layer to generative search is not your marketing copy, your schema, or your backlink graph. It is the sentences your customers wrote. Most stores have not noticed.
What the search engine became
In a single calendar year, the box you typed into stopped giving you links and started giving you sentences. The sentences are written by other people. Your customers, mostly. Whether you noticed or not.
On widgets and crawlers
The shape of what holds the writing, and why it is the wrong shape.
5 essays53 min of readingnewest first
The thirty thousand product pages Baymard reviewed, and what they said about reviews
Baymard Institute has spent fifteen years recording how buyers actually use product pages. Their findings on reviews are the most authoritative, the most ignored, and roughly the opposite of what the citation engine now needs.
Why review pages should sometimes have their own URLs
When a product has accumulated two hundred reviews, the reviews stop being a UI block under the buy button and start being a document. Documents need addresses.
The end of the review widget
An elegy for a UI component that fossilised. Twenty-three years after the first widget shipped, the shape is finally the wrong shape for what it holds.
The cumulative layout shift of a star rating
Google penalises pages whose content moves while it loads. The star widget under the buy button moves the page. Reading the score, then the cost, then the alternative.
A curl against three crawlers, three review platforms
A field experiment. Three Shopify stores, three review platforms, three AI user agents. A nine-cell grid, six of them empty. The empty cells are the story.
On the operator's work
Dashboards, replies, migrations, timing. The actual job, not its description.
8 essays1 hr 16 min of readingnewest first
Quiet metrics versus loud metrics
A dashboard reports the loud metrics because they are easy to count. The quiet metrics are what an answer engine actually weights. The gap between the two is where most of the real work hides.
The review request email no one opens
It is the most templated, lowest-craft customer touchpoint in DTC commerce. Klaviyo benchmarks open rates at 40 to 45 percent. Review-request emails specifically run 15 to 25 percent lower than that. The subject line is the tell.
Klaviyo, and the 14-day window that became a four-week window
The default review-request delay in most flows is a number from a 2018 study that no longer fits the products it gets applied to. The right window is no longer one number. It is a category.
Software that remembers
Operator software in commerce has split, quietly, into two species. The one that counts and reports. The one that remembers and acts. The first is being commoditised. The second is what wins the decade.
Re-warming review velocity after a platform switch
Fresh-review volume drops 30 to 60 percent for the first month after a migration. Three forces produce the drop and one operator playbook reverses it.
The migration that takes a month, not a weekend
Every review platform help-doc promises a CSV export and a CSV import. The actual migration is thirty days long, costs you a third of your fresh-review velocity, and quietly orphans the citations you spent two years earning.
The dashboard is not the work
A dashboard is a description of work that has already happened. The work itself is replying, deciding, fixing, writing back. Almost no software does any of it.
Hardware stores knew this in 1968
Independent retailers used to remember the specifics of every customer they had. Software was supposed to give us that back. It mostly took it away.
On compliance and evidence
The FTC rule, the schema, the falsifiable claim. The legal and the citable, read together.
2 essays16 min of readingnewest first
Self-serving review markup, and the line Google drew in 2019
In September 2019, Google quietly killed star snippets for a particular kind of schema. Most stores never noticed. Many are still violating the policy seven years later.
ClaimReview, and the schema your competitors won't ship
A schema invented for fact-checkers, sitting unused on every product page that ever made a falsifiable claim. The brands willing to put their copy under it will be the brands the engines learn to trust.
On editorial method
Studio notes on the practice of writing about commerce.
1 essay4 min of readingnewest first
On stealth
A short note on why we have, for the time being, almost nothing on this website. Stealth is a posture. It is also a kindness.