The product page is now a corpus.

In 2018, a product page on a Shopify storefront did one job. Its job was to convince a single human visitor to click the buy button. The page was, in document terms, a sales letter with a price field attached. The hero image carried the brand. The headline carried the proposition. The body copy carried the persuasion. The reviews, if any, sat in a small widget below the fold, with an aggregate star rating doing most of the work.

In 2026, the same page has three jobs. The first is unchanged: convince the human visitor. The second is newer: present a structured, server-rendered document to a class of AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended) whose citations now drive a measurable share of high-intent product traffic. The third is the largest of the three, and the one this essay is about: host the corpus of customer writing that has accumulated against the product since it launched.

The corpus shift is the new part. It is also the one DTC commerce has not architected for.

The word corpus, used by linguists and editors, describes a defined body of text studied as a unit. A poet's collected works is a corpus. The Old Bailey court records is a corpus. The published letters of a 19th-century scientist is a corpus. The defining feature is not the topic, but the boundary: the corpus has an edge, and what falls inside it can be read together, compared, indexed, quoted.

A product page in 2026 has, attached to it, exactly this kind of body. The reviews are the obvious part. The buyer questions and answers are part. The replies from the brand are part. The product-specific Reddit threads that the brand does not control are an external part of the same corpus. The press mentions, the blogger writeups, the YouTube transcripts: all of these are documents about the same product, written by humans, sitting at various URLs, increasingly indexed and increasingly cited.

For a product that has been on the market for three years, the corpus typically runs to between fifty thousand and three hundred thousand words. Most stores have not read it. Most pages display less than one percent of it. The corpus is the largest body of writing the brand owns, and the one the brand pays the least attention to.

In the dead text of e commerce, we wrote about a store that has, on average, twenty thousand sentences from its customers and reads almost none of them. The corpus framing is the architectural form of the same observation. The text exists. The text is owned, or co-owned, or at least hosted. The text describes the product better than the brand can. The product page can either be built to surface that text, or built to ignore it. The default architecture ignores it.

The three jobs of a 2026 product page correspond to three readers. Each reader has different requirements. The page has to satisfy all three at once, which is harder than it sounds.

Reader one: the human buyer. The buyer wants to know whether the product is right for her, what other people like her thought of it, whether the brand stands behind it, and how to think about the small worries she brought with her ("does it sting," "will it fit," "does it run small"). The buyer reads slowly when she is reading, scans when she is not, and decides on a mix of the page's signal and her own intuition.

Reader two: the AI crawler. The crawler arrives via GET request, in many cases without executing JavaScript, and reads the HTML of the page as a static document. The crawler is looking for passages that contain language relevant to a future buyer query. It is more likely to extract and cite passages that have first-person verbs, dated context, named authors, and topical specificity. We have walked through this in what chatgpt reaches for.

Reader three: the search engine indexer. Google's crawler, separately from the AI engine crawlers, indexes the page for traditional ranking. Its requirements overlap with both: it weights authentic content, recency, structured markup, and signs of a page that is being maintained rather than abandoned.

Three readers, one document. The 2018 product page was built for the first reader only. The 2026 product page has to be built for all three. The corpus, hosted properly on the page, is the part of the architecture that satisfies all three at once.

A product page architected as a corpus host is not a wholly different shape from a 2018 product page. It is the same shape with the proportions inverted. Three structural changes are load-bearing.

Change one: the brand-written body copy shrinks. The 800-word brand description, the ingredient deep-page (avoiding the banned phrase), the lifestyle aspirational paragraph: these compress into one or two short paragraphs at the top. The brand's job in the body copy is to frame, not to fill. Two sentences of "what the product is, who made it, what it does" is more useful than a thousand words of marketing prose. The space the marketing copy used to occupy is given to the corpus.

Change two. The customer corpus rises. Reviews, organized not as a widget with a star aggregate but as a sequence of bylined paragraphs with timestamps. Buyer questions, with brand replies, in a Q&A format that is closer to a customer service log than to FAQPage schema. Quoted passages from buyers, lifted into the body of the page under topical headings ("On use with retinol," "On sensitive skin," "On the smell"), with attribution and date and link to the original review. The corpus is not a separate block to be scrolled past; it is interleaved into the page's editorial structure.

Change three. The brand's voice moves. From the body copy of the page to the editorial framings between corpus excerpts. Section headers written in the brand's voice. Brief notes between groups of customer quotes ("Several buyers in their thirties wrote about the way this serum behaves in humid climates. Here are three of those notes"). Replies from named humans at the brand to specific reviews. The brand becomes the editor of the page, not the author. The author is the customer.

In reviews are language not inventory, we argued the linguistic version of this shift. The reviews are sentences, not units of inventory; they require editorial care, not counting. The corpus-hosting product page is the architectural form of that editorial posture. The page is built as if it were a magazine feature about the product, with the customer's writing as the substance and the brand's writing as the apparatus around it.

The job of the brand on a corpus-hosting page is not unlike the job of a magazine editor. Three things an editor does that the brand has to start doing.

An editor reads everything before publishing anything. The brand reads the full review corpus, the questions, the support tickets, the social mentions, the Reddit threads. Reading is the work. Without it, the framings between the customer excerpts are guesswork. The studio practice we have written about in what an editor would do with your corpus is, on the product page, a permanent ongoing practice rather than a one-time exercise.

An editor curates. Not every review goes on the page. Not every buyer question is worth a heading. The corpus is large; the page has limited real estate. The editor selects what is most representative, most useful, most surprising. The selections are themselves an editorial position. A page that shows only five-star reviews has taken a position the buyer can read; a page that shows the most thoughtful mixed reviews has taken a different position. The first is treated by the buyer as marketing. The second is treated as honest.

An editor attributes. Every customer quote on the page has a name, a date, a verified-purchase indicator where applicable, a link to the full review on the reviews page. The attribution is part of the writing, not metadata to be hidden behind an "i" icon. The byline is what makes the testimony testimony rather than blurb.

The composite shift is not subtle. The brand's marketing team, in the 2018 model, wrote the page. In the 2026 model, the brand's marketing team edits the page. The page is written by the customers. The team's value-add is judgement, framing, attribution, and the discipline of reading the corpus weekly.

This is uncomfortable for marketing teams whose value historically came from being able to write a clean piece of copy. It is the right discomfort. The clean piece of copy is no longer the highest-citation document for the highest-intent query. The carefully curated customer corpus is.

The structured-data layer of the page has to follow the content shift. Three schema changes that fall out of the corpus architecture.

The Product schema stays, but its content shrinks. The aggregateRating is still there. The brand, the manufacturer, the price, the SKU are still there. The description field, instead of holding 200 words of marketing copy, holds two sentences framing what the product is. The schema becomes thinner because the body of the page is doing the work the schema used to do.

The Review schema does heavy lifting. Each customer review on the page has its own Review markup, with reviewBody, author, datePublished, and reviewRating. The review is in the HTML of the page, in the schema, and addressable by a fragment anchor. The engine that wants to cite a specific paragraph can find it, identify its author, and verify its date.

The QAPage or DiscussionForumPosting schema (depending on how the page is structured) covers the buyer-question section. We have argued elsewhere that FAQPage schema is the wrong tag for question-and-answer content (FAQPage is for the brand's own FAQs, not for buyer-asked questions); the DiscussionForumPosting type, introduced by Google in 2024, is closer to right for the corpus shift's question-and-answer content.

The corpus is structured all the way down. The schema is the machine-readable version of the editorial structure of the page. The two are not separate concerns. They are the same concern, addressed at two layers.

A corpus-hosting product page does not all fit on one screen. It does not all fit on one URL, in many cases. The architectural arrangement we have argued for elsewhere (a dedicated reviews URL for high-volume products) is part of the corpus pattern. The product page is the entry point and the brand-edited summary. The reviews page is the full corpus. The Q&A page (sometimes a third URL) is the buyer-question portion.

Three URLs per high-volume product, where 2018 had one. Each URL has its own canonical, its own topical focus, its own schema, its own citation profile. The link graph between them is hierarchical: the product page is the parent; the reviews and Q&A pages are children. Sitemap and breadcrumb structure make the relationship machine-legible.

For products below the high-volume threshold, the corpus lives on the product page itself. The page is long, scrollable, full of bylines and dates. The buyer scrolls. The crawler reads. The brand does not pretend that the page is one screen tall.

In the half life of a product page, we wrote about the page that compounds in the language of its own customers versus the page that decays. The corpus-hosting architecture is the structural form of compounding. The page does not need to be rewritten quarterly; it grows on its own, in the contributions of the customers, framed by the brand's continuous editorial work. The page in year three is denser, more cited, more credible than the page in year one. The brand's role in year three is curation, not creation.

The brand's job on the 2026 product page is increasingly editorial: curate, frame, attribute. The promotional job is still there, but it is now a small percentage of the page's text. The customers are doing the writing. The brand is doing the editing.

The team that runs a corpus-hosting page is not the same team that ran the 2018 product page.

The 2018 team was a copywriter and a designer. The copywriter wrote the body copy quarterly. The designer arranged it. A product manager owned the conversion rate. The reviews, where they appeared, were owned by a customer-support adjacent function. The page was a sales asset and was managed as one.

The 2026 team has at least three new functions, often inside one or two people. An editor: reads the corpus, selects passages, writes the framings, writes the editorial replies. A schema engineer: maintains the structured-data layer, watches for crawl errors, updates as the engines evolve. A response writer: writes the named-human replies to customer reviews, in the brand's voice, on a weekly cadence. The conversion rate matters, but so does the citation rate. So does the freshness of the corpus. So does the response coverage.

This is the shift we have been arguing in the dashboard is not the work. The dashboard's job in the 2018 model was to display the conversion rate of the page. The dashboard's job in the 2026 model is to display the throughput of the editorial work: reviews read, passages selected, replies written, schema updated. The dashboard counts what the editor did. The editor does the work. The page is the publication.

Three steps, in order, for a brand considering the corpus shift on its top-volume SKU.

Step one is to read the corpus. Read every review of the top SKU from the last twelve months. Read the buyer questions. Read the Reddit threads. Read the press mentions. The exercise takes a day for a small store, a week for a larger one. It is the work. The downstream architectural decisions depend on knowing what the corpus contains.

Step two. Re-architect the page. Compress the brand-written body copy to two paragraphs. Surface six to ten customer passages as inline quotes with attribution. Open the question-and-answer section above the fold. Add named-human replies to the most-read reviews. Update the schema to match. The work is template-level on most stacks (Shopify, BigCommerce, headless React).

Step three is to build the workflow, since the corpus continues to grow and the page continues to need editing. A weekly cadence (read new reviews, update inline quotes, add new replies) is the minimum. A monthly cadence is too slow; the page goes stale. The work is small per week and substantial per quarter. The team change is permanent.

After those three, the work is at the level of what an editor would do with your corpus, applied as ongoing practice. The corpus is the page's substance. The brand's editorial discipline is what gives the substance its form.

The product page in 2018 was a sales letter. Its purpose was singular. Its writer was the brand. Its reader was the buyer. The architecture suited the era. The era is over.

The product page in 2026 is closer to a magazine feature, a community thread, and a public record at once. It hosts the corpus of writing the brand's customers have produced, framed by the editorial judgement of the brand, structured for the human reader and for the engines that will cite it. The brand still has a voice. The voice is now editorial, not promotional. The brand still owns the page. The page now belongs, in a different sense, to the customers who have written most of its text.

The shift is structural. The shift is uncomfortable for marketing teams that have spent twenty years writing the body copy. The shift is favourable for brands that recognise the writing has, in fact, already been done by their customers. The work is to read it, frame it, surface it, and let the page be what it now is. A document made of the testimony of the people who already bought, addressed to the people who are about to.