A category grew up around this fast. Wix, Squarespace, Webflow and HubSpot all launched products in the last year that tell a customer whether ChatGPT mentions them. HubSpot sells that as a standalone at $50 a month. Profound raised $96 million in February at a billion-dollar valuation doing it for enterprises.

Agencies sell the work itself. Published rates for a small business start around $1,500 a month and run to $5,000.

Nearly all of that is measurement and content. The dashboards report a score. The retainers cover monitoring, writing and getting other websites to mention the brand. Very little of it changes what the site actually sends to a machine, and the part that does is one line item.

Structured data is the one people mean. It is a block of machine-readable facts in the page saying what this is, who published it, who wrote it and when.

Ahrefs tracked 1,885 pages that added it between August 2025 and March 2026, against 4,000 comparable pages that did not. Citations moved a couple of percent on ChatGPT and on Google AI Mode, inside the range of noise. On AI Overviews they fell 4.6 percent. No uplift anywhere.

Google says the same thing in its own documentation. "Structured data isn't required for generative AI search, and there's no special schema.org markup." On semantic HTML it says to focus on human readability and not worry about perfect code.

Some of it is not just unproven, it is retired. Google stopped showing FAQ rich results on 7 May 2026. The how-to format went earlier. FAQ markup is still one of the most widely recommended items in this category.

And llms.txt, a plain text index of a site written for language models, is fetched by almost nothing. Ahrefs looked at 137,210 domains and found 97 percent of the files that exist received zero requests in May 2026. No AI crawler goes looking for one that is not there.

One factor sits above everything else, and it is whether the words are in the page when it arrives.

Vercel measured this at network level across their own traffic. None of the major AI crawlers run JavaScript. Not OpenAI's, not Anthropic's, not Perplexity's. They take the HTML the server sends and stop. Only Google renders.

A site that assembles its content in the browser sends those crawlers an empty shell. Whatever it says, they never see it.

What the page says carries the rest of it. The one peer-reviewed study in this area, presented at KDD in 2024, tested nine ways of editing a page against 10,000 queries. Adding statistics moved citation rates 41 percent. Adding quotations 40 percent. Citing sources 27 percent. Keyword stuffing made it worse. Nothing that worked was markup. All of it was the content being specific and sourced.

Every page is rendered on the server. The text, the headings, the tables and the lists are all in the first response. That is not a setting anybody turns on, it is how the platform builds a page, and it decides more here than anything else does.

Underneath that, every page now carries the machine-readable layer, generated from the content itself rather than written by hand. Who published it, who wrote it, when it was written, where it sits in the site. Every page, every content type, valid and current, with nothing to configure.

The blocks a page is built from emit the right HTML for what they contain. A comparison renders as a table with header cells. A sequence of steps renders as an ordered list. A quote renders as a quote with its attribution attached. A date renders as a machine-readable date. There are 105 block types and they were all reviewed for this.

The site publishes a sitemap with every page, every language and a last-modified date, and tells Bing and the other engines within seconds of anything changing. And it serves a correct llms.txt, for one reason. Google's Lighthouse audit now checks for it. Missing is fine and scores nothing. Broken fails. So the file should exist and be right, and nobody should ever pay for one.

On most platforms this is work somebody does by hand, once per content type, and again whenever the content model changes.

Webflow generates no structured data for its CMS items, so a developer writes a template per type with the fields interpolated in. Ghost emits two types and the theme layer makes adding more impractical. Squarespace needs the Business plan before a page can carry its own. Wix covers four kinds of page automatically and leaves the rest.

None of those are bad products. They are all built on the assumption that a site is a set of hand-made pages, so machine-readable output is something added per page. Giant Context stores content as structured blocks in a database, so the same output falls out of rendering. The marginal cost of getting it right for one more customer is zero, and there is no version of the site where somebody forgot.

The claim is narrower than the one the category is making. Not that this gets a business cited. Every page complete, correct and machine-readable, on every content type, without a plugin, a paid tier, a developer or a retainer.

The strongest measured signal for being cited by an AI engine is how often other websites mention the brand, and it happens entirely off the site. Reviews, press, forums, other people's writing. No rendering pipeline reaches any of it.

The other half is the writing. Specific claims, real numbers, sources named. The platform can make a page perfectly legible to a machine and the machine will still pass over it if the page says nothing.

So the division is that Giant Context handles the part that is mechanical and does it for everyone at once, and the part that is editorial stays editorial. That half is where the money in this category is currently going, and it is the part nobody has automated.