do you write the content or do we?

Either, and the retainer price follows the split. In every case we produce the page map, the briefs, the QA pass and the schema; the difference is who drafts. Teams with a writer and a subject expert usually draft themselves from our briefs, because the expert's name has to go on the page anyway. Teams without one hand it to us and sign off. What does not change is the QA before you see a draft and the byline requirement.

is this AI-generated content?

The pages are written to be quoted by AI engines, which is a different thing. Every page carries a named author with checkable credentials, every number carries a date and a source, and every page is fact-checked against your own figures before you see it. Whatever tools are used in drafting, the thing that makes a page quotable is verifiability, and no tool supplies that; a person with a name does.

how many pages a month?

As many as the page map says, which is usually fewer than a content agency would quote. In our dataset the sites missing from AI answers were not short of pages; 89% of them were short of a named author and 94% of structured data. A typical sprint ships two to four pages of the types the engines quote, plus refreshes of existing pages into answer-shaped blocks, and the panel two weeks later says whether any of them were cited.

AI content and SXO: the pages models quote

AI content and SXO is the stage where the site gets something worth quoting. SXO, search experience optimization, is the match between the page a buyer's query deserves and the page you have; AI content is that page written so a model can lift a claim from it without rewriting. The two are one job here, because the engines quote the same page types buyers convert on: a definition that answers the question in one paragraph, a comparison table, a price, a list, an FAQ whose markup matches the text word for word.

It is delivered inside the retainer, after the technical layer, because a quotable page on a site the crawlers cannot read is a page nobody will ever quote. And it is scoped by the audit's page map, which says which page types win your money queries today and which of them you do not have.

why volume is not the lever

Across the 100 audits in our published dataset (April to July 2026), the sites that were missing from AI answers were not short of content. They were short of anything a model could stand behind: 89% published anonymous content, with no about or team pages and no article authors; 94% had no structured data describing what the pages claimed. The clearest single example is the Manimama case, where the firm arrived with 1,004 blog posts and zero of them in Google US top-3, behind a competitor with 6.8 times less content (baseline 22 June 2026, the case). The work that moved the panel was not more posts. It was bylines and reviewer credentials on the money pages, a team page rebuilt with all eight lawyers and their bar numbers, and the blog merged into the product pages it had been cannibalising.

So the first thing this stage does is often subtraction. Then it publishes fewer pages, of the types the engines quote, with a named human behind each one.

what you get

content and SXO, delivered per sprint
no.	deliverable	what is inside
01	the page map	For each money query, the page type the engines quote, whether you have it, and the order to ship the rest in. Inherited from the strategy and refined every sprint.
02	page briefs	Per page: the question it answers, the 60-word definition it opens with, the table or list it needs, the claims and where each claim's source lives, the FAQ, and the schema it carries.
03	drafts with QA before you see them	Written, fact-checked against your own numbers, checked for consistency with every other page, and validated for markup parity. What reaches you is a page, not a first draft.
04	authors and reviewers	Named people with checkable credentials on every page, and Person schema that says so. If nobody on your side can sign a page, we say the page should not exist.
05	refreshes of what you already have	Existing pages restructured into answer-shaped blocks, dated, sourced and de-anonymised, which is usually faster than writing new ones.
06	the schema for each page	Article or BlogPosting, FAQPage with answers byte-identical to the visible text, Person, BreadcrumbList, wired to the organisation node.

the page types, and why these

The audit's SXO pass records what the engines cite for each category question, and the same five types recur: definitions, because a model needs a passage it can quote to say what something is; comparisons, because “who should I pick” is answered from pages that already compare; pricing, because a page with a number on it gets cited when a buyer asks what something costs (on our own weekly panel, 31 August 2026, ChatGPT cited our pricing page for exactly that question); lists, because the engines fill their shelves from them, which is also why we disclose that our own best-agencies page is one; and FAQs, because a question and a short answer is the shape the model is trying to produce.

What the engines do not quote is the thing most content programmes produce: undated, unsigned thought-leadership posts with no numbers in them. We do not write those.

how it is measured

Three ways, all from the same re-run panel. First, whether an engine names you on a category prompt at all. Second, the position it gives you. Third, and specific to this stage, whether the citation is your own page rather than a directory or a review site: in the Manimama case Perplexity cited the firm's own MiCA licensing page directly on the first run that named it, 18 July 2026, and in the INC4 case the engines went from quoting deleted pages to building the category answer out of the studio's own articles (the case). A page that is cited from your own domain is the outcome this stage is judged on; a page that ranks and is never quoted is a rankings win, and we report it as one.

Inside the health score, this stage moves the on-page and SXO category (20% of the score) and the two largest dimensions of the GEO block, citability (25%) and structural readability (20%) (how the score works).

what we will not write

Pages without a person. No named author with checkable credentials, no page.
Claims without a source. Every number carries a date and where it came from, on your site as on ours.
Markup the page does not support. No ratings without reviews, no FAQ answers that differ from the text. Markup that contradicts the page is worse than none.
Different content for crawlers and for users. It is fragile and it is the kind of trick that gets a domain discounted.
Volume for its own sake. If the page map says three pages, we ship three pages.

what it costs

Stage four is delivered inside the full-cycle retainer, from $3,000 per month (September 2026), with the final figure set after the audit. It is not sold as a separate content package, because a page written without the panel that says which questions it should answer, and without the re-measure that says whether it was quoted, is copywriting rather than GEO. Where your own team writes, the retainer covers the briefs, the QA and the markup and the number is lower; where we write end to end it is higher. The drivers are on the pricing page.

where it sits in the loop

After the strategy, which produces the page map, and after technical GEO, which makes the pages readable. In parallel with links and authority, because the pages written here are what the sources in that stage end up linking to and quoting. The whole loop is on the services page.

AI content and SXO