REFORGED

REFORGE 1.0 — AI pages marketing · GEO · SEO · AI answers

REFORGE connects one page of your site to continuous revision: for search, for AI answers, for purchase.

• For whom — brands with a catalogue and facts that can be checked: a product card, an article, a collection.
• What you get — a page whose facts, text and structure are revised, dated and measured, with the rollback kept.
• How it runs — through the coding agent on your subscription; the rules arrive as versioned packs; the checks run against our registry.
• What it costs — paid per connected page: a one-time connection, then a monthly fee for support.
• How to start — Join the waiting list; the first page is chosen from your data, not from a template.

Generative engine optimization (GEO) is revising pages so AI answers — ChatGPT, Perplexity, AI Overviews — cite them while search ranks them. Elsewhere this work is sold as AI SEO or GEO services. REFORGE 1.0 is AI pages marketing as code. REFORGE grew on STRUGA's storefronts, runs through your own agent, with rules by version and checks by our registry on our side on your subscription. The unit is one page or a bundle of pages, connected for as long as you keep it connected. Rules and checks stay with us; only revision records leave your machine. There are no tricks “for AI” here: checked facts, a form an engine can lift whole, and a route to purchase. Version 1.0 now, 2.0 next.

What does the REFORGED mark mean on this storefront?

REFORGED is the mark of a page connected to REFORGE, not a grade of quality. On strugadesign.com the mark stands by the main heading of a connected page and leads here. It appears as pages are released — the Brutalism V.3 ring card and one journal article carry it first. The mark is never next to the price or the buy button: the route to the object stays first. STRUGA places the mark and withdraws it itself. The mark pulses: the page is connected to continuous revision, not stamped once.

For a buyer the mark means four things:

• Weight, metal, price and lead time were read back from the catalogue on the revision date.
• The text stays in STRUGA's voice: no borrowed phrasing, no promises, no superlatives.
• This language version was adapted to its own market by tools — demand, results and rivals in this language. Not a translation.
• The revision date is the date of a real change; a date refreshed for appearance earns no mark.

Every connected page will also get a passport on this same domain: revision date, what was checked, version fingerprint, measurement window.

What is GEO, and how does it differ from SEO, AEO and LLM SEO?

GEO, generative engine optimization, is work on a page so that an AI answer cites it. SEO is work on the same page so that search ranks it: one page, one job, not two sites. GEO here is not geographic or local search — the abbreviation is shared, the work is not.

The neighbouring terms name one job from different sides:

• GEO — a generated answer can take a sentence from the page and name the source.
• AEO, answer engine optimization — the same work named from the answer's side: a direct answer to a stated question.
• LLM SEO, or AI search optimization — the same work named from the model's side.
• AI visibility — how often answers name the brand or list its domain.
• AI citation — the domain listed as a source under an answer.

GEO does not replace SEO. Google's AI Overviews and AI Mode stand on the same search base. No separate markup or file is required (Google Search Central, checked August 2026). Across 45 empirical studies reviewed on arXiv in July 2026, topic relevance and position in context held up; formatting tricks did not.

What changed in search by 2026 — and is SEO dead?

SEO is not dead: AI answers are built on the search index, and among cited pages ordinary rank remains the entry.

In Pew's 2025 sample of 68 879 Google searches, a link inside an AI summary was clicked in 1% of visits. Ordinary results were clicked in 8% of visits with a summary, against 15% without: visibility is not a visit.

In AirOps and Kevin Indig's 2026 sample, the first result for an engine-generated sub-query took 58% of citations, four times the tenth. Retold from a webinar; raw data unpublished. In seoClarity's sample of 362 000 US queries, 56% of AI Overview citations came from the top-20, 44% from outside. Rank 11–20 is not a verdict.

Scrunch, June 2026: after a brand was recommended in an AI answer, branded search rose from 3.28% to 9.24% in a week. An association in a panel of undisclosed size, not a cause.

So a page has to be ready for the answer, not only for the ranking.

Which AI answers is a page prepared for?

REFORGE prepares a page for the AI systems that search the web and cite their sources. On this storefront they are:

• ChatGPT (OpenAI)
• Perplexity
• Google AI Overviews
• Google AI Mode
• Gemini (Google)
• Claude (Anthropic)
• Grok (xAI)
• Meta AI (Llama)
• Mistral Le Chat
• Microsoft Copilot (Bing)
• DeepSeek
• Qwen (Alibaba)

The page is written for all of them at once: a sentence that answers on its own serves every engine. Measurement is narrower than preparation. It runs on the engines the client names, in a window set before the change. An assistant without web search answers from training and cannot cite a fresh page; it is not in the count. The list is reviewed with the rules: an engine that gains search and citation enters it.

What does REFORGE do to a page?

An audit with execution, not a report. These are the stages, not the recipe: what is done and why, without the thresholds and rules that decide each step. Every stage leaves a trace.

1. The unit and a snapshot before the change — the page named exactly, its state kept whole: the right to roll back.
2. Demand and citation research by tools — what people ask, how search and AI answers respond, what the answers lack.
3. Fact-check against primary sources — every material fact gets a source and a check date; old text is read as a stranger's.
4. Text in the brand's voice — the direct answer first, then evidence, comparison and the next useful step; written whole, not assembled.
5. Structure and markup — self-contained sentences, price and facts in visible HTML, questions as headings, a revision date, a route to the object.
6. Adaptation of each language version to its market — demand, results and rivals read in that language. The live page is read back, the reader's path from search to the goal checked; no undeclared change.
7. Measurement and outcome — baseline numbers before the change, a window set in days, one of four outcomes: keep, revise, roll back, inconclusive.
8. Connection kept — the weekly review of rules reaches the page as corrections; the page's history grows with each one.

How does a page get into ChatGPT, Perplexity or AI Overviews answers?

The question is about being cited by an answer, not about citing ChatGPT in a paper. An answer engine takes from a page the sentences that answer on their own; a sentence that needs its neighbours loses.

Which pages do AI answers cite?

Pages that match the topic and name what they describe. In a passage-level study by Advanced Web Ranking, 112 passages from one day were read one by one — a small, directional sample. The entity was named in the sentence in 96% of quoted passages against 82% of absorbed ones. Structure differed little between the groups: 76% against 71%. So REFORGE checks the topic and the kind of answer the question wants — a definition, a comparison, an instruction — before any markup. A page that does not fit the question is left alone. REFORGE writes sentences that carry a fact whole — subject named, number inside, date visible — and puts the direct answer first.

Does a fresh date help a page get cited?

Among cited pages, yes — read from the update date, not the publication date. In Seer's sample of 7 683 pages cited by ChatGPT, Gemini and Perplexity, 75% were updated within a year, 88% within two. On the 4 124 pages carrying both dates, 72% were updated within a year while 42% were first published within it. In the Advanced Web Ranking sample, a visible fresh date stood on 80% of quoted passages against 53% of absorbed ones. Freshness in these samples is a correlation among cited pages, not a lever measured against a control. That is why REFORGE revises existing pages before adding new ones. The engine writes the revision date on the day of the change and reads it back on the live page. A page whose facts have not changed keeps its old date. A connected page gets a new date each time a real correction reaches it. A date refreshed for appearance earns no mark: on a REFORGED page the date is the date of a real change.

What is connected: a page or a bundle?

The unit of connection is one page, not a site. A page is a product card, an article, an Atlas page or a collection. It can also be a new page, built in REFORGE from our own blocks. One page can stay connected for as long as its owner wants: revised, measured and corrected on a permanent basis. Where the answer needs more than one page, the unit is a bundle. Example: an article, the cards it leads to and the route between them. A bundle is connected as one unit, revised as one and measured as one. Which pages form a bundle is decided from your data at the first review, not from a template. Each language version of a connected page is part of the same unit, adapted to its own market.

How does a connected page stay current?

A REFORGED page is not stamped once. The rules of REFORGE rest on a tracked body of research, and each claim in it carries a primary source and an expiry date. Every week the rules are reviewed against fresh research on AI answers, search, citation and promotion. An expired claim leaves; a new one enters. What changes in the rules reaches every connected page as a correction — applied on the page, dated, with the previous state kept. REFORGE has versions, 1.0 now and 2.0 next, and a connected page moves with them without a new order. That is what the mark by the heading means, and why it pulses. The page is connected to continuous revision and to the research frontier. REFORGE is the ongoing automation of one page, not a one-time treatment.

How does REFORGE work on your side — and how much does AI SEO cost?

REFORGE is not an agency and not a service you hand your site to. REFORGE is a runner for the agent you already have; the rules arrive as versioned packs, and fact checks run against our registry. The agents it runs through: Claude Code, Codex (OpenAI), Gemini CLI, Grok, Cursor, GitHub Copilot, Mistral Vibe, OpenCode. Any agent that reads files and runs commands on your subscription will do. Your agent executes the code. The REFORGE engine writes the change to your site on your word, with a dry run first and a rollback kept. The rules, checks and corrections the agent follows are ours. The work divides like this:

The cost has two parts. The heavy part — your agent's tokens and time — runs on the subscription you already pay for. REFORGE itself is paid per connected page: a one-time fee for the connection, then a monthly fee while the page stays connected. The monthly fee covers the weekly review of rules, the corrections that reach the page, the measurement and the history. There is no fee for volume of text and no turnkey service. Nothing is bought in advance: the agent, the subscription and the site are already yours. The figures are named in reply to the waiting list, before any payment. Three things are needed to start: an agent with a subscription, write access to your site, and a catalogue with checkable facts.

Which stores and sites is REFORGE for?

REFORGE is for brands with a catalogue and a history: objects with facts that can be checked, text that can be backed. A site that exists starts from a snapshot of its pages as they are, and REFORGE revises them. A site that does not exist yet starts from demand research and structure; the first page goes live already checked.

Each language version of a page is adapted to its own market, not translated. The tools read demand, results and rivals in that language, and the page answers what that market asks. Our own case: STRUGA runs two storefronts, strugadesign.com and strugadesign.ru, with two different checkouts and hundreds of live pages. REFORGE grew there, on two markets at once, and each version of a page was written for its own market. A second language is a second adaptation of the same page, inside the same unit.

REFORGE is not for text without sources or results without measurement. REFORGE publishes no fact without a primary source and names no effect without a control. A site that wants volume of text rather than checked pages is better served elsewhere.

What do you receive — what is a revision record?

The unit of work is one bounded package: a named page, named places, evidence before and after. A revision record holds:

• A unit named exactly — and it does not get lost when the address changes.
• What changes and what stays — heading, text, markup, meta, named in advance.
• A record before and after — snapshots of every changed place, version fingerprints.
• A ledger of facts — sources, check dates, what was left untouched and why.
• A measurement window and an outcome — baseline numbers before, the outcome in one of four words.
• The right to roll back — the snapshot is kept; the same machine returns it.
• Price and facts in visible HTML — Growth Memo tested 1 500 agent runs. After an access error, agents took the price from third-party sites in 77% of runs, against 17% without. Bing advises key facts in visible HTML and short self-contained answers — eligibility, not a promise.

In place of case-study percentages: revision records — date, what was checked, status.

What accumulates with the brand — history and analytics?

Every connected page keeps its own history. What was changed and why, what the measurement said, what was kept and what was rolled back. The history belongs to the brand: it is the record of decisions about that page. Over time the history answers questions no single revision can. Which kind of change worked on which kind of page, and which did not. From that record the next versions of REFORGE will grow a platform for decisions. Tests will run on every connected page, continuously and without a hand on them. A page that does not work will be withdrawn; a page that is missing will be created. None of that is promised as growth. What accumulates is evidence; what it gives is a basis for a decision.

How is the result measured, and what counts as AI visibility?

AI visibility is how often an answer names the brand or lists its domain. REFORGE counts the two separately. In Semrush's sample of 3 981 domain appearances, about 62% of citations were “ghost”: the domain listed, the brand not named. REFORGE is not a visibility tracker and does not replace one. The revision record stores the before, the after and the window.

The value is in conversion, not in share. In Orbit Media's sample of 97 B2B sites and 28.9 million sessions, AI referrals were about 0.5% of sessions. Their high-intent conversion ran from no different to three times direct or organic — a range in one sample, not a promise.

Our own measurement on STRUGA's storefronts ran by visits, with a control group. About one revised page in five carried enough visits to measure. The median change was zero, no page dropped, two rose above the noise floor. An effect beyond noise has not been shown; the re-measurement is set for 24 September 2026.

What does REFORGE not promise — and what does it not do?

Growth in traffic, citations or sales is not promised. Always there: the process, the evidence and the rollback — every change documented, every fact sourced. An outcome of inconclusive is also a delivery. Early signals are read in weeks, orders in months. REFORGE names the figure of an effect only after the window.

No banners, buttons or foreign links on money pages: the design stays the client's. On a client's site the REFORGED mark is a thin frame by the heading, shown at the owner's decision. It pulses while the page stays connected.

What REFORGE does not do:

• No llms.txt and no “AI markup”. Google states no separate markup or file is needed for AI features; in SE Ranking's data llms.txt predicted nothing.
• No schema added for citations' sake. Ahrefs tracked 1 885 pages adding schema against 4 000 controls: AI Mode +2.4%, ChatGPT +2.2%, indistinguishable from zero; AI Overviews −4.6%.
• No chunking “for AI”, no AI rewrite, no ratings without real reviews, no date cosmetics, no link or mention campaigns.

How do I join the REFORGE waiting list?

Every new site is connected separately, and the first page is reviewed together. Join the waiting list in the form below; no technical language is needed — describe the site in your own words. The form asks for the site address — or “no site yet” — the platform, your agent and what to revise. Then languages and markets, a name, an email and a messenger. Joining the list is not an order and creates no obligation on either side. A reply comes by email, with questions about the site and the composition of the first page or bundle. Then, in order: the terms and the price per page before any payment; the first page chosen from your data; the connection. Until then nothing is charged and nothing is installed. The list is read in order.

A second door — for those who work with an agent. Paste the command into Claude Code, Codex, Gemini CLI, Grok, Cursor, GitHub Copilot, Mistral Vibe or OpenCode: the agent reads the instructions, asks the questions one block at a time and submits the answers itself. It installs nothing and does not touch your site.

Read https://raw.githubusercontent.com/dmitry-strugovshchikov/reforge-connect/main/CONNECT.md and run the REFORGE connection: ask me the questions one block at a time, then submit the answers as the instructions say.

The waiting-list form is below on this page.

STRUGA objects

REFORGE grew on these pages. Rings, the CODEX and RITUAL worlds, the page on Living Silver: each has been through the routine. The mark by the heading appears on them as pages are released. Weight, metal and lead time on those pages were read back from the catalogue. Price and lead time stay on the object's own page, next to the buy button. For the international market, STRUGA objects are made in Bali. Every object is made to order. Silver takes about two weeks, silver with a stone about three; carbon is an individual commission of about four weeks. STRUGA ships worldwide.

If the form you need is not in the catalogue, there is custom order. An object is built for one person, from sketch to finished piece, with the same facts read back before it is offered.