An AI SEO agency that runs itself, checked by rival models

The same loop runs for every site the engine manages, once per article.

• Discovery. It reads the queries your site already appears for in Search Console, pulls volume and People Also Ask data from DataForSEO, and picks topics you can realistically rank for.
• Research. It fetches the pages currently ranking for each topic, mines forum threads for the questions buyers actually ask, and pulls your first-party data — product specifications, datasheets, support Q&A — where it exists.
• Writing. A Claude model writes the outline, then the article, with a hero image, FAQ and JSON-LD schema.
• Checking. Gemini and GPT models fact-check the draft, score its novelty against the competitors it read, critique E-E-A-T and scan for AI tells. Failures go back for revision, up to three rounds.
• Publishing. Passing articles land on your site with internal links resolved, and IndexNow pings the search engines.
• Measurement. A nightly job reads Search Console, checks whether GPT and Perplexity cite your pages, and queues refreshes for content that is decaying.

Articles arrive as WordPress posts, through the @itmaster/sdk for Next.js, a REST pull feed or a signed webhook to any endpoint. Setup details are in the docs.

Autonomous does not mean unsupervised. Someone still owns what a site publishes. That seat is the editor-in-chief: one named person whose job is strategy rather than daily operations — which topic clusters to build, which claims the site must never make, what the brand sounds like, and when a page comes down. At IT Master that person is Rodd Azad, a software engineer and the founder.

Those decisions become configuration. A glossary of house terms, a list of facts the writer must never state, a brand-voice sample and a set of target queries feed every writing stage and are enforced deterministically, not left to a prompt. The seat writes the rules once; the pipeline applies them to every article.

There are also things software does not do well, and we say so. A human agency still wins on strategy that depends on knowing your market personally, on brand positioning, on digital PR and link outreach, and on the awkward conversation about what your site should stop doing. IT Master handles the volume layer — researched, checked, published articles — and leaves those judgement calls to people. The controls are post-hoc: take an article down, pause a site, or switch publishing off entirely.

A draft that is only ever read by the model that wrote it inherits that model's blind spots. IT Master separates the two jobs. Claude models write. Gemini and GPT models judge. The model that writes is never the model that checks.

Every draft runs four checks in parallel:

• Fact-check by a different vendor, against the research dossier and your first-party data.
• Novelty score against every competitor page the engine read — a draft that only restates the top ten does not pass.
• E-E-A-T critique for experience, expertise and specificity.
• AI-tell detection for the formulaic phrasing and hedged filler that make a page read as machine-written.

A failed check returns the draft to the writer with specific fixes, up to three rounds. A draft that still fails is rejected and is not charged.

The pass rates are measured rather than asserted. Across 186 Standard-tier runs, 21% of drafts cleared every check on the first attempt with no revision, and 51% got there within the three revisions allowed; the rest never published. Pro-tier judges are stricter: 25% cleared every check across a small sample of 16 runs.

Three ways to get search content produced, and what each is genuinely good and bad at.

A human agency. You pay a monthly retainer for a team's attention: strategy, briefs, writers, editing, reporting, and a person to ask. Strongest where the work depends on knowing your market personally — positioning, digital PR, link relationships, a migration, a category nobody has written about yet. Weakest on volume economics: the fee is the same in a busy month and a quiet one, so you carry the utilisation risk, and article throughput is capped by how many writers they can staff.

A content engine. You pay per finished article. Strongest where there is real search demand and you hold data nobody else has — a catalogue, datasheets, a support inbox — and where the bottleneck is throughput rather than insight. Weakest where the value is one person's voice, where there is no first-party data to ground anything in, or where every page needs a qualified human to sign off a regulated claim.

In-house. A salaried writer or SEO lead. Strongest on continuity, brand knowledge and being in the room for product decisions. Weakest on cost per article at low volume, and on the breadth of technical work one person can cover.

The practical test, in three questions:

• Do you know what to write? If not, buy strategy — from an agency or a consultant — before buying production of any kind.
• Do you have something to say that others do not? Specifications, measurements, support answers, install notes. If yes, an engine has raw material and will out-produce a retainer on cost per page. If no, fix that first; nothing writes originality out of nothing.
• Is anything you publish now getting indexed? If Google is not crawling what you have, more pages is the wrong purchase this quarter.

These are not exclusive. The common arrangement we see is an agency or consultant owning strategy and links, an engine owning production and refresh, and one named person inside the business owning what may and may not be said.

A human agency is usually sold as a retainer: a fixed monthly fee for a team's time, with articles as one output among reports, calls and audits. The fee is the same in a busy month and a quiet one, so the buyer carries the utilisation risk.

IT Master inverts that. You pay per published article from a prepaid balance. There is no subscription and no minimum term. If the gauntlet rejects a draft you are not charged for it, because you did not receive a finished article. Four tiers — Lite, Standard, Pro and Premium — set how much research and how strict a judging panel each article gets, so a supporting glossary entry and a cornerstone guide do not cost the same.

For transparency, here is what the engine costs us to run. Median generation cost per article: Lite $0.41, Standard $0.85, Pro $1.45 (our cost, not the customer price). Customer prices and what each tier includes are on the pricing page.

What a per-article price does not buy is a strategist's afternoon. If you need someone in your planning meeting, a retainer buys that and this does not.

It suits a site that has something to say and nobody to say it at volume: an ecommerce catalogue of products and datasheets that nobody has turned into buying guides, a trades or B2B business whose support inbox answers the same questions over and over, a SaaS with a documentation set and an empty blog, an agency that wants to sell content at a margin without hiring writers. Live tenants today include UK CCTV trade and retail sites.

It suits less well:

• A brand whose entire value is one author's personal voice.
• A site with no first-party data and no search demand. The engine grounds articles in what you know and what people search for; with neither it has nothing to work with.
• A business that needs links, PR or a rebrand before it needs content. Do those first.
• Anyone expecting guaranteed rankings. Nobody honest sells those.

If you are unsure which case you are, the free site check reads your site and tells you.

A section that costs us sales, kept because the alternative is worse.

• We do not guarantee rankings. Position depends on competitors, links, the site's history and ranking systems that change without notice. No supplier controls those, and a contract that promises them is either meaningless or about to be renegotiated.
• We do not guarantee citations. A generative engine can paraphrase your page without naming it, credit a competitor who quoted you, or answer with no sources at all. Its output varies between runs and between readers. We can measure the rate honestly, on a fixed schedule; we cannot fix it.
• We do not claim the writing is undetectable as AI. Detectors are unreliable in both directions, and building a product on beating one would be building on sand. What we do claim is checkable: the model that writes is never a model that checks, every check includes a model from a different vendor, and the pass rates are published.
• We do not promise a timeline. Crawling, indexing, ranking and citation each move on their own schedule. What we can commit to is cadence and measurement, not a date on which a number changes.
• We do not speak in your name without asking. Anything posted under your identity on an external platform is opt-in per site and per platform.

This is also published on our about page, and it is deliberately harder to walk back there than in a sales conversation.

Most of these can be answered in a sentence. Vague answers are the signal.

• Which model writes, and which model checks? If it is the same model, or the same vendor, the checker inherits the writer's blind spots. Ours writes with Claude and checks with panels that always include a different vendor.
• What is the first-pass rate, and where is it published? Anyone producing content at volume knows this number. Ours is on the AI Content Index, read live from the job ledger, including the tiers where it is unflattering.
• What happens to a draft that fails? Ask how many revision rounds, and what happens after the last one. Ours gets three rounds and is then abandoned, not published.
• What am I charged for a failure? If a rejected draft costs the same as a published one, nothing has an incentive to reject anything.
• What is the content grounded in? If the answer is "the top ten results", the output is the average of the top ten results.
• Who is named on the page? A byline with no accountable person behind it is the thing E-E-A-T guidance exists to catch.
• How is publishing gated, and how do I take something down? Ask for the mechanism, not the promise. Ours is per-site pause, a global off switch, and a takedown that pushes a tombstone connected sites remove.
• What do you measure, how often, and does the probe use live search? A visibility score from a model answering without searching measures brand fame, not your pages.
• What share of what you publish for me gets indexed? This is the number that says whether publishing became visibility, and it is one query.
• What do you refuse to promise? Anyone guaranteeing rankings or AI citations is describing something they do not control.

Ask the same ten of us. The answers are on how it works, pricing and the two research pages, not held back for a call.

The fastest way to find out whether this works for your site is to let the engine read it. The free site check reads your pages and reports where the gaps are, at no cost.

If you want to see where you stand in AI search first, the AI visibility checker and the AEO checker look at whether assistants cite you and how quotable your pages are, and the AI crawler check confirms that AI crawlers can reach those pages at all.

When you are ready, connect a site. WordPress users install the plugin and paste one key; articles arrive as posts with the Yoast and Rank Math fields filled. Next.js sites use the SDK. Anything else takes the REST feed or the signed webhook, we set up Shopify and Webflow on request, and Wix or Squarespace sites get a copy-paste queue instead. Top up a balance, choose a tier, and the first researched, checked article is on its way. You pay per published article — see pricing.

The external references behind the claims on this page, all read on 4 September 2026, plus our own published numbers.

• Google Search Central, "Spam policies for Google web search" — scaled content abuse is defined by volume without value, not by authorship: developers.google.com/search/docs/essentials/spam-policies
• Google Search Central, "Creating helpful, reliable, people-first content" — the E-E-A-T self-assessment, including "who, how and why" and the expectations on a named author: developers.google.com/search/docs/fundamentals/creating-helpful-content
• Google Search Central, "AI features and your website" — that no separate optimisation is required to appear in AI Overviews or AI Mode: developers.google.com/search/docs/appearance/ai-features
• IndexNow protocol documentation — the discovery push the engine uses after publishing: www.indexnow.org/documentation
• Our own figures: first-pass rates, drafts rejected and median cost per article by tier on the AI Content Index, read live from the engine's job ledger, and the weekly AI Citation Index.