This article is writing itself

Watch the engine draft — then watch a rival model try to kill it.

One model writes, grounded in your first-party data. Four checks then fact-check it, score its novelty and hunt for AI tells — live, on the right — run by models that did not write it, each including a different vendor. What ships has cleared every bar, and is grounded in data a competitor cannot copy.

galley proofcctvline.co.uk · UK-CCTV · published

gauntlet · live
fact-checkqueued
GPT-5.2 · OpenAIdifferent vendor than the writer
novelty 0.71queued
Gemini 3.1 · Googlevs every top-10 competitor
e-e-a-tqueued
Haiku + GPT-5.2panel · Anthropic + OpenAI
ai-tellqueued
Haiku + Gemini 2.5para 06 reads too uniform

254,943 words published. 21.1% of drafts cleared every check with no revision at all. 135 of 275 never went live at all.

generate one for your site~4 min · ~$0.92 →

the receipt

Every figure is a live query, not a slogan.

as of 2026-09-11
254,943
words published, live on client sites
136 articles across 5 active sites
[src]
SELECT sum((payload->>'word_count')::int)
FROM published_articles WHERE status='published';
9,234
changes shipped to client sites
8,259 meta patches · 839 rewrites · 136 articles
[src]
-- passed seo_patch_jobs + content_rewrite_jobs + published_articles
21.1%
of drafts clear every check on the first attempt
50.9% pass within the three revisions allowed · the other 135 of 275 never published
[src]
SELECT count(*) FILTER (WHERE status='passed' AND attempts = 1)::float
       / count(*) FROM content_jobs
WHERE status IN ('passed','failed') AND cost_usd > 0;
$2.05
median cost per finished run, top tier
$0.41 at the cheapest tier · our cost to produce, not the price you pay
[src]
SELECT tier, percentile_cont(0.5) WITHIN GROUP (ORDER BY cost_usd)
FROM content_jobs
WHERE status IN ('passed','failed') AND cost_usd > 0
GROUP BY tier;
745
AI-assistant answers citing a client site
from 6,931 probes of ChatGPT and Perplexity, all sites, all time
[src]
SELECT count(*) FILTER (WHERE cited), count(*)
FROM ai_visibility;  -- no date or site filter

why a competitor can't copy it

Grounded in your data — and checked by a model that didn't write it.

grounded · first-party

…keep the horizontal offset to no more than 30° from the lane centreline[1].

[1]Zobook · Hikvision DS-2CD7A26G0/P datasheet — your product data, live stock and spec. Exists nowhere else online.
adversarial · rejected

In today's fast-paced world, it's important to note that leveraging cutting-edge ANPR technology can seamlessly revolutionise your security infrastructure and unlock powerful new capabilities.

⚑ ai-tell→ rewritten before it ever shipped

In practice, a correctly mounted DS-2CD7A26G0/P reads UK plates at 30–40 mph in the dark once the IR is calibrated — the difference between a barrier that opens and a queue that builds.

The writer never checks its own work. A different vendor's model does.

The four checks, in full

Every draft is judged by four checks in parallel, none of them run by the model that wrote it, and each one including a model from a different vendor. Any miss loops back to generation with specific fixes — up to three retries. Only content that clears every bar ships.

Fact-check
GPT · OpenAI

A different vendor than the writer flags fabrication and contradiction.

Novelty
Gemini · Google

Scored against every ranking competitor; must clear the threshold.

E-E-A-T
Cross-vendor

Experience, expertise, authority and trust signals are audited.

AI-tell
Cross-vendor

How human it reads — generic AI patterns are penalised, not shipped.

Pick your tier

Pay per article — no subscription, no per-seat. Simple per-article pricing by quality tier; the figures below are for a typical ~3,000-word article and scale with length.

Lite

≈ £0.40≈ $0.51/ typical article

Volume at the lowest cost

Writer
Claude Haiku 4.5
Validators
Cross-vendor — 1-2 models / check
Retries
3
  • Fast, economical writer model
  • Cross-vendor fact-check still enforced
  • Up to two validation retries
  • First-party RAG grounding
  • AI-designed covers (Gemini imagery)
Choose Lite
Most popular

Standard

≈ £1.60≈ $2.03/ typical article

The smart default — balanced

Writer
Claude Sonnet 5
Validators
Cross-vendor consensus — 2 / check
Retries
3
  • Section-parallel Sonnet 5 writer
  • Full four-check cross-vendor gauntlet
  • Up to three retries with targeted fixes
  • Generated hero + designed cover imagery
  • Auto internal-linking + JSON-LD
Choose Standard

Pro

≈ £4≈ $5.08/ typical article

Cornerstone-grade, golden bars

Writer
Claude Opus 5
Validators
Top cross-vendor panel — 2 / check
Retries
3
  • Claude Opus 5 writer for every piece
  • Strictest validators + up to four retries
  • Web-grounded Perplexity cross-check
  • Premium AI covers (Gemini 3 Pro Image)
  • Best for pillar / cornerstone content
Choose Pro

Premium

≈ £7≈ $8.89/ typical article

The flagship — our most capable writer

Writer
Claude Fable 5
Validators
Golden cross-vendor panel — 2 / check
Retries
3
  • Claude Fable 5 — Anthropic's most capable model
  • Opus 5-led E-E-A-T critique on every draft
  • Golden validation bars + Perplexity cross-check
  • Premium AI covers (Gemini 3 Pro Image)
  • For money pages where the article IS the product
Choose Premium

where it runs today

121 articles, 6 live sites, 2 languages

Every site below is operated by the same team as IT Master. We are saying so rather than presenting them as customer logos, because you could check it in a minute and because it is the honest reason to believe any of this: we run it on our own businesses, where a bad article costs us the sale. Click any of them — the articles are live, dated and signed.

  • CCTV Master35 articles

    UK CCTV retail

    publishing since June 2026

  • 1xai34 articles

    AI API gateway (Persian) · fa-IR

    publishing since July 2026

  • CCTV Line30 articles

    UK CCTV trade distribution

    publishing since June 2026

  • Euro CCTV15 articles

    UK CCTV trade supply

    publishing since June 2026

  • X-Ray Market4 articles

    Consumer software (Persian) · fa-IR

    publishing since July 2026

  • Mahsool Hunt3 articles

    Startup directory (Persian) · fa-IR

    publishing since July 2026

Counts are read from the publishing ledger, not typed into this page. The pass rates and costs behind them are on the AI Content Index, including how often a draft fails.

Read next

The whole argument, in the order it makes sense: what the thing is, how it is checked, who it is for, and what we can prove.

Understand the approach

Compare us honestly

See the evidence

Try it on your own site

questions people ask

Will Google penalise AI-generated content?

Google's stated position is that it rewards helpful, original content however it was produced, and acts against content made mainly to game rankings. That is the whole design here: nothing publishes until rival models — never the one that wrote it, always including a different vendor — have checked its facts, scored how much it adds over the pages already ranking, and hunted for AI tells. Drafts that fail are revised or dropped, not shipped.

If no human writes the articles, who checks them?

Four validators, none of them the model that did the writing, and at least one from a different vendor. They fact-check against sources, score novelty against the top-ranking competitors, critique experience and expertise signals, and look for AI tells. A failure goes back to the writer with specific fixes, up to three times; after that the draft is abandoned rather than published.

What does an article cost?

You prepay a credit balance and each article debits it — no subscription and no per-seat fee. The price depends on the quality tier and the finished length, and you see the per-article rate before anything runs. A draft that fails the checks is billed at a reduced rate, so a miss never costs full price.

What do I have to install on my site?

On WordPress, a plugin. On a Next.js site, the @itmaster/sdk package. On anything else, either pull finished articles from a read-only API with a per-site key, or receive them on a signed webhook. You connect once; after that articles arrive on their own.

How many articles will it publish, and how fast?

It publishes on a steady cadence rather than dumping a batch — the schedule comes from your budget, from a few articles a month up to several hundred. Drip-publishing is deliberate: a sudden flood of new pages is itself a pattern search engines treat as a signal.

Can I take something down after it publishes?

Yes. A takedown turns the article into a tombstone that connected sites remove on their next pull, and it propagates to anywhere it was reposted. You can also pause one site or switch publishing off entirely, at any time, without losing what is already live.