RESEARCH · METHODOLOGY

The three-piece stack, documented

RankedTag produces work in three stages: a human sets strategy and approves every brief, AI compresses research and the first draft, and a human rewrites, fact-checks and supplies the argument before anything publishes. No page ships unread, and the automation layer is handed over on client infrastructure when an engagement ends.

Last updated . Written and verified by Bhushan Raj Shakya, founder of RankedTag.

KEY FACTS

Stage one
Human strategy. SWOT, query selection, positioning, brief approval
Stage two
AI-compressed research and first draft
Stage three
Human rewrite, fact-check and argument
Automation
n8n workflows for triggers, enrichment, publishing and alerting
Rule that does not bend
Nothing publishes unread by a person
At handover
Workflows, prompts and documentation move to client infrastructure

The three stages, and what each is actually for

  1. Strategy, by a person

    Query selection, competitive reading and positioning are judgement work with no verifiable ground truth, which is exactly where a model is least reliable. A person decides what to target and why, and approves every brief before a word is written.

  2. Research and first draft, AI-compressed

    Reading fifty competitor pages, extracting what each covers, and producing a structured first pass is volume work with a checkable output. That is where AI earns its place: it turns a week into an afternoon without deciding anything.

  3. Rewrite and fact-check, by a person

    Every figure is traced to a primary source or cut. Every argument is supplied by a person. This stage is where a draft becomes something worth publishing, and it is not optional.

The division is not ideological. It follows from where errors are cheap to catch. A wrong competitor summary is caught by a human reading the draft, while a wrong strategic choice is not caught at all until a quarter has passed.

Why the rewrite stage is the one that cannot be skipped

The commercially relevant point is that unreviewed generated text tends to be unsourced, and unsourced text underperforms in exactly the surface it is usually produced for. The peer-reviewed GEO study found that adding quotations to a page increased its visibility in generated answers by roughly 41%, and that adding cited sources produced the largest measured gain at position five, around 115% Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024. Both of those are things a reviewing human adds and a first draft does not have.

There is also a policy dimension. Google’s spam policies address scaled content abuse, meaning content produced at scale primarily to manipulate rankings rather than to help people, regardless of whether it was produced by automation, by people, or by a combination Google Search Central, spam policies for Google web search. The distinction Google draws is purpose and quality, not authorship, which is precisely why the review stage exists rather than a disclosure badge.

What the automation layer does, and what it deliberately does not

  • Does: trigger on a published page, notify the right channel, enrich a new signup record, route an enquiry into the systems a client already uses.
  • Does: capture the attribution source recorded against a signup, which is where the first-party dataset on this site came from RankedTag, 17 production signup alerts from the Sendr engagement.
  • Does not: publish anything. A workflow can prepare a page and alert a person. It cannot be the person.
  • Does not: stay locked to RankedTag. Workflows, prompts and documentation are handed over on client infrastructure at the end of an engagement.

Where this method fails

It fails when the review stage is rushed, which is the same failure mode as any content operation and the reason intake is capped rather than scaled. It fails when a client expects volume as the deliverable: this stack produces fewer, denser pages, and a brief that asks for forty thin pages a month is a brief for a different supplier.

It also fails where there is no verifiable material to work with. A page about a product nobody has measured, in a category with no published research, cannot be made citable by better writing. In that situation the honest recommendation is to generate the evidence first, which is what the original dataset on this site is: a small, bounded thing that is genuinely ours rather than a borrowed statistic.

Frequently asked questions

Does AI write the published content?

AI writes a first draft and compresses the research behind it. A person rewrites it, checks every figure against a primary source, and supplies the argument. Nothing publishes unread.

Is AI-assisted content against Google’s guidelines?

No. Google’s spam policies target scaled content abuse, meaning content made primarily to manipulate rankings rather than to help people, whichever way it was produced Google Search Central, spam policies for Google web search.

What is n8n used for here?

Workflow automation: triggers, enrichment, publishing mechanics and alerts. It moves information between systems. It does not make editorial decisions.

Who owns the workflows at the end?

The client. They run on client infrastructure and are documented, so an engagement ending does not take the plumbing with it.

How many pages a month does this produce?

Fewer than a content mill and more than a traditional agency of the same size. The number varies with the brief, and volume is not the deliverable being sold.

Official resources

FOUR SLOTS A MONTH

That is how the work gets made.
See what it would make for you.

The 52-check founder review is run personally and comes back within 48 hours, with a real opinion even when the answer is no.