The content engine that reached page one in two months

Generating content with a language model is easy. You can produce a thousand words on almost any topic in a few seconds. Getting that content to rank on Google, bring in the right visitors and turn some of them into enquiries is a different problem entirely, and most AI content efforts fail at it.

I owned SEO and content strategy for a business and built an AI content and SEO automation platform to support it. It runs on an enterprise website I also built, one that scores 100 for performance and ranks well. Within 30 days it was bringing in qualified inbound leads, and within two months the business had first-page Google rankings for commercial keywords. This post explains how the platform works and, more importantly, why the parts around generation matter more than generation itself.

Why most AI content does not work

The failure mode of AI content is easy to recognise. Pages are long, fluent and empty. They repeat what every other page on the topic says, answer no question in particular, and read like they were written for a search engine rather than a person. Search engines have become good at recognising exactly this kind of page, and readers leave them immediately.

The problem is not that a model wrote the text. The problem is that nobody decided what the page was for, checked whether it was accurate, or held it to a standard before it went live. Generation is the easy step. Direction and discipline are the hard ones.

The shape of the platform

The platform treats content as a production workflow, not a series of prompts. It is running today on the same enterprise website where it proved itself. Each piece goes through the same stages:

1. Research. The engine either takes a topic it is given or finds one that is currently getting attention, gathers information on it, and runs keyword research to decide exactly which searches the article should target.
2. Brief. Every article starts from a brief: the search intent it serves, the audience, the key points it must cover and what the reader should do next.
3. Generate. A prompt-engineered generation step writes a draft from the brief.
4. Validate. Deterministic checks decide whether the draft is good enough to continue.
5. Assemble. The approved content is built into a properly formatted document with python-docx.
6. Publish. The document is published to Google Drive through the Google Workspace APIs.
7. Track. Everything is synchronised with Google Sheets, which acts as the content calendar and the record of what exists, where it is and what state it is in.

On the home page of this site there is a simulated replay of the effect: a rank tracker shows a page climbing from the third page of results onto the first over eight weeks, while articles go live and pass the quality gate. The replay is illustrative, but the outcome it represents, first-page rankings in two months, is real.

Research: trending topics and keyword data

The engine starts in one of two ways. Either it is handed a topic, or it looks for subjects that are currently getting attention in the business's field. Timely topics matter because searches for them grow quickly and few pages answer them well yet, so a good article can rank early.

Either way, the next step is research: gathering what is actually known about the topic, and running keyword research to find the specific phrases people search for, how competitive they are and what the pages already ranking for them look like. The article is then built to target those phrases deliberately, rather than hoping the right words appear.

Built for EEAT

Google's quality guidelines describe EEAT: experience, expertise, authoritativeness and trustworthiness. It is not a single ranking switch, but it summarises what search engines try to reward: content written from real experience, by people who know the subject, that can be trusted.

The engine is designed around it. Articles draw on real experience and specifics rather than generic statements, claims have to be supportable, facts cannot be invented, the author and business behind the content are clear, and every page is written to genuinely answer the searcher's question. The validation step described below checks for exactly these things. Content that reads as if it came from someone who has done the work is content that both readers and search engines trust.

Choosing what to write about

Before briefs, there is a more basic decision: which topics deserve a page at all. The easiest way to waste a content budget is to write about what is interesting to the business instead of what potential customers are actually searching for.

A useful way to prioritise is to sort candidate topics into three groups:

• Commercial searches. People looking for a service or product like yours, often with words such as "software", "service", "company" or a location. These pages bring in enquiries and deserve the most care.
• Problem searches. People describing the problem your service solves without naming a solution yet. These pages educate, build trust and link to the commercial pages.
• Background searches. General questions in your field. These can build authority, but they rarely bring in customers directly, so they come last.

Specific beats broad, especially early on. A new site has little chance of ranking for a two-word phrase that every competitor targets. It has a much better chance with a precise phrase that a smaller number of motivated people search for. Rankings for those specific phrases come faster, and the visitors they bring are closer to buying.

Briefs: deciding what a page is for

The most important step happens before any text is generated. A good brief answers a handful of questions:

• What is the searcher trying to do? Learn something, compare options, or buy? A page written for the wrong intent will not rank, however good it is.
• Who is reading? An operations manager and a developer searching the same phrase want different answers.
• What must the page cover? The specific questions and points that make it genuinely useful.
• What should the reader do next? Every commercial page needs a clear, relevant next step.

With a brief like this, the generation step has something to aim at. Without one, it produces the average of everything it has read, which is exactly the empty content described above.

Generation: prompts as specifications

The generation step uses carefully engineered prompts. The point of prompt engineering here is not clever phrasing. It is turning the brief into a precise specification: the structure the article must follow, the points it must make, the tone it must use and the things it must not do, such as inventing statistics or making claims the business cannot support.

Treating prompts as specifications has a useful side effect: when output is wrong, the fix is usually a clearer specification, not a different model.

Validation: the quality gate

The heart of the platform is the validation step. Every draft passes through deterministic checks before it can move on. These are rules, not opinions, so they behave the same way every time. Examples of the kind of checks that make a difference:

• Structure. Required sections and headings are present, in a sensible order.
• Coverage. The key points from the brief actually appear in the draft.
• Length and density. The article is long enough to be useful but not padded.
• Forbidden content. No invented statistics, no competitor claims, no phrases the business does not want associated with it.
• Formatting. Links, lists and headings are well formed so the document assembles cleanly.

A draft that fails goes back for regeneration or human editing, with the specific failure recorded. A draft that passes moves on. This is what the "quality gate" in the replay represents: nothing reaches publishing without clearing it.

The principle is the same one that runs through all my automation work. Put deterministic checks around every generative step. Language models are flexible and occasionally wrong. Rules are rigid and predictable. Together they produce output you can trust at volume.

Assembly and publishing

Once a draft passes validation, the rest of the workflow is pure automation.

python-docx assembles the content into a clean, consistently formatted document. The Google Drive API publishes it to the right folder. The Google Sheets API updates the content calendar with the article's title, target keyword, status, link and date.

This part is unglamorous but important. Manual publishing is where content operations quietly break down: files end up in the wrong folders, the calendar goes stale, and nobody knows what has actually gone live. Automating it means the record always matches reality.

One article, start to finish

Here is how a single article moves through the engine.

A subject starts getting attention in the business's field. The research step picks it up, gathers what is known about it, and runs keyword research. It finds a specific phrase with clear commercial intent and modest competition, and notes what the pages currently ranking for it cover and what they miss.

That research becomes a brief: the searcher is a decision-maker comparing options, the article must answer three specific questions the ranking pages handle poorly, it must include concrete examples from real work, and it should end by inviting the reader to get in touch.

Generation writes a draft to that brief. Validation checks it: the required sections are present, the three questions are answered, no statistic appears without a source, length is appropriate and every link is well formed. One check fails because a claim is too vague to support, so the draft goes back and is corrected.

The approved draft is assembled into a formatted document, published, and logged in the content calendar with its target keyword and status. Over the following weeks the article climbs for its target phrase, and visitors who arrive through it land on a page that answers their question and shows them the next step.

No part of that flow depends on someone remembering to do it, and no part of it skips a check because a deadline is close. The same standard applies to the tenth article and the hundredth. That consistency, week after week, is what compounds into rankings.

Why it produced results quickly

First-page rankings in two months and qualified leads in 30 days came from the combination, not from any single part:

• Every page targeted a real, specific search intent, so each one had a reason to rank.
• Commercial pages came with clear next steps, so visitors who arrived had somewhere to go.
• The quality gate kept standards consistent, so volume did not come at the expense of usefulness.
• Publishing was automated, so output was steady and nothing stalled in someone's inbox.
• The Sheets record showed what was working, so effort could shift toward the topics and formats that performed.

What I would tell a team starting AI content

• Write the brief before the prompt. Decide what the page is for and who it is for.
• Build a quality gate. Deterministic checks catch the problems that fluent text hides.
• Never publish invented facts. One made-up statistic can undo the credibility of an entire site.
• Automate the boring parts. Formatting, publishing and tracking should not depend on anyone remembering to do them.
• Measure outcomes, not output. Rankings, qualified visits and enquiries matter. Word count does not.

Summary

The AI content and SEO automation platform turns content into a disciplined production workflow: briefs define intent, engineered prompts generate drafts, deterministic validation acts as a quality gate, python-docx assembles documents, Google Drive publishes them and Google Sheets tracks everything. Built this way, AI content stopped being filler and started doing its job, with qualified inbound leads within 30 days and first-page rankings for commercial keywords within two months.

If you want to talk about building a content engine that actually brings in customers, ask Alfred or send me a message.