Inside an AI marketing team: how one article gets made

Most agencies show you the finished article and keep the kitchen closed. We would rather show you the kitchen, because if you are thinking about an AI marketing team, the real question is not "can AI write?" but "who checks it, and how?". Here is exactly how one article in this Journal goes from an idea to a page on Google, including the parts AI agents still cannot do alone.

The AI marketing team behind one article

Hadi is a team of AI agents run by a human growth expert. For content, three roles matter:

• Sami (SEO agent) chooses the topic and gathers the research.
• Zara (content agent) drafts the article.
• The human expert checks the facts, the brand voice and anything regulated, then approves or sends it back.

You can meet the full team on the team page. The other agents (Nora on the website, Rayan on ads, Maya on CRM, Omar on WhatsApp) are not involved in writing, but their work often supplies the questions we write about.

Step 1: Sami picks a topic from search intent

Sami does not start with "what should we post this week?". He starts with what people are trying to find out, and why.

The method, in plain terms:

1. List the questions buyers actually ask. Owners of clinics, setup firms and agencies ask the same things on calls and in WhatsApp: what does a marketing team cost, why do leads go cold, is an agency better than hiring. Those questions become candidate topics.
2. Check what Google shows for them. Sami looks at the current results, the "People also ask" questions and the autocomplete suggestions, to understand what kind of answer searchers expect: a price breakdown, a comparison, a checklist.
3. Classify the intent. Is the reader comparing options (close to buying), solving a problem, or just curious? We prioritize the first two.
4. Find the gap. If the top results are generic or written for another country, a specific, sourced answer for Dubai has a reason to exist.

One honest limitation: for this launch batch, we did not have keyword volume data, so topics were chosen on intent, not on search volume. We will not quote volumes we do not have. Once the articles have been live for a while, Search Console will show which queries they actually appear for, and that real data will guide the next topics.

Step 2: research and a brief

Before any drafting, Sami builds a brief: the reader, the main question, the angle, and a list of sources. Our rule is strict. Every number presented as fact needs a credible, linkable source, preferably official (DLD, Dubai Health Authority, u.ae), platform documentation (Google, Meta, WhatsApp) or reputable research. If no source exists, the number is dropped or clearly labelled as our estimate, with the reasoning shown.

Sami also reads three to five competing articles on the same topic to see what readers expect and what is missing. He does not copy them.

Step 3: Zara drafts

Zara writes to the brief: a short opening that names the problem, real headings, short paragraphs, a table or checklist where it helps, and a closing paragraph about how our team handles the topic, without hype or guarantees. Sources go inline, right where the number is used, so a reader can click and check.

This is where AI agents are genuinely strong. Turning a solid brief and a pile of sources into a clear, well structured first draft is fast work for an AI agent, and it can produce several angles or headlines to choose from.

Step 4: sources are checked again

Drafting and checking are separate steps on purpose. Generative models predict likely text, and they can state things that sound right but are not. Google says the same thing in its guidance on using generative AI content, updated in October 2026: it is critical to manually fact check and review all AI generated content for accuracy and trustworthiness before publishing, and that review should also cover titles, meta descriptions, structured data and image alt text.

So every figure is matched against its source: does the page actually say this, is it the right year, is it about the UAE or somewhere else? When a source disagrees with itself or with another source, we say so in the article or leave the number out.

Step 5: the human expert reviews and approves

This is the step we will not automate. The human expert reads the draft as a sceptical business owner would and asks:

• Is it accurate? Every number sourced, nothing overstated.
• Is it useful? Would a busy clinic owner learn something they can act on this week?
• Does it sound like us? No filler, no invented stories, no claims about clients or results we do not have.
• Is anything regulated? This matters a lot in Dubai. For clinics, the Dubai Health Authority's Standards for Medical Advertisement Content in Social Media ban wording such as "the best", "100%" or "no side effects", and posts naming a facility need its medical director's approval. For real estate, every property ad needs a DLD permit, as we explain in Property Finder and Bayut leads: why speed decides who wins.

If any answer is no, the draft goes back. Only then is it published.

Step 6: publication and measurement

Each article is published with its author agent and researcher named at the top, a list of sources at the bottom, and a short FAQ. Then we watch it in Google Search Console. The Performance report shows clicks, impressions, click through rate and average position, by query and by page. That tells us whether an article is being shown for the questions we wrote it for, or for different ones, and which pieces deserve an update.

What Google actually says about AI content

There is a lot of fear and folklore here, so here is what Google has published:

• In its February 2023 guidance, Google said it focuses on the quality of content rather than how it is produced, and that using automation, including AI, mainly to manipulate search rankings violates its spam policies.
• Its spam policies define scaled content abuse as generating many pages primarily to manipulate rankings without helping users, whether made by people, automation or both.
• Its page on helpful, people first content asks creators to think about Who, How and Why: make clear who created the content, explain how it was produced where readers would expect it, and make sure the reason for it is to help people, not to attract search visits.
• In December 2022, Google added Experience to its quality rater framework, which became E-E-A-T: experience, expertise, authoritativeness and trust, with trust described as the most important.

Our reading: AI drafting is fine. Publishing unchecked AI text at scale is not, and neither is hiding how content is made.

Where AI agents shine and where a human is still needed

Yes, this article was made this way

Transparency is the point of this piece, so here it is plainly: this article, and the other eight launch articles in our Journal, were researched by Sami and drafted by Zara, then every figure was checked against the source linked under each article before publication. Where we could not find a source, the figure is labelled as our estimate. That includes How much does a marketing team cost in Dubai, Marketing agency vs in house vs AI marketing team in Dubai, Why aesthetic clinics in Dubai lose leads after 8 pm, WhatsApp response time: what Dubai customers expect, Google Ads for business setup firms in Dubai and your startup marketing plan for the first 90 days. If you spot an error in any of them, tell us and we will fix it and say so.

How Hadi's team handles this for you

For a client, the same process runs on your topics and your rules: Sami and Zara research and draft, your human growth expert checks every fact and claim, and nothing regulated goes live without the approval of the right person on your side, such as a medical director or permit officer. You can see how the agents work together on the live OS page. If you want content that is fast to produce and still safe to publish, talk to us.