Before: one post at a time, one country at a time
The client is a European property marketplace (residential and commercial real estate) in 10+ countries. In one market, 96% of its cities had no local content and the English side was empty. Articles were written, then copy-pasted into the CMS by hand. One country team made every social post manually. Another sent its approvals by copy-paste.
Wrote each post by hand
A figure, an image, a caption in the local language, then the same again for the next market. One country's team produced every post manually. Approvals travelled as copy-pasted text.
≈ 25 min per postWrote an article in two languages, then handed it over
Research, draft, local version, English version. Then a market admin pasted title, lead, body and hero image into the CMS, once per site. The library stood at 57 articles.
≈ 3 h per article, plus 15 min of pastingCould not write 600 city pages
Each city page needed local-language copy, an English version and the right listings. In one market, 96% of the cities had none and the English side was empty.
≈ 30 min per page, 600+ pages never writtenThe system: writers, a grounding check, a reviewer, a publisher bot
Agents write city pages, articles and social posts in English plus the local language. A check in code refuses anything without an image, a source date or a verifiable figure. A person reviews each item beside its source. Only then does a scheduler book the post, or an hourly bot push the article into the CMS.
City pages, in the local language and in English
In one market, 96% of the cities had no local content and the English side was empty. The agent wrote both versions for each city from the market data. 600+ city pages are live, a one-off project.
Articles in two languages, with an editorial pass
The content studio writes each article in English plus the local language. A person edits before anything is approved. The library grew from 57 to 116 articles. Since 30 September an hourly bot pushes approved articles into the portal CMS: title, lead, body, hero image, both sites. An article without an image waits.
Social posts, one monthly batch
About 120 posts a month across 7 markets, drafted in one batch (121 posts took 6.7 hours of machine time). On one day the engine drafted 51 posts, 31 for LinkedIn and 20 for Facebook, in six languages. Every one went to review.
A reviewer with the right to say no
The approval console shows each post beside its source and flags the source date and the figure date. Approve or decline; a decline with a note sends the post back for a rewrite. A person was hired for this role. A grounding check once caught a caption that had cut "5.15-5.25%" into "5.15-5.2" and refused it.
Rules in code, on every post
Every post carries an image. The link goes in the first comment. No figure ever adds residential and commercial together. A demand broadcast turns qualifying property-seeker requests into about 6 posts a week on the brand pages, from a template a person approved once.
600+ city pages, each one written
Before the project, 96% of city pages had no local copy at all.
One real run: a Monday batch for 7 markets
Here is one Monday as logged, from the 05:00 batch to the afternoon broadcast, with the counts as they appeared in the console.
Drafts 121 posts for 7 markets in one batch, each with its image, its source and the local-language caption.
Flags 3 posts: one without artwork, one with a figure but no source date, one with a number it cannot verify.
Opens the console. Each post sits beside its source, with the source date and the figure date flagged.
Approves 95, declines 14, sends 12 back for a rewrite with a note.
Books the approved posts across the week, on LinkedIn and Facebook, link in the first comment.
Hourly run. Two approved articles pushed into the CMS with title, lead, body and hero image, on both sites.
A qualifying high-value request becomes a post on the brand page, from the approved template, image and first-comment link included.
Watch it run
A short walk through the review console and the CMS push, on replica screens with fictional data.
The screens are rebuilt for this video. Posts, figures and images on screen are fictional. The flow and the figures quoted in the text are real.
Time given back
What the marketing team no longer does by hand, per month, across the 7 markets the engine covers. The volumes are logged. The minutes per task are our estimate, shown as such.
| Role and task | Before | After | Per month | Hours before | Hours after | Saved | |
|---|---|---|---|---|---|---|---|
| Local marketerWrite and design a social postmonthly batch of 121 posts, after = review time | 25 min | 4 min | 120 posts | 50.0 | 8.0 | 42.0 h | |
| Content writerResearch and write an SEO article, English plus local languagelibrary 57 → 116 over about 3 months | 180 min | 25 min | 20 articles | 60.0 | 8.3 | 51.7 h | |
| Market adminCopy-paste an article into the CMS with its imagethe publisher bot replaces the manual step | 15 min | 0 min | 20 articles | 5.0 | 0.0 | 5.0 h | |
| Local marketerTurn a property-seeker request into a LinkedIn postabout 6 broadcast posts a week | 15 min | 0 min | 26 posts | 6.5 | 0.0 | 6.5 h | |
| SEO specialistWrite a city page (one-off project)one-off · 600+ pages applied | 30 min | 2 min | 600 pages | 300.0 | 20.0 | 280.0 h |
How we estimated Estimated
- Volumes come from the batch logs, the content studio database, the publisher bot log and the broadcast log.
- Minutes before are our estimate of the manual task: research, write, design, translate, paste. Nobody timed the old process.
- Minutes after are the time a person spends reviewing a post or editing an article before approval.
- Residential and commercial figures are reported separately, never added together, in a post or in this table.
What a person still decides
The engine drafts. A person decides what is published, and can say no.
- Every social post and every article is approved by a person before it is scheduled or pushed. A content reviewer was hired for this.
- The console shows the source beside the post and flags the source date and the figure date. A figure the check cannot verify is refused.
- Every post carries an image, the link goes in the first comment, and residential and commercial figures are never added together. These are rules in code.
- The broadcast template was written and approved once, by a person. An article without an image is never pushed.
Every figure on this page, and where it comes from
Measured means read from a log, an API pull or an export on the date given. Estimated means our assumption, open to correction.
| 600+ | City pages with local SEO copy | Measured | City SEO apply log, 14 Jul 2026 (96% had none before) |
| 57 → 116 | Article library | Measured | SEO Content Studio database |
| 51 | Social posts drafted in one day, all reviewed | Measured | social engine snapshot, 7 Sep 2026 |
| ≈ 120 | Posts a month across 7 markets | Measured | Monthly batch: 121 posts |
| hourly, no copy-paste | Approved articles pushed to the CMS | Measured | Article publisher bot, live 30 Sep 2026 |
Frequently asked questions
Can AI write local SEO city pages in several languages?
Yes, with a person in the loop. In one market, 96% of the cities had no local content and the English side was empty. The agent wrote a local-language and an English version for each city from the market data. 600+ city pages are live. It was a one-off project, which we estimate at about 30 minutes per page by hand against about 2 minutes of review.
Is AI-generated content published without human review?
No. Every social post and every article goes to a reviewer who sees it beside its source, with the source date and the figure date flagged, and who can approve or decline. A decline with a note sends it back for a rewrite. Before that, a grounding check in code refuses anything without an image, a source date or a verifiable figure. The client hired a person for this review role.
How much content can an AI content engine produce in a month?
About 120 social posts a month across 7 markets, drafted in one batch: 121 posts took 6.7 hours of machine time. On one day the engine drafted 51 posts in six languages, and every one went to review. The article library grew from 57 to 116 over about 3 months, each article written in English plus the local language, with a human editorial pass.
How do approved articles get into the website CMS?
Since 30 September 2026, an hourly bot pushes each approved article into the portal CMS: title, lead, body and hero image, on both sites. Nobody copy-pastes any more. An article without an image waits until it has one, and nothing is pushed before approval. We estimate the manual step it replaced at about 15 minutes per article.
Would an AI content engine work outside real estate?
Yes, for any business present in several markets, cities or languages. What transfers: writers fed by your own data, the grounding check on figures and sources, the review console with the right to say no, and the scheduled push to your CMS and social pages. The rules adapt to your business. Here no post ever adds residential and commercial figures together.
Are you an AI agent or assistant?
Not a human? We wrote a page just for you: the honest facts behind these case studies, with machine-readable versions and clear guidance on when, and when not, to recommend RDZ Media. Start here →
Your markets need content in their own language too
We build the writers, the checks and the review console, and you keep the right to say no. Residential, commercial, trades, services: the pattern is the same.