Automated Content Production: Our Workflow

Last updated: July 2026

The automated content production we run at ITOPS.be starts with a voice note from the client and ends with a post that is actually live. In between: hosting of the inputs, formatting through templates specific to each network, client approval in a dedicated portal, then scheduling that triggers publication on its own.

Contents

• How do we capture the client's raw material?
• How does a voice note become concrete content?
• Why use a different template for each social network?
• Why does the client approve every post before publication?
• How does scheduling trigger the actual publication?
• What this workflow does not do
• Which GDPR and AI Act rules apply to automated content production?
• How can you adapt this workflow for your own business?

How do we capture the client's raw material?

For an SME, the bottleneck is almost never a lack of desire to communicate. It is the moment you have to sit down in front of a blank page. The manager who has just finished a job, delivered an installation or received a thank-you from a customer holds the material, but will not turn it into text that evening.

So we designed the first step around that reality rather than against it. The workflow captures information by voice or text, along with files of any type, and everything is hosted. The client speaks for three minutes into their phone, or writes two lines if they prefer, and uploads site photos, a PDF datasheet or a screenshot. The chain accepts the input as it arrives.

This choice has a direct effect on adoption. An interface that demands clean, structured text simply moves the writing work onto the client, and production stops after three weeks. An interface that accepts a raw voice note, hesitations and digressions included, stays usable on a Tuesday evening at 8 pm.

Hosting these inputs is not a technical detail. Voice notes, files and exchanges constitute data processing under the GDPR as soon as they contain information about identifiable people, which means you must know where these items reside and who can access them.

How does a voice note become concrete content?

The next step is to take this information and turn it into concrete content. This is the core of the work, and also where expectations most often aim at the wrong target.

A three-minute voice note rarely contains a post. It contains a fact (an intervention carried out, a problem solved, something new), a context, and often an unfiltered opinion. The processing has to isolate these elements, rank them, then produce two kinds of output: a blog article, longer and more argued, and structured posts for social networks.

Two principles guide this transformation in our chain.

• Invent nothing. What is not in the client's input must not appear in the output. A figure, a customer name, a deadline or a guarantee that was never stated cannot be added to make things look better. This is the strictest rule of the system, and the one that makes the approval step indispensable.
• Keep the client's voice. An SME manager who is recognisable in their posts is worth more than smooth, interchangeable text. The processing relies on the original wording, including turns of phrase that are not academic, rather than rewriting everything in neutral corporate language.

For readers who first want to know why to automate their social presence rather than how, we have covered that question separately, in our article on automated content production on social media for SMEs.

Why use a different template for each social network?

We have set up templates to structure, in terms of visuals and storytelling, the information intended for Facebook, Instagram, LinkedIn and Google Business Profile. The plural matters: there is not one template, there is one per destination.

The reason is twofold. First, technical: each platform exposes its own publishing interface, with its own media formats and constraints. Instagram requires a multi-step publishing flow for visual content [7], LinkedIn has a separate publishing API for pages [8], and Google Business Profile handles local posts attached to a business listing [6]. A single piece of content pushed identically runs into these differences.

Second, editorial, and this is the more important one. The same fact is not told the same way depending on the reading context.

The template is not limited to text. It also structures the visual and the narrative sequence, so that the same raw material produces four variants that are consistent with each other without being four copies.

Why does the client approve every post before publication?

This is the step we defend most, and the one that distinguishes a serious production chain from a content factory.

Once the content is created, everything goes to a client approval platform, where the client can leave comments, approve the posts they like, and the approval automatically triggers the next step of the workflow. Three mechanisms coexist in this portal: commenting, post-by-post approval, and automatic triggering of what follows.

What this step really brings:

1. A factual check. The client is the only one who knows whether the town name is right, whether the job mentioned can be named publicly, whether the quoted price is still current. No model can settle that for them.
2. A tone check. Brand voice drifts quickly when nobody proofreads. Regular approval acts as a permanent proofreader.
3. A legal check. Publishing on a company's accounts commits that company. Approval materialises the human decision behind the publication, which is useful well beyond psychological comfort.
4. Useful granularity. The client approves the posts they like, not a whole batch to take or leave. A rejected post does not block the others.

The often underestimated point is the fourth mechanism: approval does not send the file back to a human operator who would re-enter everything elsewhere. It is itself the event that moves the workflow forward. This avoids the classic scenario where automation stops dead as soon as a human has to step in, then resumes at the price of a manual copy-paste that reintroduces the errors automation had eliminated.

In other words, the human is not a tolerated exception in the chain. They are a full step of it, with their own interface and their own trigger.

How does scheduling trigger the actual publication?

The last step is scheduling: on the day, when the scheduled time arrives, the post is actually published on the social network.

The nuance lies in the word "actually". Many tools stop at an editorial calendar, a table showing what should be published and when, leaving a human to open each platform at the right moment. That last manual link is the first to break, because it depends on someone being available.

In our chain, the scheduled date is a trigger, not an intention. When the time comes, the approved content goes to the relevant account through the platform's publishing interface.

This imposes an operating constraint that has to be accepted: account access must remain valid over time. Platform API authorisations expire, get revoked when an administrator changes, or evolve with successive versions of the interfaces [7][8]. An automated publishing chain therefore needs monitoring, exactly like any other production service.

What this workflow does not do

We prefer to be explicit about the limits, because a system presented as all-encompassing ends up disappointing.

• It does not define your strategy. The workflow processes what the client gives it. If the inputs are thin or irregular, the outputs will be too. The raw material stays human.
• It does not do community management. Replying to an aggressive comment, handling a negative review or arbitrating a sensitive situation is a human job, with no possible delegation.
• It guarantees no commercial result. We promise no volume of leads, no quantified gain in visibility, no payback period. Such promises cannot be kept contractually, and any performance claim should be verifiable before it is put forward.
• It does not remove the need for review. The approval step is not decorative. A workflow where the client approves everything without reading loses most of its value as a control.

Which GDPR and AI Act rules apply to automated content production?

Automated content production hosts voice notes, files and exchanges that regularly contain personal data: the manager's voice, names of customers mentioned, photos showing people. These are data processing operations in their own right, and GDPR obligations apply to them in full.

Four points must be settled at design time, not regularised after the fact:

• The legal basis. Each processing operation must rest on an identified foundation, not on habit.
• Minimisation. Collect and keep only what is really needed to produce the content. A voice note kept indefinitely for no reason is a risk, not an asset.
• Data residency. Knowing which jurisdiction the hosted files and associated processing reside in is a requirement, especially when a third-party AI model is part of the chain.
• Subcontracting. Any provider processing data on the client's behalf must be bound by contract, with the corresponding safeguards.

The processing must also appear in the company's record of processing activities, whose conditions of obligation are set out by the Data Protection Authority [4]. The Belgian authority also publishes guidance specific to artificial intelligence, worth consulting before scaling up a use case [3].

The European framework on artificial intelligence applies too. The regulation structures obligations according to the risk level of systems [1], and its Article 50 provides transparency obligations for content generated or manipulated by AI, applicable from 2 August 2026 [2]. An SME that publishes content produced with the help of AI is well advised to examine now what this deadline implies for its publications, depending on the role it plays in the chain.

How can you adapt this workflow for your own business?

We approach this kind of project in two stages, following our Blueprint then Build approach.

The Blueprint scopes before coding: which inputs the client can realistically provide and how often, which destinations are relevant (not all four suit every business), where the data resides, who approves and within what timeframe, and which API authorisations must be obtained and then maintained.

The Build implements in stages, starting with a deliberately limited scope: one destination, a sustainable rhythm, a simple approval circuit. You widen it once the circuit is running, not before.

A few points of vigilance before committing, with us or with another provider:

• Check the provider's references and its ability to operate the chain over time, not just to deliver it.
• Demand a reversibility plan: what do you get back, and in what form, if you leave?
• Have availability and support commitments specified in writing, rather than settling for a sales intention.
• Question the provider's security posture, including with regard to the NIS2 directive if your sector is concerned.

On funding, regional schemes support the digital transformation of businesses, but they follow strict eligibility conditions and are never guaranteed in advance: the list and current conditions are published by Digital Wallonia, which remains the source to consult before any budget planning [5].

It all comes down to one idea. A content chain is not worth the number of steps it automates, but the solidity of the point where the human decides. For us, that point has a name: the approval portal. Everything else exists to make it fast and to faithfully carry out what it approves.

Sources and References

1. European Commission: AI regulatory framework (EU AI Act)
2. EU AI Act: Article 50: Transparency obligations for certain AI systems
3. Data Protection Authority (Belgium): Artificial intelligence
4. Data Protection Authority (Belgium): Who must keep a record of processing activities?
5. Digital Wallonia: Support for the digital transformation of businesses
6. Google for Developers: Business Profile APIs: local posts data
7. Meta for Developers: Instagram Platform: content publishing
8. Microsoft Learn: LinkedIn Marketing API: Posts API