AI can write your marketing. Approval is what makes it publishable.

The marketing pitch of 2026 is autonomy.

Agentic tools now offer to research an audience, draft the posts, schedule them across every channel, watch the numbers, and shift budget between platforms without stopping to ask. For a business with no marketing team, that sounds like the whole problem solved.

It is not, because producing marketing content was never the expensive part for a small business.

The expensive part is being accountable for what gets published. Who approved that claim about turnaround times? Which version went out on LinkedIn? Was that image AI-generated, and does anyone need to know? Did any of it produce an enquiry?

We are building our own marketing system for exactly this problem. It is called Figlix, and it is designed to take one goal and produce the strategy, the channel-specific content, and the approval workflow around it. Designing it has made one thing obvious. Generation is cheap and getting cheaper. The approval step is the part that carries the value, and it is now also the part that regulators and platforms are looking at.

The generation step got cheap. The review step got valuable.

Ask a decent model for thirty social posts about your services and you will get thirty usable drafts in a minute.

Read them properly and you will usually find the same three problems.

The first is invented specifics. Models fill gaps confidently, so a draft acquires a response time, a client count, a certification, or a testimonial that nobody ever agreed to.

The second is flattened voice. Everything comes back competent and interchangeable, which is a real commercial problem when the audience is about to see a great deal more of exactly that.

The third is missing context. The model does not know that you stopped taking that kind of job in March, or that a claim you made in 2024 is no longer accurate.

None of these are fixed by better prompting alone. They are fixed by someone who knows the business reading the draft before it goes out, and by a system that records that they did.

We covered the general shape of this in where AI automation saves time before hiring more admin.

Platforms are labelling AI content, so your customers will notice

Meta has been attaching automatic "AI info" labels to content its systems detect as AI-generated for a while. This year Instagram added a profile-level "AI creator" label that accounts can switch on, which shows a note saying the profile posts content generated or modified with AI.

It is optional and it is aimed at creators rather than plumbers or accountants. The direction still matters.

Disclosure is moving from a thing nobody mentioned to a visible attribute of an account. Over time, customers will form opinions about businesses whose marketing looks entirely machine-made, in the same way they formed opinions about stock photography and fake reviews.

That is not an argument against using AI. It is an argument for using it where it saves you real time, and keeping something recognisably yours in the parts customers actually judge you on: the service pages, the examples, the answers to hard questions, the reply to an enquiry.

The EU transparency rules started on 2 August 2026

On 2 August 2026 the transparency obligations in Article 50 of the EU AI Act became applicable, alongside the Commission's Code of Practice on transparency of AI-generated content.

Two parts are worth understanding even if you are an Australian business.

Synthetic images, audio, and video need machine-readable marking so systems can detect that they were artificially generated or manipulated. This is a technical marking obligation that mostly lands on the providers of the tools, and the Commission has acknowledged no single technique currently meets the standard on its own, so a layered approach using metadata and watermarking is expected. A limited grace period runs to 2 December 2026 for marking on systems already on the market.

AI-generated text has a narrower rule. It must be disclosed when it is published to inform the public on matters of public interest, and the obligation does not apply where the content has been through human review or editorial control and a person or organisation holds editorial responsibility for it.

Read that carve-out again, because it is the useful part.

The regulation's own answer to AI-generated text is a named human who reviewed it and is responsible for it. The Commission is explicit that this means genuine examination and judgment, not a superficial or purely procedural tick.

Scope matters too. Article 50 reaches non-EU providers where the output of the system is used in the EU. Most local service businesses posting about their own services are not publishing on matters of public interest, and are not in scope for the text rule. If you sell into Europe, run an EU-facing site, or your product embeds an AI feature customers interact with, this is worth a proper look rather than a blog paragraph.

In Australia, the rule is simpler than people expect

There is no Australian AI Act, and no general requirement to announce that a caption was drafted by a model.

The Australian Consumer Law still applies, and it does not care who or what wrote the copy. Misleading or deceptive conduct is misleading or deceptive conduct, with no turnover threshold and no small business exemption. If a model invents a qualification, an availability claim, or a saving, the business made that claim.

The ACCC's compliance and enforcement priorities for 2026-27, announced in February, include manipulative and false practices in digital markets, alongside misleading pricing and discount claims. That is the environment your automated marketing runs in.

One more date to put in the calendar. From 10 December 2026, new privacy policy obligations under Australian Privacy Principles 1.7 to 1.9 apply where a business has arranged for a computer program to use personal information to make a decision that could reasonably be expected to significantly affect someone's rights or interests. Writing a caption is not that. Automatically deciding who gets a quote, a price, or a rejection may well be, and the OAIC has said guidance is coming before commencement.

What a workable approval step looks like

You do not need enterprise software for this. You need the steps to exist and to leave a trace.

Write the brand facts down once. Services you actually offer, service areas, the claims you are allowed to make, the claims you are not, tone, words you never use, and the offers currently running. Every draft should be generated from that, not from a blank prompt.

Give content a real status. Draft, in review, approved, scheduled, published. Nothing skips from draft to published because someone was in a hurry.

Record who approved what. A date and a name against a specific version. This is what "editorial responsibility" means in practice, and it is also what saves you when a customer quotes a post back at you six months later.

Keep the provenance. Which model, which prompt version, which source facts, whether the image was AI-generated. Store it even if you never show it publicly, because you cannot answer questions about content you cannot trace.

Run a claims check before approval, not after. Numbers, dates, credentials, guarantees, comparisons, and testimonials get verified against something real or they get cut.

Keep the human in the loop where the stakes are. Approvals, spend changes, and anything that speaks on behalf of a person should not be autonomous, no matter what the tool is capable of.

Manual publishing is genuinely fine

There is a quiet trap in agentic marketing tools, which is that connecting them to your accounts is often the hardest and slowest part.

Publishing access to the major platforms means business verification, developer applications, review processes, and scopes that can take weeks or months to obtain, and some platforms charge for API tiers that a small business will never justify.

Meanwhile a scheduled reminder, ready-to-paste copy, the right image at the right dimensions, and a tracking link works today.

We are building Figlix manual-first for exactly this reason. If the system produces content that is on-brand, approved, correctly sized for the channel, and tagged for tracking, a person spending ten minutes on Monday morning pasting it in captures nearly all the value, with none of the integration risk and none of the "the bot posted it at 3am" stories.

Automate the thinking and the preparation first. Automate the button press last.

Measure the outcome, not the output

The failure mode of automated marketing is a beautiful dashboard of things that do not matter.

Posts published, impressions, reach, and engagement are output metrics. They tell you the machine ran.

The number worth building is the cost of an actual customer. That means tracking links with a consistent naming convention on every published piece, a defined conversion on the website, and a straight line from a channel to an enquiry to a paying customer.

Cost per click is not cost per enquiry, and cost per enquiry is not cost per customer who stayed. Most small businesses discover the cheap channel and the profitable channel are not the same one.

We covered the tracking foundation in how to track website traffic that turns into leads.

What to set up this month

If you are adding AI to your marketing, do these before you add more volume:

1. Write your brand facts, approved claims, and banned claims into one document.
2. Decide which content types require human approval. In most small businesses, that is all of them.
3. Add a simple review status to your content, even if it lives in a spreadsheet.
4. Record who approved each piece and when.
5. Verify every number, date, credential, and guarantee before publishing.
6. Note when an image or video is AI-generated, and keep that note with the asset.
7. Put consistent tracking links on everything you publish.
8. Check whether AI decides anything about a person in your business before 10 December 2026.
9. If you sell into the EU, get advice on where Article 50 applies to you.
10. Review one month of published content and ask which pieces produced an enquiry.

The takeaway

The tools got good enough to write the marketing. They did not get good enough to be accountable for it.

Platform labels are making AI content visible to customers. The EU's answer to AI-generated text is human review and editorial responsibility. Australian consumer law has always held the business responsible for its claims, whoever typed them.

All three point at the same small piece of process: a person who knows the business, reading the draft, approving a version, and leaving a record.

Build that step properly and you can safely let AI do a great deal of the rest.

Read more about practical AI ideas for small business websites, AI agents updating websites, or contact Vritul if you want a marketing and automation workflow that keeps human approval in the loop.

Sources: Article 50 of the EU AI Act, European Commission FAQ on Article 50 transparency obligations, Engadget on Instagram's optional AI creator labels, Social Media Today on Instagram AI creator labels, ACCC compliance and enforcement priorities 2026-27, OAIC consultation on transparency in automated decision-making.