Generative AI in marketing: where it helps and where it hurts

Generative AI multiplies how much content you can produce, not its quality. If what you produced was generic, you will now produce a great deal more generic content, and that costs rankings rather than winning them.

I have lived both sides: I use these tools daily, and I have also had to clean up sites full of articles generated without judgement. This article is about where the line sits.

Which tasks are worth handing over?

Task	Worth it?	Caveat
First draft of a text	Yes	With your material, not from nothing
Ad variants	Yes	Where it saves the most time
Summarising research and interviews	Yes	Very good at this
Adapting a text to several formats	Yes	Article to email to script to post
Translating and localising	Yes, with review	Local nuance escapes it
Interpreting data you supply	Yes	Excellent
Generating data or statistics	No	It invents them with total confidence
Writing your opinion or your method	No	It is precisely what differentiates you
Client case studies	No	It invents plausible figures

The last three rows are not nuances: they are the difference between content that builds authority and content that destroys it the day somebody checks a figure.

Why does content generated without judgement fail to rank?

Because it adds nothing that is not already on ten other pages. Search engines do not penalise using AI; they penalise results that do not help.

What makes a text worth something, and no AI can supply for you:

• A figure you measured. The only thing that is truly yours.
• A specific experience. What happened with a real client.
• An opinion that can be argued with. Generated text defaults to neutral and agreeable.
• What does NOT work. It almost never appears in automated content.

My practical rule: if the article could be published as-is on a competitor’s blog with the logo swapped, it is not worth publishing.

So how do you use it well?

The flow I follow:

1. I supply the material. Data, transcripts, notes, what I know about the client.
2. The AI builds the structure and first draft.
3. I add what only I have: the number, the anecdote, the position.
4. The AI polishes and adapts to formats.
5. I verify every figure and every claim.

Step five is not optional. Anything that looks like a sourced fact, a percentage, a study, an institution, must be verified or removed.

What about personalisation at scale?

It is the most-sold promise and the one that requires most care in Colombia.

You can genuinely adapt the message by segment, behaviour or funnel stage, and it works. But there are two limits:

The legal one. Law 1581 requires prior, express and informed authorisation to process personal data. Personalising with data the customer did not authorise, or combining databases without consent, is illegal here.

The perception one. There is a point where personalisation stops being pleasant and becomes uncomfortable. Mentioning something the customer does not remember telling you breaks trust in a second.

What mistakes do I see most?

• Publishing without verifying figures. The most serious and the most common.
• Losing your own voice. Fifty articles in the same neutral tone build no brand.
• Producing for the sake of producing. More posts is not more results.
• Pasting client data into tools without reading their terms.
• Believing it saves time on everything. Reviewing a bad draft takes longer than writing from scratch.

Frequently asked questions

Is generative AI useful for marketing? Very, for drafts, ad variants, summaries, format adaptation and interpreting data. Not for generating data, writing your opinion or drafting client case studies.

Does Google penalise AI-written content? It does not penalise the use of AI as such. It penalises content that adds no value, however it was made. The problem is not the tool, it is publishing the generic.

Can I ask ChatGPT or Claude for statistics? Not without verifying them. Models produce plausible, false figures with complete naturalness, sometimes attributing them to real institutions. Every figure must be checked at source.

How do I stop my content sounding generic? By adding what only you have: a figure of your own, a specific experience, an arguable opinion, and what does not work. If the text works equally well for a competitor, it is not worth publishing.

Is personalising with customer data legal in Colombia? Yes, with prior, express and informed authorisation, and using the data for what was authorised. Cross-referencing databases without consent is not legal.

How much time does it really save? A lot on repetitive, format-based tasks. Little or nothing on strategy. And if the draft comes out poor, reviewing costs more than writing from scratch.

Should I disclose that I used AI? There is no legal obligation for marketing copy, but you should not attribute to a person something they did not write, nor present as original research something that is not.