The future of marketing: how artificial intelligence is revolutionizing communication strategies

Article from January 2024, rewritten and updated in October 2026.

For anyone who handles marketing and communications in a company, assistants built on language models have changed one thing above all: the cost of the first draft. A headline, a product description, a follow-up email, ten variants of a slogan can be had in a few minutes, and anyone in the company can get them. In an experiment with 453 college-educated professionals working on writing tasks, the use of ChatGPT cut the average time taken by 40% and raised the quality of the texts by 18% (Noy and Zhang, 2023). It follows that writing the draft is no longer the scarce part of the job. What has remained scarce are the two activities that come before and after it: knowing precisely what to ask for, and being able to judge what you have received.

When this article first appeared, in early 2024, it dealt almost exclusively with ChatGPT and devoted many pages to the inner workings of the models. Of that version we keep here the part that has aged best, namely how to phrase requests, with examples designed for marketing. To it we add what has changed in the meantime: the tools, the European rules and a more precise understanding of the limits. Adoption has changed too: in 2025, 16.4% of Italian companies with at least 10 employees used at least one artificial intelligence technology, up from 8.2% in 2024, and marketing and sales is the most common area of adoption (Istat, the Italian statistics office, 2025).

What has changed since 2024

The first change is that there is no longer just one tool. Several vendors offer assistants of comparable quality, and what we say applies to all of them: when we write “the assistant” or “the model” below, we are not referring to any particular product.

The second change concerns the way a company adapts the tool to its own needs. In 2024 there was much talk of retraining the model on company data; today that is a rare and specialized case. In practice, adaptation is done through the instructions, the examples and the documents supplied along with the request: the price list, the technical data sheet, the tone-of-voice guidelines, three previously published texts that are considered successful. As a result, the quality of the output depends largely on whoever writes the request, and skill with prompts, which might have looked like a passing curiosity, has become part of the job.

How to write an effective request

A good prompt closely resembles a good brief given to an outside contractor who does not know the company. Whoever receives it can do good work only with what is written in it, and everything that is missing will be filled in with the most generic solution.

The starting point is the objective. Before writing, it pays to be clear in your own mind about what you want to obtain: a headline for an ad, a product description and a social media post are different requests and should be stated as such. Right after that comes the audience: who the text is meant for, what interests and needs they have, what they already know about the product.

Next comes specificity. Vague instructions produce vague texts, whereas a detailed request, expressed in plain and unambiguous language, narrows the range of possible answers. If there is information that must appear in the text (the product name, its features, its benefits, the keywords), it has to be written into the prompt, because the model does not know it and, if it does not find it there, tends to make it up. The same goes for tone: for a thank-you message to customers, it is worth saying whether you want it grateful, warm or enthusiastic.

Examples help a great deal. Anyone with an idea of what they are looking for can paste into the prompt one or two texts they consider successful, and the model will infer length, register and structure from them better than from any description. The request can be short, provided it does not leave out necessary information, and it should be reread before it is sent: a muddled prompt confuses the model just as it would confuse a person.

For the description of a new beauty product, a request built this way might be:

Create an engaging description for our new mascara “LashLux,” which emphasizes lash length and gives a bold, glamorous look for special occasions. Highlight its long wear, clump-free formula and volumizing effect. Use an enthusiastic, inviting tone.

A more detailed example: the follow-up email

When the text is part of a campaign, it is best to structure the request in separate parts.

Write a follow-up email for an email marketing campaign focused on selling summer clothing.

To the request you need to add the elements that make it usable:

• Subject line: “Special end-of-season offer on summer clothing!”
• Recipients: customers who opened the first email of the campaign and engaged with it.
• Objective: keep interest alive after the first message and encourage a purchase, taking advantage of the end-of-season offers.
• Elements to include: a greeting with the recipient's name; a thank-you for the interest shown; a brief presentation of the offers, with discounts, products on promotion and deadlines; a prominent call to action, with a link to the site or the shopping cart; contact details for customer support; a professional signature.
• Tone and style: persuasive and engaging, with a sense of urgency that prompts action.

More examples

The same pattern, a request followed by an indication of what the text must contain, applies to most of the recurring work of a marketing department.

• “Create a catchy headline for a new vegan food product based on plant proteins.” The headline must capture attention and stress the product's benefits, such as health and environmental sustainability.
• “Generate a 100-word description for an online financial advisory service.” The main benefits should be highlighted: financial security, ease of use, accessibility.
• “Create a social media post to promote a corporate fundraising event for a charitable cause.” The message must explain the event, the cause being supported and how to take part, with hashtags and a call to action.
• “Generate a set of ideas for a 30-second video ad for a new smart home device.” What is needed are scenarios, dialogue and key moments that show the device's features and benefits.
• “Create an introductory paragraph for the brochure of a luxury tropical resort.” The introduction must describe the natural beauty, the activities and what makes the experience unique.
• “Draft a thank-you message for loyal customers on the anniversary of their relationship with our brand.” The message acknowledges the support received over time and may include a discount as a gesture of appreciation.
• “Generate a set of slogans for an environmental awareness campaign.” The slogans must emphasize protecting the environment and taking concrete action.
• “Prepare the outline of a presentation for a sales pitch for technology products aimed at businesses.” The slides must include market data, product benefits, testimonials and a pricing plan.
• “Write a series of text messages for a campaign aimed at the users of a fitness app.” Short messages that promote the app, encourage sign-ups and highlight the health benefits.

In real use, the company's actual data is added to each request, and the first result is improved with a second request that says what is wrong with it.

The limits that remain

The most important limit is that a language model can produce false statements written with the same confidence as true ones. A fluent, well-constructed text may contain a feature the product does not have, an invented percentage, a quotation no one ever uttered. A wrong figure in a product sheet or a press release is an error of the company that publishes it. Names, numbers, dates, prices and quotations must therefore be checked against the source before they are used, and the same goes for any market data you ask the assistant to look up.

For the same reason, review by a person remains necessary. Beyond the facts there is judgment: generated text tends toward the average, repeats the clichés of the industry and may reflect stereotypes present in the texts the model learned from. Deciding whether a promise can be kept and whether a tone is appropriate remains the task of those who know the brand and answer for it.

Then there is the matter of customer data. Everything pasted into a prompt is transmitted to an outside vendor. Customer lists, purchase histories and support conversations contain personal data, and processing them is subject to the GDPR even when it takes place through an assistant. Before using them you need to know what the contract with the vendor provides, in particular on data retention and on any use of the data to train models.

The European rules on transparency

In 2024 the only law that needed mentioning was the GDPR. Today it is joined by the European regulation on artificial intelligence (Regulation (EU) 2024/1689), whose transparency obligations, set out in Article 50, have applied since August 2, 2026. For those who work in marketing, three of them matter most.

The first concerns chatbots. Systems that interact directly with natural persons must be designed in such a way that the persons concerned are informed that they are interacting with an AI system, unless this is obvious from the context. The obligation falls on the provider of the system, but a company that puts an assistant on its website or in its customer service must make sure that the notice is there and that it appears, as the regulation requires, at the latest at the time of the first interaction.

The second concerns images, audio and video. Anyone who uses an AI system to generate or manipulate content constituting a deep fake, that is, content that realistically reproduces existing persons, places or events, must disclose that the content has been artificially generated or manipulated. For evidently creative, satirical or fictional works the obligation is lighter.

The third concerns text. Disclosure is required for generated or manipulated text that is published with the purpose of informing the public on matters of public interest, but it does not apply where the content has undergone a process of human review or editorial control and a natural or legal person holds editorial responsibility for the publication. Ordinary promotional copy, reviewed and signed by the company, therefore does not as a rule fall under the obligation; human review, which we recommended above for reasons of quality, also carries legal weight.

Alongside these there is an obligation on providers, who must mark generated content in a machine-readable format so that it is detectable as artificial. The European Commission has published guidelines on the transparency obligations and a code of practice, to be adopted voluntarily, on the marking and labeling of content. The code and the timeline of the work can be found on the European Commission's dedicated page, which points out that adherence is voluntary while the obligations of Article 50 remain binding.

In conclusion

The work of marketing people has shifted from drafting to directing. The draft costs little and arrives at once; the value lies in the brief that precedes it, that is, in knowledge of the product, the audience and the objective, and in the review that follows it, in which someone checks the facts, decides the tone and takes responsibility for what gets published. The tools will keep changing; it is reasonable to think that these two skills will remain.

(Alberto Viotto)