Artificial intelligence has produced two things in marketing at once: tools that save real hours of work and a wave of promises impossible to verify. Both land on your desk in the same emails, with the same big words — and if you run a company of 10–100 employees, with a limited marketing budget, the difference between them is measured directly in money wasted or saved.
This article is a filter. It does not recommend tools — the market changes from one quarter to the next — but gives you the criterion for judging any “AI marketing” offer, a list of the uses that produce verifiable results today, a list of those that are usually noise, and a way to test everything on your own company within 90 days.
The filter: one number before, one number after
The rule is almost embarrassingly simple: a use of AI in marketing is real if it changes a figure you were tracking anyway — the cost of a new customer, the conversion rate, the production time of a piece of content, the average order value. If nobody can say which figure will move, in which direction and in how long, it is not an investment, it is an attempt.
The corollary: the filter requires you to have the “before” numbers. Many small companies do not — they do not know what a lead costs them today or how many hours a campaign consumes. The first intelligent investment in “AI marketing” is, paradoxically, measuring the current situation. Without it, any vendor can tell you any story.
What is measurable today
Content production. Writing product descriptions, ad variants, emails and support articles is the area where the gain shows fastest, because the unit of measure is banal: hours of work per piece, before and after. A caution, though: AI accelerates production, it does not guarantee quality — a person who knows the company must remain the final editor.
Variants and testing. Where you used to write two versions of an ad, you can now test ten. The measure is the click-through rate and the conversion on each variant — and A/B testing tells you objectively which one works. The gain does not come from “AI creativity”, but from the number of hypotheses you can afford.
Segmentation and personalisation. Different messages for new customers, loyal customers and inactive customers, generated and sent automatically. The measure: open rate, click rate and sales per segment, compared with the single newsletter sent to everyone.
Response speed to enquiries. A chatbot that picks up questions and qualifies enquiries outside working hours changes a concrete figure: the time to first response — and, in many fields, the enquiry goes to the company that answers first. The measure: response time and the percentage of qualified enquiries.
Notice what all four have in common: none of them promises “more visibility” or a “modern presence”. All of them end in a column of numbers. For the numbers to be real, though, you also need their instrumentation: goals defined in your website analytics platform, distinct phone numbers or forms per campaign, the question “how did you hear about us?” recorded systematically for every enquiry. Without attribution, even good results remain unproven.
What is usually noise
- The “AI-powered” label on old tools. Many products have added an AI button and doubled their price. The control question: what can it do now that it could not do last year — and which of my figures does that change?
- Vanity metrics. Impressions, likes, “engagement” — if they do not demonstrably connect to quote requests or sales, they are decoration. Pretty reports are not results.
- Mass-generated content, with no distribution and no reader. Fifty AI-written articles a month that nobody reads are not a strategy. Volume without quality dilutes the brand — and, more subtly, makes you identical to every competitor using the same tools with the same generic instructions.
- Tools bought without a process. A subscription to an AI platform is not a marketing plan. If nobody in the company has time to use it systematically, the money leaves every month for nothing.
Real differentiation comes from the raw material your competitors do not have: your data, your cases, your people's expertise. AI is an amplification tool — it amplifies the absence of your own content too.
From 2 August, transparency becomes law
There is one more thing to factor in, with a precise deadline: from 2 August 2026 — less than two weeks away — the transparency obligations of the AI Act will apply, together with the full penalty regime. For marketing, concretely: AI-generated content will have to carry a machine-readable marking, deepfake materials will have to be labelled, and chatbots will have to tell the user there is no human at the keyboard. The “Digital Omnibus” amendments from June postponed other parts of the regulation — the obligations for high-risk systems — but not transparency.
In practice, if you use AI in producing materials or in communicating with clients, ask your vendors how they handle marking and disclosure. It is the kind of detail that costs little when solved in advance and a lot when solved after a penalty.
The 90-day pilot: how to test without getting burnt
- Pick a single channel and a single objective. For example: emails to existing customers, with the objective of increasing repeat orders.
- Measure the current situation for four weeks. The “before” numbers are half the pilot's value.
- Define the target in writing. Which figure must move, by how much, by when — agreed with the vendor or with your marketing person before starting.
- Run for eight weeks, without changing anything else. If you simultaneously change the prices, the website and the campaigns, you will not know what produced the effect.
- Compare and decide. Expand, adjust or stop — based on the numbers, not on impressions. A detail that saves many pilots: also set the stopping threshold in advance, the figure below which you end the test without further debate.
If the terms trip you up along the way — language model, prompt, hallucination, conversion rate — they have short explanations in the IT glossary.
The next step
Before any subscription or contract, take stock of the numbers you have: cost per new customer, production hours per campaign, conversion rates. The list of gaps shows you the first step by itself. Answers to the questions that come up most often are gathered at the frequently asked questions about AI and digital marketing, and if you want to build the 90-day pilot with someone who works on such projects daily, Neoxis does this as part of its AI and digital marketing services.