How to use AI in branding: the three phases of the new operator
There is a measurable difference between those who have learned to use AI in branding and those who still only have access to it. Four independent reports published between April and May of 2026 are beginning to draw that difference clearly. Read together, they point to a learning curve, not to a technology problem.
Brand operations in 2026 have three phases. Before the prompt is judgment, during the prompt is literacy, after the prompt is curation. AI works well in the middle phase. Brands that are investing in the new operator, and not only in the tool, are the ones that will get through homogenization without losing their voice.
In this article
- Four reports, four markets, one shared reading
- The Italian experiment gives the effect a name
- The three phases of brand operations in 2026
- The distribution context: Alphabet’s Q1 2026
- The production context: IAB Brasil’s Digital AdSpend 2026
- An applicable diagnosis: the five-prompt test
- The productive discussion in 2026 has already moved on
- Frequently asked questions
- Sources and references
Four reports, four markets, one shared reading
On April 23, 2026, Forbes amplifies the paper by Chaoran Liu, Tong Wang and S. Alex Yang from London Business School. The American Marketing Association publishes its Future Trends in Marketing 2026. Alphabet reports the first quarter of 2026 with Google Network revenue down 4% and queries at an all-time high. On April 29, IAB Brasil releases the Digital AdSpend 2026, with R$ 42.7 billion invested in digital advertising in 2025 and 51% already going through direct buying, with no agency in between.
Four independent sources. Four different methods. Four markets. Read together, they point to a learning curve, not to a technology problem.
The Italian experiment gives the effect a name
On March 31, 2023, the Italian data protection authority banned ChatGPT across the entire country for thirty days. Researchers from London Business School used the interval as a natural experiment and measured the marketing content of small Italian businesses on Instagram before, during and after the block.
During the ban, without access to ChatGPT, local restaurants posted with 15% higher lexical diversity, 12% higher syntactic diversity, and got 3.5% more engagement. All of this with shorter text and a lower publishing frequency. When ChatGPT came back, the three metrics regressed to their previous level.
The useful reading of this experiment is that the gain did not come from the absence of the tool. It came from the editorial work that the small businesses took back out of necessity during the ban, work they had fully outsourced before. With no other option, they went back to defining what they wanted to say before typing. They went back to looking at the finished text before publishing. The gain lies in those two reclaimed operations.
The three phases of brand operations in 2026
Brand operations in 2026 have three phases. Before the prompt, during the prompt, after the prompt. AI works well in the middle phase, executing in seconds what used to take hours, and that is real efficiency. The invisible cost shows up when the operation is reduced to the middle phase. Those who use AI well in branding are the ones learning to operate in the other two phases with the same seriousness they devote to the prompt.
Before the prompt: judgment
The first phase is judgment. An explicit definition of what sets this brand apart: which words never appear in its voice, which argument structure is forbidden, what the narrative hierarchy is, which phrases have become assets. Without documented judgment, the prompt starts from a blank page and ends where all the competition ends, at the statistical center of the training corpus. With documented judgment, the prompt starts from something the model does not have, and the output already arrives filtered through something only this brand can offer.
During the prompt: literacy
During the prompt, the operation is called literacy. Understanding how the model works in order to know what to ask for. Generative models work by conditional probability. Each word that comes out pulls the text toward the average of what was seen in training. Those who master this change the way they prompt: they ask for contrast, demand negative examples, provide voice references, ask for the output in variations and then choose. Those who do not master it ask for a post on a given topic and receive the market average.
The difference between a literate operator and a beginner operator today is of the same order of magnitude as the one between those who knew how to write a brief and those who did not in 2010.
After the prompt: curation
After the prompt, curation. Looking at the output as a draft. Working on top of it, always. The work there involves constant discards: the paragraph that sounds generic, the sentence that smells of the model, the passage that is good but is not the brand’s. Brands that treat the output as a deliverable publish the average. When the operation recognizes the output as a beginning, something that belongs to the brand itself comes out.
The distribution context: Alphabet’s Q1 2026
Alphabet’s Q1 adds an important layer. Google Network revenue fell 4% compared to the same quarter of the previous year, even with total query volume at an all-time high. The difference was absorbed by AI Overviews, which answer inside Google itself and remove the need to visit the site that produced the content.
In Brazil, the case became a CADE administrative proceeding in April 2026, with the hypothesis of improper use of journalistic content to feed Gemini. Content distribution is narrowing at the same time that production is standardizing. The two compounded movements mean that brand content now needs more distinction to pass through the same narrowed discovery filter. Volume does not offset convergence.
The production context: IAB Brasil’s Digital AdSpend 2026
IAB Brasil shows the other end of the chain. In 2025, 51% of Brazilian digital advertising was direct buying, with no agency. That share grew 16% in the year. Buying through an agency fell 2% in the same period.
The immediate reading of this figure is media internalization, and it is correct. The additional reading is that the in-house teams taking on this work need to learn the three phases of AI operations, or they will produce content identical to the competition’s using the same tool with the same lack of judgment. The advantage of internalization depends on the literacy of whoever internalizes.
An applicable diagnosis: the five-prompt test
There is a simple diagnosis for any brand to measure where it stands on the curve.
- Select the five prompts most used by the marketing team over the last sixty days.
- Run each of them in three different generative models (for example, ChatGPT, Claude and Gemini), with no additional brand instruction.
- Compare the outputs with the brand’s published communication over the last six months.
- Calculate the average similarity between what comes out of the models and what is published.
If the similarity exceeds 60%, the symptom is an operation truncated at the extremes of the prompt. Judgment was missing before, curation was missing after. This is not a problem with the tool. It is a position on the curve, and you climb a curve with method.
The productive discussion in 2026 has already moved on
As for whether AI fits into marketing, that is settled. It fits, it works well in execution, it is here to stay. What is still open is how to learn to use the tool so that the output belongs to the brand and not to the market. That learning has a method, has a measurable curve of return, and has a cost of not-learning that shows up late. Not in this month’s budget. In the pricing power curve two years later.
Brands that are investing in the new operator, and not only in the tool, are the ones that will get through homogenization without losing their voice.
Frequently asked questions about using AI in branding
Does using AI in branding hurt content performance?
The 2026 data show that the problem is not the tool, it is the operation. When AI use is reduced to the prompt, with no judgment before and no curation after, content converges toward the statistical center of the training corpus. When the three phases are relearned, AI works as a real productivity gain with no cost to distinction.
What are the three phases of using AI in branding?
Before the prompt is judgment. During the prompt is literacy. After the prompt is curation. AI executes well in the middle phase. The other two need to be relearned at a new scale.
What did the 2023 Italian ChatGPT ban experiment show?
Researchers from London Business School used the 30 days
