Who does what, step by step
Every product on this site runs on the same division of labour. The machine does the parts that are wide: reading, iterating, measuring, drafting. A person does the parts that are narrow: choosing, deciding, saying no. Here is the split, per step of the cycle.
| Step | What AI does | What a person does |
|---|---|---|
| Think: research | Reads the market, five competitors, thousands of reviews and forum threads; runs ten category questions through four assistants; builds synthetic segments to narrow the hypotheses | Writes the list of questions the research has to close; runs the five customer interviews; reads it all and writes the document that says where the money is |
| Think: strategy | Structures interview transcripts, audits channel data, drafts alternative structure maps and positioning options | Sits in the sessions, draws the structure map, chooses the positioning, writes the plan with a closed price on every block |
| Build | Produces the message, copy in two languages and design on day one; generates the code, schema and entity data; runs QA across devices | Decides whether the direction is right, edits every line of copy that ships, signs off the page, hands it over |
| Run | Produces content from the brand context at the level's volume; reads performance daily; runs the monthly AI visibility measurement; drafts the report | Decides what goes out, adjusts what performance shows, writes page one of the report and the one decision it asks for |
The person, in most cases, is Dmytro Lobod. Behind him, the studio: designers, writers, developers, producers and media buyers who have worked together for ten years. AI did not replace them; it changed what they spend their day on.
The pipeline, in numbers you can check
- A market in a day. The research pipeline reads the category, five competitors and the audience's own words in the first two days of a Research Sprint, with every figure tied to a source. What used to be three weeks of desk research is now the part before the interviews.
- A page draft before lunch. On day one of a Launch Page, message, copy in English and Spanish and design run in parallel, and by the end of the day you see one direction worked through to the end, chosen by a person. Not a shortlist to sort through: a decision to approve or send back.
- A page in 72 hours. Brief on Monday, live on Thursday, on your domain, with the repository. The counter on the home page is not a metaphor.
- A brand measured in four assistants every month. The same thirty questions, in clean sessions, through ChatGPT, Gemini, Perplexity and Google AI, scored as share of voice against five competitors with a confidence interval. How that measurement is built is on the audit method page.
The brand context
The single most important file in every project. One package, updated monthly, that says who the brand is in a form a tool or a person can use on the first day: the voice in examples of what the brand sounds like and never sounds like; the facts, so nothing is invented; the do's and don'ts; the products, offers and season, current; the assets and which may be used.
Every prompt starts with it. Every freelancer gets it before the brief. Every template has it inside. It is why two brands that use the same tools through us do not sound alike, and why in-house teams set up through Marketing OS keep sounding like themselves after we leave. Context narrows what the machine produces; the editor chooses from what is left.
The library
Ten years of running an agency and 350 brands in twelve countries left behind something more useful than a portfolio: playbooks. What a launch in the beer category needs that a launch in banking does not. Which mechanics people actually play with, and which they abandon on the second screen. How a baby-care brand talks to a first-time parent, and how a car brand talks to a dealer. What a press moment needs to land on a page rather than in an archive.
The context of every new client is assembled from that library and then made specific: the playbook for the category, the structure that worked for that kind of business, the season calendar that fits the market. It is the part of the work AI cannot do from a blank prompt, because it comes from having shipped the thing before, with a client's name on it.
Synthetic research and real customers
AI reading what real people have already written, and simulating segments from it, is very good at one thing: narrowing. It finds the patterns in thousands of reviews in a day, and it tells you which three of your ten hypotheses deserve a real conversation. Validation studies published in 2026 put the accuracy of synthetic respondents between roughly 47% and 88% depending on the task: strong on segments and aggregates, weak on predicting what one person will do.
So every Research Sprint uses synthetic research to narrow and then talks to real customers about what is left: five short interviews or a quick survey, the same week. We do not skip the second part when the first one looks convincing, and we do not present the first part as the truth. Where a number comes from a simulation, the document says so.
What we don't do
- We don't train models on client data. Nothing a client shares goes into training, fine-tuning or improving any model. Tools are configured not to retain inputs. Customer data never enters an AI tool at all; that rule is written into every Marketing OS manual.
- We don't promise a spot in an AI answer. Nobody controls what an assistant says. We measure it the same way every month, publish the method, and move what can be moved: the entity work, the citable content, the press alignment. Anyone selling a guaranteed position is selling something they do not have.
- We don't count hours. Billing by the hour in the age of AI means charging less for getting better. Every product has a scope, a fixed fee or a monthly fee, and a date. The speed is our margin, not your discount.
Where to go next
- The first product on the ladder: the Research Sprint, five days.
- How the monthly measurement works, question by question: the audit method.
- Every number on this site with its source: sources.
- The whole cycle, on the home page.