• Series voir-pages
  • page 06 / 26
  • department 1 of 8
  • channels implemented 0

Marketing is not “write a post”. It is six channels, one bus and a measurement return.

The marketing department in VOIR AI is built like a mixing desk: the owner's brief comes in on the top bus, splits across six channels — content, SEO, social, campaigns, visuals, email — and every channel drops its artefact onto the bottom measurement bus, from where the reading goes back into strategy. Below is the honest picture: what the department should do in the target architecture, what already runs inside our Marisova ecosystem as in-house tooling, and how the department sits inside the AI CEO cycle.

VOIR AI product · nothing implemented Marisova ecosystem · some functions run

See the department desk Honest statuses

Demonstration · department desk

One brief, six lanes, a single return bus

Marketing without a measurement return degenerates into text production. That is why the desk is closed-loop: the brief enters at the top, the channels work in parallel, each artefact lands on the bottom bus together with its reading, and the reading goes into the report and into the next cycle. When a consequence cannot be measured, the artefact travels with an explicit “no data” mark instead of an invented figure.

In the demonstration scenario: 6 department channels, 8 neighbouring departments supplying data, 5 steps of the management cycle, 0 channels implemented in the product.

Desk · brief → bus → channels → measurement → report

Demonstration · prepared scenario
SIGNAL IN · owner brief RETURN BUS · measurement CNTSEO SOCCMP VISEML ContentSEO SocialCampaigns VisualsEmail · leads IN Brief OUT Report
  1. INThe owner's brief enters the top bus together with acceptance criteria and constraints — fixed before any channel starts working.
  2. CNTContent channel: positioning, copy, editorial plan, drafts for a specific outlet and audience.
  3. SEOSearch channel: keyword mapping, page structure, technical audit, control of indexing and snippets.
  4. SOCSocial channel: content calendar, publications, audience replies, reach and retention readings per format.
  5. CMPCampaign channel: segments, offers, ad drafts, budget model and testable hypotheses.
  6. VISVisual channel: covers, illustrations, short vertical clips — material serving the other channels.
  7. EMLEmail and lead generation: list segmentation, email drafts, qualification of inbound leads.
  8. MTRReturn bus: every artefact arrives here with its reading or with an explicit “no data” mark.
  9. OUTOwner report: what was done, what was measured, which decisions stayed with a human, what changes in the next cycle.

The demonstration runs a prepared scenario and is not connected to a live AI service. The faders rising in the animation illustrate the order in which channels work — they are not data about real activity. No channel is implemented in the VOIR AI product, which is why every knob sits at zero in the static state.

Target architecture

Six channels: what each one owns

A channel is not “an area of activity in general” but a contract: which artefacts it produces, which data it consumes, how it is measured. Every channel on this page carries two statuses: the first is its implementation in the VOIR AI product, the second is whether a working tool already exists inside our Marisova ecosystem. The second status never means availability to customers.

CNT Target architecture

Content

Product positioning, copy for each outlet, an editorial plan for the period, drafts tailored to a channel and an audience. The content channel does not invent meaning from scratch: it takes it from research data and from the product record.

  • positioning and messages
  • editorial plan for the period
  • drafts per channel

VOIR AI product — not implemented. Marisova ecosystem — an SEO article pipeline runs: texts are generated, edited by a human and published; a Telegram Mini App studio generates images.

SEO Target architecture

Search

A keyword map driven by the business goal, page structure, control of meta tags and headings, technical audit: speed, indexing, broken links, duplicates. A separate duty of this channel is to never break live pages for the sake of new ones.

  • keyword map and page structure
  • technical audit with priorities
  • meta tag and snippet control

VOIR AI product — not implemented. Marisova ecosystem — SEO articles are published in the news section of marisova.ru; a local SEO baseline is maintained with Lighthouse measurements and audits of several dozen URLs. It is an internal tool, not a service.

SOC Target architecture

Social

Content calendar, preparation of posts, replies to the audience, reach and retention readings per format. Publishing outward happens only with human confirmation: the channel prepares the material and the estimate, but does not press the button itself.

  • content calendar
  • post drafts
  • reach summary per format

VOIR AI product — not implemented. Marisova ecosystem — a local Instagram Reels analytics system runs for one account: a localhost dashboard, scheduled metric snapshots and a content assistant that breaks clips down. It is not a public service.

CMP Target architecture

Campaigns

Audience segments, offers, ad drafts, a budget model and hypotheses that can actually be tested. In the target architecture a campaign must have a stop criterion: without it spend grows while the decision stays “based on a feeling”.

  • segments and offers
  • ad drafts
  • budget model and stop criterion

VOIR AI product — not implemented. Marisova ecosystem — the GTM pipeline runs in dry-run: research, lead qualification, communication drafts, a local CRM. Real sends are blocked by a hard gate and the paid ad budget in the configuration is zero.

VIS Target architecture

Images and video

Visual material for the other channels: covers, illustrations, short vertical clips, editing with subtitles and a narrator. This is a service channel — it does not set its own tasks, it takes them from the editorial plan and the calendar.

  • covers and illustrations
  • short vertical clips
  • a material pack per campaign

VOIR AI product — not implemented. Marisova ecosystem — photo and video generation runs through an external API (qwen-image and wan model families), including joint speech-and-video generation; there is a vertical clip pipeline and a long-form documentary pipeline with quality control.

EML Target architecture

Email and leads

List segmentation, email and sequence drafts, qualification of inbound leads, handover of the hot ones to a human. This channel handles other people's personal data, so it runs in the strictest mode: nothing goes outward without confirmation.

  • segments and email sequences
  • lead qualification
  • a handover note to a human

VOIR AI product — not implemented. Marisova ecosystem — email drafts are generated by the pipeline and stored in a drafts folder; the real-send flag is off and protected as a hard gate rather than as a default setting.

Measurement

The return bus: what exactly we count

Every channel drops onto the bottom bus not “a report on work done” but a reading with a named source. The source is mandatory: a metric without one is not accepted by independent QA and never reaches the owner report. Below is the minimal set of the target architecture. Not a single performance figure appears on this page: there are no customers and no sales.

Reach and impressions

How many people saw the material and how often. Counted per channel rather than as a marketing-wide average — otherwise it is impossible to tell what actually worked.

source: platform API · when unavailable — a “no data” mark

Clicks and depth

Clicks through to the page, time on it, whether the reader reached the target block. This shows the material brought a person who kept reading, not just traffic.

source: on-site analytics counter · UTM tags mandatory

Requests and qualification

How many enquiries arrived and how many of them are on-target against criteria fixed before the campaign started. Qualification without pre-agreed criteria turns into fitting the answer to the result.

source: pipeline CRM records · criteria from the task card

Cost of acquisition

Channel spend divided by the number of on-target enquiries. Counted only when there is real spend: if no budget was allocated the metric is marked “no data”, not zero.

source: ad account + spend log

Content health

The share of material that passed independent acceptance on the first attempt, plus a list of recurring discrepancies. This measures process quality rather than reach: it shows where a channel is systematically wrong.

source: QA verdicts in the task trail

Return into strategy

Which decisions changed as a result of the reading: a channel reinforced, reduced, closed, a hypothesis dropped. A metric without this step is a reference book — it changes nothing and gets repeated next cycle.

source: the AI CEO cycle decision log

Statuses

Honest matrix: the product and the ecosystem are different things

The most common failure of pages like this one is presenting an internal working tool as if it were a product capability. We keep two columns apart. First: what is implemented in the VOIR AI product (right now — nothing). Second: what already runs inside our Marisova ecosystem as our own tooling. The second column is not for sale and is not provided to customers.

Status of marketing department functions as of 2 November 2026. Verified against code, configuration and live URLs — not against plans.
Function In the VOIR AI product In the Marisova ecosystem (internal)
Content and copy not implemented
Target architecture: the content channel		 works
An SEO article pipeline: generation → human edit → publication in the news section of marisova.ru
Image generation not implemented
Target architecture: the visual channel		 works
An image generation studio inside the Telegram Mini App via an external API, with a credit balance
Video and editing not implemented
Target architecture: the visual channel		 works
A 9:16 vertical clip pipeline with subtitles and an AI narrator; a separate long-form film pipeline with quality control
Social analytics not implemented
Target architecture: the social channel		 works locally
A Reels analytics system for a single account: scheduled metric snapshots, dynamics reports, clip breakdowns. Localhost only, not a service
Lead generation and CRM not implemented
Target architecture: the email and lead channel		 dry-run
The GTM pipeline: research, qualification, a local CRM (209 lead records as of 02.11.2026 — data from our own test run, not customer results), email drafts. Real sending is disabled by a hard gate
Technical SEO audit not implemented
Target architecture: the search channel		 works locally
An SEO baseline: Lighthouse measurements across several sites, audits of dozens of URLs, meta tag control. An internal team tool
Advertising campaigns not implemented
Target architecture: the campaign channel		 limited
There is working access to an ad platform API in a separate project; no campaigns are run within VOIR AI and the paid budget in the configuration is zero
Publishing outward not implemented
By architecture — always through a human		 works
The rule is enforced in practice: external publications and sends happen only after the owner confirms

The works label refers to our internal tools that we use in our own projects. It is not a VOIR AI product feature and not a promise to a customer. The actual project status as a whole is published in the ship's log on the About the platform page; the platform itself is described on the AI Operating System page.

Link to the AI CEO

The department never assigns work to itself

Marketing in VOIR AI is the executive part of the AI CEO cycle. The AI CEO takes the task, decides which share of the work goes to marketing, and consolidates the result into the owner report. Department artefacts pass independent acceptance by a role that did not perform the work. The list below shows what the department does at each of the five steps.

  1. 01 Task setting The department receives the brief and the acceptance criteria fixed before the start: what counts as the result, which channels are involved, what the budget and tone constraints are. Marketing does not set its own task — otherwise the channel starts producing content for the sake of content.
  2. 02 Analysis and allocation Headquarters decides which of the six channels this task needs and at what volume, requests data from the research department, and fixes the autonomy grade: what the channel does on its own and what goes to the owner for confirmation.
  3. 03 Execution Channels work in parallel: content and visuals do not wait for the SEO audit, campaigns do not wait for the whole editorial plan. Each channel keeps a trail of actions and hands its artefact to the return bus together with a reading or a “no data” mark.
  4. 04 Independent verification The QA role checks every artefact against the step 1 criteria: consistency with positioning, a named source for each metric, no invented facts, compliance with the ban on external publication. The verdict is VERIFIED or REJECT with a list of discrepancies.
  5. 05 Acceptance and report Headquarters consolidates the verified artefacts into a single owner report: what was done, what was measured, which decisions stayed with a human, which hypotheses were dropped, what changes in the next cycle. Rejected artefacts and missed deadlines go into the report as they are.
  • Research → marketing Data about the audience, the market and competitors arrives with a source list. Marketing does not run its own research and does not fill in figures that are missing from the summary.
  • Marketing → sales Qualified leads and deal materials go to sales together with the communication history — without it the next step is taken blind.
  • Marketing → development Landing pages, UTM tagging, counters and forms go to development as requests with acceptance criteria, not as “make it look nice”.
  • Marketing → finance The campaign budget model and the spend forecast go to finance for review before launch; the channel never initiates spend without an agreed limit.

How a single executor works is on the AI agents page; how headquarters assembles several departments for one task is on the AI teams page; roles across departments are on the AI employees page.

Boundaries

What the marketing department never does

Marketing is the most tempting area to automate: it is easy to publish, to spend and to promise. That is why the constraints are built into the channel contracts rather than into settings — no channel can bypass them.

Does not publish outward on its own

A post, an email, an ad, a comment on behalf of the company — only after the owner confirms. Drafts move freely inside the department; nothing goes outward without a human.

Does not spend the budget

Launching a paid campaign, raising a bid, buying placement — these are financial decisions. The department prepares the model and the options; a human makes the call.

Does not promise results

No conversion forecasts, no growth guarantees, no “industry averages” without a source. When there is no data, the artefact carries a “no data” mark.

Does not invent customers

Testimonials, case studies, logos and rollout figures appear only when there is a real customer and their consent. There are none yet — so none appear on this page.

Does not accept its own work

The channel's self-check remains, but only the independent acceptance verdict goes into the report. An artefact that never reached QA does not exist for the report.

Does not break live assets for new ones

Edits to published pages, bulk meta tag changes and any action on a production site go through a backup and an additive approach. Removing what already exists is not part of the channel's authority.

Questions

What people ask about AI marketing

Is the AI marketing department working yet?

Not as a VOIR AI product. All six channels on this page describe the target architecture: none of them exists in the product code. The desk demonstration runs a prepared scenario and is not connected to a live AI service. Separately, the Marisova ecosystem already runs some marketing functions as our in-house tooling — an SEO article pipeline, Reels analytics, photo and video generation, a GTM pipeline in dry-run. They are not part of the VOIR AI product and are not offered to customers.

What does the status “Target architecture” mean?

It is an honest mark that a function is described in the project but not implemented in the product. We keep two statuses apart: “Target architecture” — how the department is designed to work in VOIR AI; and “works in the ecosystem” — a tool that already exists and is used in-house. Conflating them would be misleading: the second status never means a customer gets that capability.

Does the marketing department publish content on its own?

No, and this is an architectural constraint rather than a setting. Any external publication is escalated to the owner. In the Marisova ecosystem the rule is already enforced in practice: the GTM pipeline runs in dry-run, the real-send flag is off and protected as a hard gate, and articles are edited by a human before publication.

Which metrics does the department track?

Reach and impressions, clicks and depth, requests and their qualification, cost of acquisition, the share of artefacts that passed acceptance on the first attempt, and the decisions taken as a result of the reading. Every metric must name its source; when no source exists the mark is “no data”. Performance figures are absent from this page deliberately — there are no customers and no sales yet.

How is marketing connected to the AI CEO?

The department never assigns work to itself. The AI CEO takes the task and the criteria, decides the share of work for marketing, consolidates the result into the owner report and decides what changes in the next cycle. Department artefacts pass independent acceptance. Details are on the AI CEO and AI CEO Cycle pages.

Can the marketing department be used apart from the platform?

Architecturally it would not make sense. The channels run on shared platform data — tasks, audience, product record, communication history, decision log — and on the shared independent acceptance mechanism. A standalone “AI marketer” without the management cycle turns into a text generator with no measurement of consequences.

Where to go from this page

Marketing is one of eight departments: the AI CEO manages it, and agents do the work.

Management

AI CEO

Who takes the task, distributes it across departments and answers for the final owner report.

Open → Process

AI CEO Cycle

The five steps inside which the department gets the brief, works, passes acceptance and reports.

Open → People and roles

AI Employees

Roles across all eight departments: which of them serve the marketing channels.

Open →

See pricing Ask a question about the platform