TL;DR: The Media Comment Generator is a 49-node n8n pipeline that produces voice-matched expert media comments in under 60 seconds. It uses Yandex Search MCP for Russian-market research, Claude AI for voice-calibrated writing, and a dedicated humanization pass to remove 24 types of AI markers. PR teams save 10+ hours per week, with each comment matching the expert's actual vocabulary and reasoning style — no template patterns, no recognizable AI tone.
Marina is a PR manager at a mid-size branding agency. Tuesday, 2 PM — three hours until deadline. A journalist from Forbes Russia sends a Telegram message: "Need three expert quotes by 4 PM. Topic: how AI is reshaping marketing strategies in 2026."
Previously, this meant: open five browser tabs with research, draft a response, ask the spokesperson to review it, get it back with edits, finalize, send. A minimum of two hours. Today, Marina opens the bot, pastes the article link and the question — and 55 seconds later receives a finished expert comment in the first person, written in the right tone, with current data, and without any telltale AI markers.
This is not a demo. This is the daily reality of agencies already running our media comment generator.
What Is an AI Media Comment Generator?
An AI media comment generator is a pipeline that transforms an article link and a journalist's question into a fully publication-ready expert comment. Unlike generic AI writing tools (Jasper, Copy.ai), our system understands the context of the Russian media market: publication format, editorial tone, typical comment length for specific sections.
The system does not simply generate text. It:
- finds and analyzes the original article from the link
- researches the topic within the Russian-language information landscape
- generates the comment in the voice of a specific spokesperson
- removes template AI patterns, leaving a natural, recognizable style
- validates the result against formal criteria (length, structure, absence of filler)
The output is a comment that cannot be distinguished from one written by a professional PR consultant.
How Our Pipeline Works: 49 Nodes from Request to Publication
The system architecture runs on n8n — an open-source workflow automation platform. We chose n8n because it lets us visually design complex pipelines with multiple branches, without requiring the dev team to write custom code for every integration.
Here is the full processing cycle for a single request:
1. Routing and parsing (nodes 1–6). The Telegram bot receives a message in the format "link + question." The system validates fields, loads the spokesperson's profile from GitHub (title, expertise, typical phrasing, a do-not-say list).
2. Research (nodes 7–14). The Research Agent, powered by Claude Sonnet 4.5, runs two parallel processes: full article extraction via Tavily and supplementary source research via Yandex Search MCP. This is critical for the Russian market — Yandex surfaces results that are relevant specifically to the Russian-language information space.
3. Comment generation (nodes 15–24). The Comment Generator Agent receives the research output and the spokesperson profile. It produces a structured comment: context analysis, the comment itself (1,200–2,000 characters), and a self-check against quality criteria. Every comment is generated in the unique voice of the spokesperson — with their typical constructions, set phrases, and professional register.
4. Humanization (nodes 25–34). The Humanizer Agent, powered by GPT-4o, applies our v2.11 methodology: it removes 24 types of AI markers (template phrases, excessive politeness, language-model-typical constructions), adds natural turns of phrase, rhythmic variation, and a recognizable personal stance. This stage is what transforms "good text" into "human text."
5. Quality control and delivery (nodes 35–49). Automated Compliance Review checks text length, presence of Cyrillic, absence of AI markers, and question relevance. If something fails — a retry loop triggers a regeneration. The final result is sent to Telegram, and a callback is dispatched to the web app for status tracking.
The entire process takes 30 to 90 seconds. In 95% of cases — under 60 seconds.
Key Features That Set Us Apart from Generic AI Writers
The market has no shortage of AI text generation tools. Jasper, Copy.ai, any ChatGPT bot can "write a comment." But there are three critical differences that determine whether the output will be published or end up in the "not suitable" folder.
1. Russian media source integration. Generic AI does not know that Kommersant expects concise, first-person quotes, while Secret Firmy prefers a more conversational tone with examples. Yandex Search MCP in our pipeline finds exactly the sources that your future competitors and peers are citing.
2. Telegram approval workflow. After generation, the comment arrives in Telegram. The PR manager sees the text, can request a revision (iteration via the bot) or send it straight to the spokesperson. No need to switch between web interfaces. One environment, one process.
3. Unlimited feedback loops. We do not cap the number of iterations. If the first comment does not "land" — send "make it more critical" or "add a specific example from your practice." The system accounts for the context of previous iterations and refines the generation. In practice, most comments pass on the first attempt; 20% require one revision.
Real Example: From Media Request to Published Comment in 55 Seconds
Here is a real case from our agency's practice.
Situation: A journalist from SportBusiness sent a query: link to an article on sponsorship contract trends in Russian sports, question — "How have brands' criteria for selecting sponsors changed in 2025?"
What used to happen (manual process):
- PR manager searches for the article, opens 3–4 sources on the topic (15 minutes)
- writes a draft, tailoring it to the publication format (30 minutes)
- sends to spokesperson for approval, receives edits (20–40 minutes)
- finalizes and sends to the editorial office (5 minutes)
Total: 1.5–2 hours. And the deadline on a media query is often 2–3 hours.
What happens now (with the generator):
- PR manager sends to the bot: "LINK: [link] / QUESTION: How have brands' criteria for selecting sponsors changed in 2025?"
- 55 seconds later, a ready comment arrives in Telegram
- PR manager checks it, makes one revision ("add an example with a hockey club"), sends a follow-up request
- 40 seconds later, the final version arrives and gets sent to the editorial office
Total: 8 minutes. Of which 5 are spent checking and one round of revision.
The comment was published the next day with the spokesperson's title and company affiliation.
Cost Comparison: AI Comments vs. Junior Staff vs. Manual Work
Let us calculate honestly, without marketing fluff.
| Method | Time per comment | Cost (estimate) | Quality |
|---|---|---|---|
| Manual work (PR manager) | 60–120 min | $20–40 / comment | High, skill-dependent |
| Junior content (outsource) | 40–90 min | $10–20 / comment | Medium, requires review |
| Generic AI (ChatGPT, Jasper) | 15–20 min | ~$3 (tokens) | Low for Russian media, needs rework |
| DDVB Tech Generator | 1–3 min | $5–10 / comment | Publication-ready, in spokesperson's voice |
The key difference is not just speed — it is the "last mile": generic AI produces text that needs to be largely rewritten. Our generator produces text that needs only to be checked.
Step-by-Step Workflow for Your Agency
If you are a PR director or agency owner looking to implement the system:
Step 1. Define your spokesperson list. Create profiles for 3–5 key experts at your agency: title, expertise, typical phrasing, topics the spokesperson does not comment on.
Step 2. Connect the Telegram bot. The bot is your team's interface. No need to retrain on new software; PR managers already live in Telegram.
Step 3. Configure the profiles. For each spokesperson — a separate JSON profile with context. This is the foundation of voice replication.
Step 4. Test on 10 comments. Take real media queries from the past month and generate comments for them. Compare with what you sent manually.
Step 5. Connect to your web app. If your agency has multiple PR managers and you need analytics — connect the webhook to your CRM or task tracker.
Best Practices for Publication-Ready Comments
After two years of operation and hundreds of generated comments, we have developed a few rules.
Rule 1: The more specific the question, the better the result. "What do you think about industry trends?" is a vague query, produces a vague comment. "How has the average B2B sponsorship contract duration changed in 2025?" is a precise query, produces a precise comment.
Rule 2: One comment — one thought. Journalists typically quote 2–3 sentences. If your comment is a stream of consciousness, the editor will cut the most valuable part. Formula: one thesis + one example + one assessment.
Rule 3: Numbers and names increase citability. Comments with specific figures ("a 23% growth year-over-year") and proper nouns ("as the Spartak team did with their 2024 rebrand") are cited 2.5× more often than abstract reflections.
Rule 4: Reflect the editorial line. Before sending a comment to a media outlet, check what the publication has already covered on the topic. A comment that continues or challenges the position of a previous piece is always more interesting than a neutral expert response.
Frequently Asked Questions
Will this replace our PR director?
No. The system is a tool for the PR team, not a replacement for it. The PR director remains the strategist: deciding which spokesperson comments on what, how to position the company, how to build long-term relationships with editorial offices. The generator handles the routine — daily operational responses to media queries.
Will the comment look like typical AI text?
No. This is precisely why the pipeline has two separate stages: generation and humanization. The Humanizer Agent is trained to remove 24 categories of AI markers. After humanization, the text goes through an automated AI-detection check. In our internal stats — 91% of comments pass compliance on the first iteration.
Can the system learn from our published comments?
Yes. This is one of the customization steps: we load 15–20 previously published comments per spokesperson and configure the profile. The system learns not just the language, but the logic of argumentation — what examples the spokesperson uses, what constructions they favor, what topics they elaborate on in more detail.
How is confidentiality handled?
All data passes through our n8n instance on Yandex Cloud. On request — deployment on your own server is possible (sovereign AI infrastructure). Spokesperson profiles are stored encrypted.
What publications are supported?
Any. The system does not work from a publication database — it analyzes each specific article from the link. Supported: Sostav, Kommersant, Forbes Russia, Vedomosti, RBC, Secret Firmy, SportBusiness, and other federal outlets.
Want to see the system in action on real media queries?
Book a 15-minute demo — we will show the full cycle: from a journalist's request to a published comment.
Or start with one comment — free, no strings attached. Write to the Telegram bot and test the result on your next media query.
