How a Sitario blog article is written
An article from the blog add-on isn't a language model dumping text from memory. Before the first sentence is written, the Sitario content engine does real research: it searches the web at generation time, reads the pages it finds in full and discards the ones that add nothing. Only then does it write — which is why the text contains specifics, not filler.
Below is the whole process, stage by stage — in exactly the order you watch it in the panel.
5 stages — from topic to draft #
The samples below come from generating an article on "heat pump running costs in 2026" — this is what the stages look like for a typical commission.
-
Stage 1/5
Topic analysis and query building
The system breaks your topic and key phrases down into the concrete questions the article should answer, then turns them into real search queries — several variants, the way a researcher would. One generic query returns random results; several precise ones return material you can actually write from.
From the panel
heat pump running costs 2026 detached house heat pump subsidies 2026 eligibility heat pump vs gas boiler operating costs -
Stage 2/5
Live web search
The queries hit the web at generation time, so the results come from today's internet — not from model knowledge frozen at a training date. Prices, regulations and figures in the article have a chance of being current, because the system finds them in fresh publications.
From the panel
searching live
energy-costs-guide.com heating-advisor.net installer-magazine.com home-renovation-forum.net -
Stage 3/5
Reading sources in full
The system fetches the pages it found and reads them whole — not just the titles and snippets from the results list. It sees the full context: tables, figures, caveats and definitions that never make it into summaries, and builds its working material from that.
From the panel
- energy-costs-guide.com/heat-pump-costs read
- heating-advisor.net/subsidies-2026 read
- installer-magazine.com/pump-vs-gas reading…
-
Stage 4/5
Source selection
Not everything that can be found deserves to be a source. Thin pages, duplicates and purely promotional texts are dropped — only the most substantive sources make it into the material. Three good sources beat ten random ones.
From the panel
- energy-costs-guide.com — cost data, tables
- heating-advisor.net — subsidy conditions
- business-directory.net — purely promotional page
- news-aggregator.net — copy of another site's article
2 of 4 pages found make it into the material
-
Stage 5/5
Writing and SEO polish
Only now is the text written — sentence by sentence, on the collected material. At the end the article gets its SEO polish: headings arranged around your phrases, a meta title and description, and a structure that reads well for both people and search engines. The finished draft lands in your Blog tab and waits for your edit — nothing publishes itself.
From the panel
SEO polish- H2/H3 headings around your phrases
- Meta title and description
- Phrases placed naturally in the text
What you see in the panel #
Generation is not a black box. On the article's page you watch, live, which stage is running: completed steps tick off one by one, and below them the list of sources grows — with the addresses of the pages the system is reading.
You can close the tab and come back later — the process runs server-side. When you return, you'll see the current state or the finished draft in the Blog tab.
Generating the article — live view
- Topic analysis and query building
- Searching the web
- Reading sources (2/3)
- Source selection
- Writing and SEO polish
Sources found (4)
- energy-costs-guide.com
- heating-advisor.net
- installer-magazine.com
- home-renovation-forum.net
Why this produces a better text #
Fresh data
Research happens at generation time, so the article is built on what's on the web today — not on a state of knowledge from two years ago.
Specifics instead of filler
The text is written on source material, so it contains what readers are looking for: figures, examples and answers to concrete questions.
You stay in control
The result is a draft, not a publication. You read it, edit whatever you like and decide yourself when the article goes live.
Common questions #
How long does an article take?
Usually a few minutes. The exact time depends on the text length and the number of sources the system has to read — you can watch the progress in the panel the whole time.
Is the text unique?
Yes. The pages found serve as factual material, not as text to rework — the article is written from scratch, around your topic and phrases. The system doesn't copy paragraphs from sources.
What happens to my character pool if generation fails?
The characters go back to the pool. If the process doesn't finish because of an error, you lose nothing — your pool only shrinks for articles that were actually produced.
Can I provide my own guidelines and phrases?
Yes. In the commission you give the topic, the key phrases you care about and the text length — the system arranges the research and the article's structure around them. You fine-tune the rest while editing the draft.
Commission your first article
Enable the blog add-on in the panel — you'll usually see the first draft within a few minutes.
20,000 characters a month on the PLN 49/month subscription, or a one-off 20,000-character pack for PLN 69.