An editorial team that starts with the sources
Pennao combines finding publications, writing and quality control in one process. You see every step, can stop it and correct it, and the finished text goes straight to your system.
13 stages of one article
Each stage has its own model, chain of fallback models and settings. Stages that do not depend on each other run in parallel.
-
01
Brief and queries
The keyword, intent and site profile turn into queries for scholarly databases, search engines and news, also in other languages.
-
02
Source discovery
OpenAlex, Semantic Scholar, Crossref, PubMed, Europe PMC, arXiv, CORE, DOAJ and web search. Duplicate URLs and DOIs are merged.
-
03
SERP analysis
Headings of the top ten pages and questions from “People also ask”. The plan answers those questions better than the competition.
-
04
Source ranking
A score from 0 to 100: source type, citations per year, recency, relevance, open access and agreement between providers.
-
05
Fetching content
Pages and PDF files, with protection against private addresses. When the full text is unavailable, the abstract remains citable material.
-
06
Fact extraction
Facts, figures, definitions and quotes from every source go into the project knowledge base, each with its exact origin.
-
07
SEO plan and critic
The planner lays out headings for the keyword and intent, the critic checks topic coverage. You can approve or edit the plan before writing.
-
08
Writing sections
Sections are written in parallel with the context of their neighbours, and a separate pass smooths the transitions between them.
-
09
Inline links
Source links land on the exact words. Plus internal links to your pages and the required links from the brief.
-
10
Validation and fixes
LanguageTool, fact-checking, the AI-slop filter, an SEO linter and readability. Text below the threshold goes back for fixes.
-
11
Images
Graphics from an image model or photos from Openverse, Wikimedia Commons, Unsplash and Pexels, with author credit and license.
-
12
Translations
Each language has its own keywords and goes through the same check. Versions are linked with hreflang tags.
-
13
Assembly and publication
Markdown, HTML or JSON with JSON-LD, WebP images in several sizes and a webhook to your system.
-
The whole process on one timeline
The dashboard shows the time, cost and model of every stage, and when a fallback model takes over, the reason. A typical 1,500-word article is ready in 10–15 minutes.
See an example
Pennao knows your site before it writes the first sentence
When you add a project, an agent reads your site: offer, tone, audience and pages. The resulting profile goes into every order.
- Internal links to your pages, matched to each section
- Products and required headings set once for the whole project
- Domains in several languages, each with its own page index
- Preferred and blocked sources, for example competitor sites
Fix a passage, not the whole text
Select a sentence or paragraph and choose an action: rewrite, shorten, expand, add a source or write your own instruction. The model changes only the selection.
- A diff where you accept each change separately
- A Markdown editor and version history with authors
- Team comments on specific passages
- After every change the text goes through the quality check again
Works out of the box, scales with your keys
Language models need your own key. Everything else works without setup, and when you need higher limits you connect your own account or server in Settings → Services.
Your own servers need a public HTTPS address. A server behind Cloudflare Access connects with a service token. Test connection sends one test request and saves nothing.
One brief, ten language versions
The first language is the main one, the others are translations with their own keywords. The AI-slop filter and style rules have separate dictionaries for English, Polish, German, French and Spanish.
- EN English
- PL Polish
- DE German
- FR French
- ES Spanish
- IT Italian
- PT Portuguese
- NL Dutch
- CS Czech
- UK Ukrainian
Read your first article today
Enter a keyword, add a model key and read a sourced article in about fifteen minutes.