Deep Research: Real Web Sources Inside Every Article

Most AI writers either ignore the web or skim a few snippets and hallucinate the rest. Deep Research is KoalaWriter's mode for producing articles grounded in real, current sources, and it runs by default on every section, every time.

How Deep Research works

1. SERP fetch per section: at outline time, KoalaWriter queries Google for the section's topic and pulls back the top results with titles, snippets, and metadata.
2. LLM-assisted URL selection: a small model picks up to 5 URLs per section that are most likely to contain useful primary information.
3. Source scraping: each selected URL is fetched and its readable text extracted (limited to 2,000 tokens per source to keep context manageable).
4. Research caching: sources are written to KV (1-day TTL) so the outline step and writing step share the same evidence.
5. Inline citation: during writing, the model is instructed to cite sources with keyword-rich anchor text in markdown. No footnotes, no endnotes.

Why this matters

An AI article without research is recycled training data plus filler. Deep Research changes the shape of the prompt: the model isn't being asked to recall facts, it's being asked to synthesize specific source text. The output reads less like AI slop and more like a writer who actually went and looked things up.

Real-time data sources

You can switch the research mode per article:

• Web (default): general Google search.
• Google Scholar: academic + research-paper sources.
• Google News: recent news sources.
• Custom: use advanced search operators (site:, filetype:, date ranges, etc.) to constrain sources.

Combining Deep Research with Brand DNA

For brand-attached articles, the writer combines two layers: Deep Research provides external evidence (industry studies, news, competitor pages), while Brand DNA provides internal voice and credentials (services, FAQs, brand-specific framing). The result is content that's both factually grounded and authentically yours.

Citation format

KoalaWriter forces inline markdown citations. No footnotes, no endnotes, no separate reference section. Anchor text is keyword-rich so citations double as internal SEO signals. You can toggle citation behavior via the shouldCiteSources flag.