Synthiz for journalists
Synthiz transcribes your interviews, podcasts and videos, summarises reports and articles, then links concepts across sources. Its search finds the exact source behind a claim before deadline, with a timestamped quote.
The real bottleneck is reading in, not writing out
Filing a 1,200-word piece takes two or three hours. What comes before it takes twenty: two hours of interview to replay, a 180-page report, a three-hour committee hearing, fifteen competing articles. That intake work is the actual job — and almost all of it is mechanical.
The existing tools split the problem without solving it. A transcription service hands back a wall of undifferentiated text. A note manager files what you have read without reading it for you. A general chatbot summarises fast and occasionally invents a quote, which is disqualifying in a newsroom.
The cost is not only lost hours. It is the annexe report skimmed, the hearing half-watched, the contradiction between two sources that nobody caught.
The workflow with Synthiz
- Create one memoire per investigation — "Nursing home funding", "City council election" — each with its own sources, synthesis and generation language.
- Put everything in. A link is enough for an article, a YouTube video or a podcast; PDFs upload directly. The Chrome extension saves the open page in one click and can capture the transcript of YouTube videos you watch.
- Check without leaving the tab. You are reading a department press release and a sentence rings a bell. The extension's search panel queries your own base from the current page and shows the surrounding excerpt: you see immediately whether you already hold a source that says the opposite.
- Subscribe to recurring sources: an agency's RSS feed, a committee's YouTube channel, a watchdog's podcast. Sync runs on its own — hourly, every six hours or daily — and content arrives already processed.
- Read the summary, then drop to the source. Every concept carries timestamped citations: one click puts you in the transcript to the second, and that is the sentence you write down.
- Cross-analyse the sources. The cross-analysis surfaces convergences, tensions and blind spots — two experts contradicting each other, a point nobody raises.
- Write thoughts — an objection, a lead, a contact to call back — anchored to their sources. That is the draft of your angle, not machine output.
- Export to Markdown or PDF into your writing tool at deadline.
The "I know I read that somewhere" moment
It is 6 p.m., the piece files at 7, and a claim has no source attached. You know a speaker gave that figure — hearing, report or podcast, you no longer remember which, and their exact wording escapes you.
This is where most note systems give up. Obsidian, a Zettelkasten, any second brain: excellent for thinking and linking ideas, weak on retrieval. Finding the source of an idea six months later means guessing the term you used at the time, or having kept a tag discipline nobody actually keeps.
Synthiz keeps the linking and treats retrieval as a separate problem. Search is hybrid: a semantic layer that finds a passage by meaning even without the document's own words, and a full-text lexical layer that catches proper nouns, acronyms and exact figures. The two scores are fused.
The detail that matters is how relevance is decided. There is no fixed threshold: Synthiz measures a document's distance from the distribution of its own query, plus an absolute floor. On a corpus of 948 embeddings, a query with 5 relevant documents and a query matching nothing peak at 0.09 apart — a fixed 0.65 threshold returned nothing to someone who did in fact hold 5 sources. At 6 p.m.: you find the source without recalling its phrasing, and an empty corpus returns zero rather than three adjacent documents.
What it changes in practice
| Step | Before | With Synthiz |
|---|---|---|
| One-hour interview | Full replay, 90 min to 2 h | Transcript and concepts ready, targeted checking |
| 150-page report | Skimmed, points missed | Structured summary, jump back to the exact passage |
| Institutional monitoring | Manual site checks | Synced subscriptions, already processed |
| Finding a quote | Scrubbing through audio files | Search across the corpus, timestamped to the second |
| Contradiction between sources | Caught by luck | Surfaced by the cross-analysis |
The net gain is not writing speed. It is the volume of material you actually get through: the annexe report you would have ignored for lack of time now makes it into the corpus.
The honest limits
It is worth being blunt about what Synthiz does not do for a newsroom.
It is not a source of truth. Summaries are generated by language models. Traceability exists precisely because they are not self-sufficient: nothing goes to publication without being re-checked in the original document.
It is not a tool for sensitive sources. Content is processed by third-party AI services. For protected material — a whistleblower, a confidential document, an off-the-record interview — use SecureDrop, Signal and offline handling.
It is not a collaborative writing tool. No real-time co-writing, no sub-editing workflow, no article versioning: writing stays in your CMS.
Transcription is not perfect. Noisy field recordings, speakers talking over each other, strong accents, technical jargon: errors are real. It speeds up locating a passage; it does not replace your ear on a line you intend to quote.
Nothing for data journalism. Analysing a procurement spreadsheet or joining datasets remains the job of a spreadsheet, Python or OpenRefine: Synthiz handles text, audio and video.
No evidentiary archiving. No legal timestamping, no admissible capture of a page that might disappear: keep your usual screenshots and web archives.
Frequently asked questions
- Can Synthiz transcribe an interview I recorded on my phone?
- Synthiz handles video and audio files you upload, plus podcasts and YouTube videos from a plain link. A personal recording goes in like any other source and comes back with a transcript and a summary.
- Can I quote directly from what Synthiz produces?
- No, and that is not the intended use. Summaries are a triage layer: they take you to the right minute of the transcript, and it is that original quote you verify and put in your copy.
- How do I keep several investigations apart?
- Each investigation becomes a memoire — a topic space with its own sources and its own synthesis. Anything you capture without knowing where it belongs lands in Capture, the default memoire, and you file it later.
Published 2026-09-05