Methodology

Prices are not generated by AI. They are fetched from third-party market-data providers by scheduled sync jobs and stored with the time of the last update. Each asset page shows the latest stored value.

Quotes may be delayed by the provider (typically up to 15 minutes for exchange-listed securities). Forex reference rates are published once per business day by the ECB and are not live dealing prices. Fund NAVs are published by the fund manager, usually once per day. When a provider stops returning data for an asset, the page keeps the last known value; pages whose price is missing or stale are excluded from search indexing (see section 5). The full provider list is on our Data Sources page.

Every few minutes an editorial pipeline collects headlines from publisher feeds and news APIs, removes duplicates (same URL or near-identical headline), and scores each story for market relevance. Selected stories are rewritten into an article by a large language model (Google Gemini), using the original source text and our stored market data as its only reference material.

Before an article is published in full, the draft must pass automated guards:

• Numbers must exist in the sources. Every figure in the draft (prices, percentages, amounts) is extracted and matched against the source article and the market data supplied to the model. Formatting differences and rounding are tolerated; a figure that appears nowhere in the sources is a violation.
• No invented quotes or attributions. Quoted speech must appear verbatim in the source, and statements attributed to people or firms ("X analysts said") may only name people or firms that appear in the source text.
• Date checks. A draft whose dominant year is in the past is rejected unless the source itself is about that year, so old stories are not presented as new.
• Sentiment sanity check. A bullish/bearish label that flatly contradicts the headline (e.g. "shares plunge" tagged bullish) is corrected.

A draft that fails is sent back to the model once with the list of violations. If it still fails, the full article is not published: only a short summary with a link to the original source goes live, marked noindex so it does not appear in search results. Every article lists its sources at the end and links to the original reporting.

Each show (for example Crypto Pulse, Forex Focus, Stock Spotlight) runs on a fixed schedule. At air time the system pulls the current market data for that show, and a language model writes the narration and storyboard. The prompt for every public show carries the same hard rule: use only the numbers in the data provided, skip anything with no data, and never invent figures, quotes, ratings or price levels. Shows are written to explain market moves, not to recommend trades.

The narration is voiced with an AI text-to-speech voice; charts and figures are rendered from the same data, and some backgrounds use licensed stock footage (Pexels). The full narration is published as a transcript under every video.

After rendering, an automated quality gate inspects the output file: a video that is unreadable, shorter than 60% of its target length, missing its audio track, or too small to be a valid render is marked failed and never published.

Articles and videos are published automatically; they are not reviewed individually by a person before publication. Human oversight happens around the pipeline: the rawfin.tv team (operated by 1B RACE SAS) writes and maintains the prompts, data rules and fact-guards described above, reviews published videos and logs issues against them, monitors failed and rejected drafts, and handles correction requests.

Found an error? Email [email protected] with the page URL and what is wrong. Confirmed errors are corrected or the content is unpublished. See the corrections policy.

Not every automatically generated page deserves to rank. We only ask search engines to index pages with substantive content: major cryptocurrencies and listed companies with a current price, major and G10 currency pairs, benchmark commodities and sovereign yields, funds with a recent NAV, full-length news articles that passed the guards, and video episodes of at least 45 seconds with a complete transcript. Other pages remain available to readers but are marked noindex.

• AI-written text can still misread a source or describe a move imprecisely even when every number matches.
• The number check confirms a figure exists in the sources, not that the source itself is correct.
• Data can be delayed, interrupted or revised by the provider; a video always reflects prices at its air time.
• Coverage is broad but not complete: some assets have limited history, no market cap or infrequent quotes.
• Voice-overs are synthetic; pronunciation of tickers and names can be imperfect.

Nothing on rawfin.tv is financial, investment or tax advice, or a recommendation to buy or sell any asset. Do your own research and consult a qualified, licensed adviser before making investment decisions.