How Blogo finds official sources

The first search asks for the pages that state the facts first-hand: the official site, documentation, help centre or pricing page of the product or organisation the topic is about; regulators, government and standards bodies; original data and research. It is told not to list blogs, buying guides, listicles or review sites that are themselves trying to rank for the same search.

Blogo keeps up to eight candidates, one per website. Each is marked as an authority or not. Government and education domains and a fixed list of international bodies and major publications are marked straight away.

If fewer than three candidates are marked as authorities, a model is asked to name up to four organisations whose own website states the facts: the company that makes or runs the product, the regulator, the standards body, the institution behind the original data. It is asked for the most specific site, such as a help centre rather than a corporate home page.

The model only supplies names and domains. It is not allowed to supply page addresses. For each domain, Blogo runs a search restricted to that site, and every page address comes from the search results.

Being on the right website is not enough. Example: a payment company’s developer reference sits on its official domain but is no use to a reader who wants to set up an account.

So the pages found are shown to a model as a numbered list, and it is asked which ones state facts this reader needs: rules, limits, steps, requirements, definitions, test methods. It answers with numbers from the list and cannot add a page. At most six are kept, and they go to the top of the candidate list.

If there are still fewer than three authorities, the remaining candidates from pass one get one more look. A model judges, site by site, whether each is a citable first-party or institutional source. This can promote a candidate but never demote one, and a known competitor is never promoted.

The candidates are shown at the References step. You can choose which to use; if you do not, Blogo takes up to five, authorities first.

Only chosen pages that are marked as authorities are offered to the writer as sources it may link. Blogo then reads those pages and copies out sentences that contain a figure and mention the topic, so that any figure the writer borrows already has its source attached.

If none of the chosen pages qualifies, Blogo falls back to pages on recognised authority sites that its own research notes cited, keeping only those that still load. If that also yields nothing, the list stays empty.

If the model names the wrong domain, the site search returns nothing. If the authority publishes its rules only as PDFs, they are skipped. If the site search is unavailable, the second pass yields no pages at all. And many product topics have no regulator or standards body behind them.

When no approved source exists, Blogo does not invent one. The writer is told to use no figures except yours, the draft is not held merely for having no outside link, and the page stands on your own answers. If the draft links only weaker pages, the final review suggests a stronger source.