Automatic research versus user-supplied sources

By Luis Rita · 15 October 2026

An AI draft can only be as accurate as what it reads. There are two ways to give it something to read: you supply the sources, or a search tool finds them from your brief. Neither is always better. Each fails in its own way, and knowing how is what lets you choose, or combine them.

The two approaches

• Sources you supply. You pick the pages, and only those are read. The draft rests on evidence you chose and would cite yourself.
• Automatic research. Search queries are written from your brief, a search service returns pages, and the best results become the sources. You get coverage without having to look.

How they compare

When to supply your own

Choose the sources yourself when:

• The facts belong to someone. A client’s products, prices, policies and terms should come from the client’s own pages, not from a reseller’s summary.
• Mistakes are expensive. For health, legal, financial and safety topics, start from official bodies, regulators and the primary documents.
• You already know the authority. If there is an obvious primary source, a search can only add noise around it.
• The detail is local. Timetables, opening hours and local rules are often best on small official sites that rank poorly.
• The knowledge isn’t online. Interviews, your own tests and internal data have to be supplied; a search won’t find them.

When automatic research helps

Let search find sources when:

• The topic is new to you, and you don’t yet know who the authorities are.
• You need breadth: several perspectives, or more examples than you have time to collect.
• Figures change, such as prices, statistics and rules, and the search can ask for this year’s.
• You’ve supplied a few pages and want the rest of the article covered.

The best of both

On most articles, combine them:

1. Seed the brief with the primary sources you know: the official page, the client’s page, the original study.
2. Let search fill the gaps around them, up to a sensible total; five to eight sources is plenty for one article.
3. Review the list before anything is written. This is where weak sources are cheapest to remove.
4. Check the claims after drafting against the sources that survived.

A source checklist

Before a page is allowed to support your article, ask:

• Who published it? An official body, the company itself, a recognised publication, or someone you can’t identify?
• Is it primary? Does it state the fact, or repeat someone else’s statement of it? Prefer the original.
• How old is it? Check the publication or update date, not just whether the page still loads.
• Is it on topic? Same product, place, period and audience as your article?
• Would you link to it yourself? If you wouldn’t cite it in your own name, it shouldn’t support a claim.

Drop aggregators, unsigned listicles and pages that exist mainly to rank. A shorter list of strong sources makes a better article than a long list of weak ones.

Where tooling helps

AI Blog Writer with Sources works from up to eight URLs you supply, fetched by your own server, and cites them inline. Pages published more than five years ago are marked as possibly stale, and in Guided mode the job pauses so you can read each source’s extracted notes and untick any before the outline is written.

AI Blog Writer Pro adds automatic research. It writes up to three search queries from your brief, aimed at primary sources and, when the topic depends on changing figures, at current information. It then searches with your own Tavily key, and adds the most relevant pages until the article has eight sources in total. Your own URLs always come first, and pages found this way are labelled found automatically, so you know which to check most closely. The free Tavily plan includes 1,000 credits a month, and an article uses about 3. Pro’s claim-to-source check then marks each factual claim in the draft as supported, weakly supported or unsupported by the sources that survived your review.