Source Evaluation in the AI Era

In 2016, a study found that 82% of middle school students couldn't distinguish a real news article from a sponsored content piece. By 2025, the problem had mutated: it was no longer just about distinguishing news from advertising. It was about distinguishing human-written content from AI-generated content designed to look authoritative, complete, and sourced.

The old source evaluation playbook — check the domain, look for an author, see if it has citations — still applies. But it's no longer sufficient.

What AI Has Changed About Source Evaluation

Before AI, producing large volumes of authoritative-looking content required significant effort. A fake research site needed writers, editors, and time. This created a natural friction that limited the scale of content manipulation.

That friction is gone. A single person with access to a capable AI model can now produce hundreds of well-formatted, citation-rich, authoritative-sounding articles in a day. Many of those articles will pass the old surface-level checks: they have authors, they have citations, they have consistent formatting, they read fluently.

The citations may not exist. The author may not exist. The institution in the byline may be a domain registered last week.

This doesn't make source evaluation impossible. It makes it more specific. You need to look for different signals.

Three Signals That Still Hold Up

Provenance — where did this originate?

Don't evaluate the source you're looking at. Trace the claim upstream to its origin. If an article cites a study, find the study. If a site references an expert, find the expert's actual profile. Provenance asks: where did this claim first appear, and is that origin credible? AI-generated content typically breaks down at this step because fabricated citations don't trace to anything real.

Track record — has this source been wrong before, publicly?

A source that has made verifiable claims over time — and been held accountable for errors — is more trustworthy than one that has no history. This is why academic journals, established newsrooms, and government statistical agencies carry more weight than newly created sites. They have track records. AI-generated sites typically don't.

Independence — who benefits if you believe this?

Every source has a position. The question is whether that position is disclosed and whether the source has something to lose by being wrong. A pharmaceutical company's press release about its own drug is not independent. A peer-reviewed study with disclosed funding and methodology is held to a different standard. Ask who funded the research, who published it, and who benefits from the conclusion.

A Practical 90-Second Check

When a specific claim matters enough to verify, run this sequence:

First, search for the claim and the source independently — don't just evaluate the article you're reading. Second, check whether the cited study or expert appears in independent databases (PubMed, Google Scholar, institutional directories). Third, look for coverage by outlets with different known positions — if only one type of source is reporting something, that's a flag.

None of this takes more than 90 seconds for most claims. The ones that can't be verified in 90 seconds are the ones worth spending five minutes on.