ChatGPT now flags unverified vendor claims six times more often after GPT-6 switch

David Konitzny, writing on LinkedIn on 14 October, reported that ChatGPT started hedging citations from listicles and comparison pages soon after GPT-6 became the default on 7 October. His team analysed close to 3 million chats that cited this type of source.

Two patterns stood out. Warnings that a ranking had not been independently verified rose from 0.56% to 3.46% of chats, about six times the earlier rate. Warnings about vendor-supplied performance numbers rose from 1.01% to 2.27%, more than double. Both jumps landed within two days of the switch and have held since.

Konitzny cites sample model replies. One reads: "This is not an independently verified ranking." Another: "These case studies were published by the company itself, which has an interest in promoting its products." The model is naming the commercial interest behind the source in the same breath as the citation.

What the warnings do to a citation

Listicle placement still pulls brands into ChatGPT answers. Konitzny is clear that getting featured helps a brand get cited. The change sits in what the model says around the citation when the source is a ranking or a vendor claim.

A reader who sees the brand name alongside a line about unverified vendor data is reading a different endorsement than one who sees the brand name alone. The citation count on a dashboard will not show that difference.

How to check this on your own answers

Pull a sample of ChatGPT answers that cite you, from queries you care about. Read the full reply around each citation. Mark the ones where the model adds a hedge about independence or commercial interest, and whether that hedge is attached to your name or a competitor's.

If the hedges cluster on sources you control, such as your own case studies, own benchmarks or sponsored listicles, that is the signal to act on. Independent reviews, third-party benchmarks and verifiable data carry the citation without the warning attached. Konitzny makes that point directly, and the numbers in his sample support it.

We think the useful read here is not that listicles stopped working. It is that the credibility layer around a citation is now measurable in the model's own words, and brands relying mostly on self-published proof will see that reflected in how they are introduced.