Does AI cite sources for its top brand recommendation?

An attached citation is a cited page linked to that brand in the same answer. AI attached at least one citation to 113,609 of 215,349 top-ranked brand recommendations, or 52.8%. The remaining 101,740, or 47.2%, had no citation attached to that brand in the same answer.

This refines the assumption that a top AI recommendation is automatically source-backed. Semrush's ghost-citation study starts with cited domains, while Ahrefs' brand-link study starts with brand mentions. This study starts with the top-ranked brand. An attachment also does not prove that the cited page supports the ranking or mentions the brand.

Google AI Mode attached a citation to 66,268 of 109,734 top-ranked recommendations, or 60.4%. ChatGPT Search did so for 47,341 of 105,615, or 44.8%. The difference was 15.6 percentage points.

Audit both engines separately. The combined rate hides a large difference, and the observed answers do not explain its cause.

The top-ranked brand had an attached citation in 113,609 of 215,349 recommendations, or 52.8%. The rate was 42.0% for ranks two and three, with 156,382 of 372,215, and 30.9% for rank four or lower, with 176,122 of 570,048.

Audit the full shortlist, not only the winner. Lower-ranked brands were less likely to have an attached citation in this window. The pattern does not show that recommendation rank caused the citation difference.

Answers naming one to three brands attached a citation to the top pick in 28,381 of 49,654 cases, or 57.2%. The rate was 54.4% for answers naming four to six brands and 45.8% for answers naming seven or more.

Compare citation attachment within similar answer lengths. A long recommendation list is a different review job from a short shortlist. This descriptive result does not show that list length caused the lower rate.

Among the 113,609 top-ranked recommendations with an attached citation, 47,211, or 41.6%, had one attached source. Another 32,629, or 28.7%, had two or three, and 33,769, or 29.7%, had four or more. The median was two attached sources and the average was 2.9196.

A single citation count does not describe the evidence available around a recommendation. Record how many sources are attached, then inspect whether those pages independently support the brand claim.

Among industries with at least 1,000 top-ranked recommendations, the rate ranged from 57.7% for Sustainability, with 578 of 1,002, to 49.1% for Collaboration, with 529 of 1,077. Professional Services reached 57.6%, Transportation 57.6%, Information Technology 51.2%, and Apps 51.1%.

The headline result was not confined to one market. Industry remains useful context, but the observed range is too narrow to treat category alone as an explanation for whether a winner has an attached citation.

LinkedIn was the top-ranked brand 2,965 times and had an attached citation in 1,847 cases, or 62.3%. Datadog followed with 2,322 top recommendations and a 44.7% attachment rate. Alphabet had 2,309 and a 41.5% rate.

Recommendation volume and citation attachment are separate measures. Use the leaderboard to find high-volume brands for source review, not to infer brand quality or recommendation accuracy.

Exact brand-name matching attached citations to 112,585 of 215,349 top recommendations, or 52.3%. Merging parent and sub-brand names recovered 1,024 matches and produced the 52.8% headline, a 0.48-point change. The June 1 through July 16 sensitivity cut returned 52.8%.

The study merges related brand names before matching citations and excludes one answer with two different top-ranked brands. The stable sensitivity check reduces the chance that the opening partial week or name variants determine the headline.

Record the top-ranked brand and whether a source is attached to that same brand. Keep the rate separate by engine and recommendation rank. Review longer shortlists independently because their winners had a lower attachment rate in this window.

For high-volume recommendations, inspect the cited pages for brand mention, claim support, and source ownership. Rerun the fixed-window study next quarter with the same brand-merging and ambiguity rules. Do not treat an attached citation as proof that the recommendation is accurate.