Does being cited in AI answers mean being recommended?

A cited pair is one brand in one AI answer with at least one cited page attributed to that brand. The pick is the brand the same answer ranks as its #1 recommendation. In 260,895 of 1,189,703 cited pairs, or 21.9%, the cited brand was the pick. That is about 1 in 5.

The other 928,808 cited pairs, or 78.1%, ended without the pick. Semrush's ghost-citations study showed that a citation often arrives without a brand mention. This result measures the next step: even a citation attributed to a named brand usually does not make that brand the recommendation.

In 1,181,943 of 1,189,703 cited pairs, or 99.3%, the cited brand appeared on the answer's ranked shortlist. The average shortlist held 6.0 brands, and any single ranked brand was the pick 15.7% of the time.

Citation and shortlist membership almost always occur together. The scarce outcome is the pick, because each shortlist has one #1 slot and about six candidates.

Within answers that ranked brands, cited brands were the pick 22.0% of the time. Brands named in the same answers without any attributed citation were the pick 10.4% of the time. That is a 2.1 times difference.

The Ahrefs correlation study found that brand mentions predict AI visibility far better than backlinks. This result adds the outcome side: a citation is associated with about double the pick rate, but this is an observed difference, not proof that the citation caused the pick.

Cited brands backed by one attributed cited page were the pick 17.2% of the time. The first-place rate was 23.6% with two pages, 25.3% with three to four pages, and 31.6% with five or more.

The first-place rate rises with each additional group of cited pages. Even at five or more cited pages, the brand missed the pick 68.4% of the time.

On ChatGPT Search, cited brands were the pick 23.5% of the time (129,119 of 550,469 cited pairs). On Google AI Mode the rate was 20.6% (131,776 of 639,234). The gap is 2.8 percentage points.

The two rates are close enough that the headline conclusion holds on both engines: on each one, roughly 4 in 5 cited brands were not the pick.

4,635 brands had at least 100 attributed cited pages in the window. Their median first-place rate among cited answers (the share of a brand's cited answers where it was the pick) was 21.7%. 574 of those brands, or 12.4%, ranked first in under 5% of their cited answers, and 80, or 1.7%, were never the pick in any answer at all.

Many brands receive citations without being the first recommendation. Looking at each brand separately shows why citation counts alone can overstate its position in the answer.

Among 9,758 brands with at least 20 cited answers, the rank correlation between citation volume and first-place rate among cited answers was 0.2309. The average first-place rate rises from 17.6% in the lowest citation-volume quintile to 27.5% in the highest.

Volume mostly reduces the chance of never being picked: the share of brands never picked falls from 16.9% in the lowest quintile to 0.15% in the highest. It does not sort the winners; brands with similar citation volumes rank first at very different rates.

OFX had 979 attributed cited pages across 362 cited answers and was never the pick in any of them. Recurly ranked first in 1 of 387 cited answers, Medium 1 of 271, and SOAX 1 of 309.

These brands are cited as evidence and listed on shortlists, and the answers pick someone else. The leaderboard shows the 10 most-cited brands whose first-place rate stayed under 2%.

Alphabet led the index with 27,835 attributed cited pages and ranked first in 23.6% of its cited answers. LinkedIn ranked first in 52.8%, Datadog 46.7%, and Cloudflare 51.4%. At the other end, Notion ranked first in 13.2% and Dynatrace 16.0%. AiDash ranked first in 77.8%, the highest rate on the list.

Citation volume and first-place rate tell different stories even for the most-cited brands. Two observability brands on the same list, Datadog and Dynatrace, ranked first at rates of 46.7% and 16.0%.

Among brands picked in at least 150 answers, The Walt Disney Company had the thinnest citation footprint: 244 picks on 198 attributed cited pages, or 0.81 cited pages per pick. OpenAI took 3,197 picks on 4,215 cited pages, or 1.3 per pick.

Zero-citation repeat picks are rare: of 3,675 brands picked in at least 20 answers, only 12, or 0.33%, had zero attributed cited pages. Picks without citation support exist, but almost no brand builds a repeat-pick record on them.

Across 23 industries with at least 10,000 cited pairs, first-place rate among cited answers ranged from 30.7% in Consumer Electronics to 20.4% in Sales and Marketing. Consumer categories cluster at the top of the range and several B2B service categories sit at the bottom.

One plausible reading is recommendation concentration: where AI answers repeat one winner, when that winner is cited, it often also ranks first, and where shortlists are long and varied, most citations back a brand that is not picked. This study does not measure concentration directly, so treat that as a hypothesis.

Weekly first-place rate among cited answers stayed between 19.8% and 24.2% across the seven full weeks of the window. There is no trend large enough to change the headline.

The stability matters because pooled rates can hide period effects. Here the 21.9% headline is close to every individual week.

The window starts July 6, 2026 because recommendation-order data reaches full coverage in the index in early July; inside the window, weekly shortlist coverage for cited brands never dropped below 99.1%. Earlier weeks would understate recommendation outcomes as a coverage artifact, so they are excluded. A citation counts for a brand only when the answer attaches the cited page to that brand, which also means every cited pair involves a named brand by construction.

Two product-level rows and one duplicate brand identity were excluded from the leaderboards; the receipts document them. The reverse question, whether recommendations carry citations, is a different study: 45.1% of ranked brand appearances in this window had at least one attributed cited page. All results are measured in the Parse index over the stated window, and none of the differences here prove cause.

The 21.9% rate divides first-place recommendations by cited brand-answer pairs. It is conditional on the brand being associated with a citation. An appearance rate instead divides answers naming the brand by all eligible answers. Those denominators answer different questions and should not be substituted for one another.

Illustrative example: across 100 answers, a brand is cited in 20 and first in five of those 20. Its cited-pair first-place rate is 25%, while those five answers represent 5% of the full panel. Neither figure measures sales conversion. Compare buying intent and the number of alternatives before attributing a low first-place rate to weak content.

Citations and first-place recommendations measure different outcomes. In this sample, 21.9% of cited brand-answer pairs were also first-place recommendations. That association does not establish that earning a citation causes a brand to appear or win.

Report citation presence and first-position rate separately. If a brand is cited frequently but rarely picked, inspect question intent and the actual comparison before changing content. An answer may cite a brand for a limitation, an alternative or a factual detail. No sales or lead-conversion outcome was measured.