The Source-to-Mention Gap: When Google AI Overviews Read Your Domain But Recommend Competitors

Most AI visibility dashboards treat citations and brand mentions as two views of the same signal. They are not.

Cleanlist's September 2026 study of 205 B2B data and GTM buyer queries found that Google AI Overviews routinely cite one vendor's domain while naming a different vendor in the answer text. Apollo appears in 47 of 73 readable AI Overviews but is cited on apollo.io in only 17. Cleanlist ranks 5th among cited domains yet is named in only 2 answers — finishing last on the ratio that matters.

The gap between being read (your domain in the source list) and being recommended (your brand in the answer) is now measurable. Teams optimizing for citations alone are optimizing for half the buyer journey.

What Cleanlist measured

On September 1, 2026, Cleanlist pulled 205 buyer-intent queries in the B2B data enrichment and GTM tooling category via DataForSEO's SERP API — Google US desktop, one live pull per query.

The 73-answer base matters. One hundred twenty-three AI Overviews returned as stubs without readable text — a collection artifact from synchronous SERP pulls where Google loads Overview content asynchronously. Every citation and mention rate on this page divides by 73, not 196, because only those 73 could produce either signal.

Cleanlist published the full dataset under CC BY 4.0 so counts can be recomputed independently.

Finding 1: Two leaderboards that disagree

When you rank domains by how often AI Overviews cite them versus how often they name brands in answer text, the lists diverge sharply.

Most-cited domains (once per AI Overview, n=73):

Most-named brands in answer text:

Google reads Cleanlist 5th and names Cleanlist 15th (tied). Google reads Apollo 4th and names Apollo 1st. Same answer box, opposite outcomes.

Finding 2: The source-to-mention ratio

Cleanlist introduces a metric worth adopting: source-to-mention ratio = brand mentions in AI Overview text ÷ AI Overviews citing that vendor's own domain.

Vendor ratios (registrable domain rollup, vendors with ≥5 citations):

Apollo's ratio is 18.3× Cleanlist's on this sample. That is not a rounding error — it is the difference between a vendor whose name travels through third-party evidence and a vendor whose pages get harvested without attribution.

Finding 3: Citations are neither necessary nor sufficient for mentions

The study's most actionable number: of 217 brand-mention events, only 33.2% included a citation to that vendor's own domain in the same answer.

Five vendors were named without their domain ever being cited anywhere in the corpus: Clearbit (5 mentions), Bombora, Prospeo, Smartlead (2 each), Kaspr (1).

Apollo was named in 47 answers and cited on apollo.io in only 17 — meaning 30 mentions arrived with no Apollo page in the reference list.

This inverts the assumption that "get cited, get recommended." Something on other people's pages is doing the recommendation work. That aligns with earned media driving 84% of AI citations and with Parse.gl's finding that 38.55% of reviewed citations point to pages that do not mention the attributed brand at all.

Finding 4: AI Overviews are near-universal in B2B buyer queries

98.5% of usable GTM buyer queries triggered an AI Overview on September 1, 2026. Only three exceptions: "hubspot data enrichment," "uplead pricing," and "best mobile number data provider."

For category leaders, the practical reading is blunt: the traditional blue-link SERP has effectively stopped existing on the queries buyers type before they reach your pricing page. Your share of voice is decided inside the answer box, not on page one.

What drives the gap

Cleanlist's data does not isolate causation, but the pattern matches what other September 2026 studies show:

Third-party evidence sets the shortlist. When Apollo is named 47 times but cited 17 times, the other 30 mentions likely come from ZoomInfo pipeline pages, YouTube comparisons, Reddit threads, and editorial roundups that discuss Apollo without linking apollo.io. Google's synthesis reads those sources and names the brands they discuss.

Your own domain can be a source without making you the hero. Cleanlist is cited 14 times but named twice — the inverse problem. Its pages may supply data or context while the answer recommends Clay or Apollo based on broader consensus signals elsewhere in the retrieval set.

Mentions and citations are different retrieval behaviors. A citation means a page entered the evidence set. A mention means the synthesis layer judged that brand relevant to the buyer's question. Those layers do not always align — a point BuzzStream's July 2026 study made across 12,000 responses when it found only 23.1% of brand mentions also carried a citation to that brand's domain.

Practical playbook

Measure both signals separately

Track citation share (how often your domain appears in source lists) and mention rate (how often your brand name appears in answer text) on the same prompt set. Compute your source-to-mention ratio. A ratio below 1.0 means you are feeding answers without earning recommendations.

Diagnose which side is failing

Work the third-party layer first for recommendation lift

If your ratio is below 1.0, another FAQ rewrite will not fix it. Prioritize:

1. Presence on domains Google already cites — YouTube comparisons, Reddit threads at BOFU, G2/Capterra profiles, editorial roundups
2. Brand mentions on cited pages — not just backlinks, but named inclusion in the comparison text AI reads
3. Category rankings you control — vendors with high ratios often publish credible "best X in our category" content that doubles as citable evidence (see second-hand visibility on best X queries)

Do not treat citation share as winning

A dashboard showing rising citation share while mention rate stays flat is a warning, not a win. You are becoming source material for answers that recommend competitors.

How Obsurfable fits

Obsurfable records both signals on the same buyer prompts: whether your brand is named in the answer and whether your domain is cited as a source. The Visibility Director flags prompts where competitors get mentioned while you only appear (or do not appear) in citations — the exact failure mode Cleanlist's ratio captures.

Instead of one blended "AI visibility score," Obsurfable separates retrieval (citation) from recommendation (mention) so budget goes to the layer that is actually broken.

For the broader recognition-versus-recommendation gap, see AI Recognizes 96% of Brands — Then Mentions Almost None. For when AI uses your content without naming you, see Ghost Citations.

FAQ

Should I stop caring about citations?

No. Citations still matter for retrieval eligibility, referral traffic, and understanding which pages AI reads. But citation share alone misstates whether you are winning the recommendation.

Is the source-to-mention ratio stable across categories?

Cleanlist measured one B2B GTM category on one day. Ratios will differ by vertical, query type, and platform. The metric is portable; the thresholds are not. Run it on your own prompt set.

Why divide by 73 instead of 196?

Only 73 AI Overviews returned readable answer text with reference lists. The other 123 were empty stubs. Dividing by 196 would understate every vendor's visibility by mixing measurable and unmeasurable answers.

Does this apply outside Google AI Overviews?

The ratio concept applies anywhere answers separate brand names from source lists. ChatGPT and Perplexity behave differently — ChatGPT often names brands without citing anything at all. Measure each surface separately.

Bottom line

On 205 B2B buyer queries, Google AI Overviews triggered 98.5% of the time — but only 33.2% of brand mentions came with a citation to that brand's own domain. Apollo's source-to-mention ratio is 2.61; Cleanlist's is 0.14. Being read and being recommended are different jobs. Measure both, compute the ratio, and fix the layer that is actually failing.