Google started putting rows of clickable article cards inside AI Mode answers, and the only study that splits AI citation rates by company size puts a solo operator at 1.4 percent.

Robby Stein, VP of Product for Google Search, announced it on X with the wording "Now live in AI Mode (already in AI Overviews): link carousels for developing topics." The carousel is a horizontal row of article cards placed between sections of the AI response, each card carrying an image, a headline, a source name and a publication date. Google says it appears "for some searches" about developing topics. Preferred Sources can be highlighted inside the carousel, and Google's stated count of unique sources selected as Preferred Sources is now over 600,000. The same carousel has been in AI Overviews since June. Neither report states device, country or language scope, so treat it as unbounded until Google says otherwise.

The 24 August brief covered Bot Preference Sync as robots.txt automation and missed the substantive part. Cloudflare set four conditions a mixed-purpose crawler (one doing both search and training) must meet to keep reaching sites that disallow training. Verbatim: the bot "must respect, via any mechanism, a 'no training' preference in robots.txt"; "they give site owners a way to opt out of AI summaries"; "they provide URL-level visibility into which pages were made available for training, as well metrics on search results, so you can see how your content was used"; and "they can show publicly that Disallowing Training does not hurt your traditional search results." Cloudflare's framing is "making Transparency the price of admission." The feature was stated as shipping "in the coming week" from 21 August, so it is due now and I have not confirmed it is live.

Method: 200 instrumented sites over 16 weeks, 12 January to 28 April 2026, four engines (ChatGPT Search, Claude with web search, Gemini, Perplexity), 80 buyer-intent prompts per site times four engines times three runs, so 960 observations per site and 192,000 in total. Citation rate is the share of prompts where the domain appears as a citation, blended equally across engines.

Median citation rates by headcount band: solo and 0 to 10 people (n=50) 1.4 percent, with a 90th percentile of 3.0 percent; 10 to 50 people (n=80) 3.4 percent; 50 to 500 (n=50) 6.2 percent; 500 plus (n=20) 11.4 percent. So the largest band earns roughly eight times the median of the smallest.

Three limits, all of which matter. Attrifast sells citation-tracking software, so this is a vendor study. The cohort is dominated by B2B SaaS, fintech and consumer software, with no coaching or local services isolated, so headcount here is a proxy for company type as much as for size. And the study's own caveat is worth quoting: "Citation rate is the first checkpoint in a four-step funnel: presence, click, session, payment. This study measures only step one." As with the SE Ranking model reported yesterday, it counts pages cited, not businesses named.

The detail that is arguably more useful than the headline: the engine list is ChatGPT, Claude, Gemini and Perplexity. Google AI Overviews is absent.