Key findings
- Google says more than 345,000 unique sources had been selected as Preferred Sources and users were twice as likely to click a selected source.
- Preferred Sources is personalized, while Highly Cited highlights original or influential reporting; Google does not describe either as a general ranking factor.
- Pew found traditional-result clicks in 8% of visits with an AI summary versus 15% without one; cited-source clicks occurred in only 1% of visits.
- Geolix.ai’s English fintech case separates citation from visibility: ChatGPT own-domain citation rate was 22.9%, while brand mention rate was 6.9%.
- After four Geolix.ai articles were published, ChatGPT own-domain citation rate rose from a 9.7% pre-publication average to 26.7% afterward; the change is observational, not causal.
What Google changed
On May 27, 2026, Google extended Preferred Sources into AI Overviews and AI Mode. Users can select domains or subdomains they want to see more often, and eligible links may display a Preferred badge. Subdirectories are not separate Preferred Source entities. Google also expanded Highly Cited labels to identify primary or influential reporting and added link treatments for timely articles and first-hand perspectives.
Sources: Google product announcement | Google Search Central guide
The update should be read as a discovery and trust signal, not a shortcut to AI inclusion. Preferred Sources depends on a user’s selection. Highly Cited has no application process and does not mean that link volume alone produces AI visibility. The durable strategy is to publish material that other credible sources can verify and cite.
Why citations matter even when clicks decline
Pew Research Center analyzed 68,879 Google searches, including 12,593 with an AI summary. Users clicked a traditional result in 8% of visits with an AI summary, compared with 15% without one. They clicked a source cited within the AI summary in only 1% of visits. Yet 88% of the summaries cited at least three sources. Citation can therefore shape the generated answer even when it produces little referral traffic.
Sources: Pew Research Center click study
A separate Wix Studio AI Search Lab study, reported by Search Engine Land, analyzed 75,000 AI answers and more than one million citations. Listicles, articles and product pages together accounted for 52% of citations, while query intent was strongly associated with source format. This does not establish a universal template; it suggests that citable content must answer a specific evidence need.
Sources: Search Engine Land / Wix citation study
Geolix.ai evidence: citation, mention and recommendation are different
Geolix.ai's English fintech case monitored nine English purchase-intent questions for a GEO service brand in the fintech category, targeting Singapore and APAC. From July 20 to 31, 2026, ChatGPT API, Gemini and Perplexity produced 15,495 valid answers and 223,196 citation records. Google AI Mode and Google AI Overviews were excluded because collection success was too low for statistical use.
| Engine | Valid answers | Mention rate | Top-1 rate | Own-domain citation |
|---|---|---|---|---|
| Perplexity | 3,150 | 75.3% | 51.9% | 85.5% |
| Gemini | 3,195 | 38.6% | 25.6% | 25.5% |
| ChatGPT API | 9,150 | 6.9% | 0.8% | 22.9% |
The contrast is clearest in ChatGPT: the brand’s domain was used in 22.9% of answers, but the brand was named in 6.9%. Citation indicates that content entered the evidence pool; mention indicates that the brand entered the answer; Top-1 rate indicates recommendation prominence. Treating these as one metric would overstate or understate performance depending on the engine.
Source composition also varied. Vendor and service-provider websites accounted for 90.3% of ChatGPT citations, 91.3% of Gemini citations and 88.4% of Perplexity citations in this purchase-intent case. That pattern is relevant to this category, but it should not be generalized to all fintech questions, especially regulatory or news queries where official and editorial sources may carry more weight.
What happened after Geolix.ai published new evidence
Geolix.ai published four articles on July 22. In ChatGPT monitoring, average own-domain citation rate was 9.7% before publication and 26.7% afterward, a 2.8-fold difference. Average brand mention rate moved from 2.2% to 8.3%, a 3.8-fold difference, and the average number of cited own-domain URLs increased from two to 7.2. The results are directionally consistent with broader source coverage, but the period included multiple simultaneous changes and no control group, so the publication cannot be claimed as the sole cause.
A second batch of six pages published on July 27–28 produced another useful observation. The most-cited new page accumulated 2,081 citations, while its Chinese version accumulated 1,891. Perplexity’s first-day own-domain citation rate reached 88.8% and brand mention rate 76.8%; by July 31, mention rate reached 81.9% and Top-1 rate 74.0%. Gemini mention rate rose from 32.1% to 44.8% over four days. Because several pages and actions launched together, the data cannot prove that title matching or any single page feature caused the change.
What fintech brands should publish
- Original benchmarks with sample size, collection dates, definitions, methodology and limitations.
- Stable product and regulatory fact pages covering legal entity, licence, markets, fees, eligibility, risks and update dates.
- Decision-oriented comparisons and market maps that define inclusion rules and avoid pay-to-play rankings.
- Research summaries that cite primary regulators, filings, company announcements and peer-reviewed work.
- English and Chinese versions that preserve the same controlled facts while adapting examples and distribution to each ecosystem.
- A complete References section and durable URLs so journalists, analysts and AI systems can verify claims over time.
How to measure authority without overstating it
- Track citation rate, brand mention rate, recommendation rank and referral traffic as separate outcomes.
- Segment results by engine, prompt intent, language, market and date; never rely on one composite score.
- Record whether citations come from the brand, regulators, media, directories, communities or other vendors.
- Use pre/post monitoring for diagnosis, but label it observational unless a controlled design supports causality.
- Audit factual accuracy separately; the current Geolix.ai production export does not contain a factual-accuracy field.
Methodology and limitations
The Geolix.ai figures come from an anonymized, read-only production-monitoring export dated July 31, 2026. That case used 100 scheduled repetitions per question per day, but engine coverage periods differed: ChatGPT covered 12 days, while Gemini and Perplexity covered four. Google AI Mode succeeded on 1 of 450 scheduled runs and Google AI Overviews on 62 of 450, so both were excluded. Before-and-after findings are observational and may reflect multiple concurrent changes. These data support measurement and operating recommendations, not claims about Google’s ranking algorithms.
Frequently asked questions
Is Preferred Sources a ranking factor?
Google has not described it as a general ranking factor. It is a personalization feature that may highlight content for users who select a domain.
Does an AI citation guarantee traffic or a brand mention?
No. Pew found very low clicks on cited AI-summary sources, and Geolix.ai observed that ChatGPT cited a brand domain much more often than it named the brand.
Can a pre/post uplift prove that publishing caused the result?
Not without stronger controls. It is useful operational evidence, but other content, engine and timing changes may contribute.
References
- Google — New ways to find your favorite sources and original content in AI Search: https://blog.google/products-and-platforms/products/search/original-high-quality-content-search/
- Google Search Central — Preferred Sources guide: https://developers.google.com/search/docs/appearance/preferred-sources
- Google Search Central — Optimizing for generative AI in Google Search: https://developers.google.com/search/blog/2026/05/a-new-resource-for-optimizing
- Pew Research Center — Google users are less likely to click when an AI summary appears: https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/
- Search Engine Land — Preferred Sources and Highly Cited coverage: https://searchengineland.com/google-ai-overviews-ai-mode-gain-preferred-sources-plus-new-perspectives-carousel-and-highly-cited-labels-478766
- Search Engine Land / Wix Studio AI Search Lab — Citation format study: https://searchengineland.com/ai-citations-favor-listicles-articles-product-pages-study-472364
- Geolix.ai — Official website and engine coverage: https://geolix.ai/
- Geolix.ai — GEO bilingual case dataset: Geolix.ai internal production-monitoring export, anonymized, July 31, 2026.