LLM Output Dependency on Google and Bing Rankings: Independent AI Response Test Results #8

ChatGPT, Google AI Overview, and other AI assistants already process millions of queries daily, offering ready-made answers instead of the list of pages that regular Google search provides. For SEO professionals, this means the emergence of a new obvious challenge: for effective client promotion, it’s not enough to simply reach Google’s TOP rankings — you need to find out whether users will see their site in AI responses.

But is it necessary to be in the TOP rankings for this, or do rankings not affect it at all? Do LLM systems use traditional search rankings as the basis for their recommendations? If so — to what extent? Understanding these mechanisms is critically important for adapting SEO promotion strategies (and marketing in general) in the new era of AI search. So, faced with these questions, we began looking for answers.

Research Hypothesis

We put forward two key hypotheses based on our own observations and opinions:

Hypothesis 1: A site’s appearance in ChatGPT output depends on its position in Google and Bing’s organic search results. Sites in the TOP-30 have a significantly higher probability of being mentioned in AI responses.

Hypothesis 2: Google AI Overview has an even stronger dependency on its own organic search results since it’s a product of Google’s ecosystem, and the algorithm most likely prioritizes its own top results.

Additional assumption: The dependency may vary depending on query type (local vs. general), language (Ukrainian vs. English), and account type (logged in vs. incognito), which influenced our query selection for testing.

Experiment Methodology

To ensure maximum objectivity, we immediately describe the conditions we set for ourselves.

Geography:

• Ukrainian GEO (via VPN)
• US GEO (via VPN)

Using VPN allowed us to avoid the influence of personalization based on geolocation and obtain “clean” search results that a typical user from Ukraine and the US would see. We’ve already described the impact of GEO on ChatGPT results, as we tested this hypothesis first.

Account Types:

• 2 logged-in accounts
• 2 accounts in incognito mode

Logged-in accounts allow tracking the influence of search history and personalization, while incognito mode shows basic results without considering user context.

Research Volume:

• 10 queries in Ukrainian
• 10 queries in English

The selection of 20 queries provided a sufficient statistical base for identifying patterns, covering various intent types: informational, commercial, local, and navigational.

Test Query Set

Where?	Ukrainian Queries	English Queries
Google	Послуги SEO просування	SEO promotion services
ChatGPT	Компанії з послуг SEO просування	Give me a list of companies providing SEO promotion services
Google	кафе неподалік	cafe nearby
ChatGPT	в яке кафе неподалік мені піти	which cafe nearby should I go to?
Google	Автосалон	Car showroom
ChatGPT	Напиши список із 10 автосалонів	Write a list of 10 car dealerships
Google	Купити Біткоін	buy btc
ChatGPT	Де купити біткоін	where to buy btc
Google	доставка їжі	food delivery
ChatGPT	де замовити доставку їжі	where to order food delivery
Google	додаток для психічного здоров’я	mental health app
ChatGPT	додаток для психічного здоров’я	mental health app
Google	платформа онлайн-навчання	online learning platform
ChatGPT	платформа онлайн-навчання	online learning platform
Google	SEO-агентства в Україні	seo agencies in California
ChatGPT	SEO-агентства в Україні	seo agencies in California
Google	криптогаманці для початківців	crypto wallets for beginners
ChatGPT	криптогаманці для початківців	crypto wallets for beginners
Google	платформа для потокового передавання музики	music streaming platform
ChatGPT	платформа для потокового передавання музики	music streaming platform

Data Collection Process

Data collection occurred in several sequential stages, each requiring result recording. The process was completely standardized for all 20 queries, ensuring data comparability. Using automated tools to collect TOP-100 results avoided human error and missed positions. Our action plan was as follows:

1. Organic search collection: For each query, we captured TOP-100 results from Google and Bing
2. AI Overview collection: Recorded all Google AI Overview results (if they appeared)
3. ChatGPT querying: Ran identical queries through all 4 account types
4. Match analysis: Compared domains and URLs from AI responses with organic search positions
5. Metric calculation: Determined hit rate (share of mentioned sites), average position, citation probability for different position ranges

All data was compiled into structured tables for statistical analysis.

Research Results

Part 1: Dependency on Google (Ukrainian and English Results)

AI Overview

Ukrainian Results:

• Average dependency on Google TOP-100: ~26%
• Peak in TOP-30: 70–80% matches
• For crypto/service queries (“Buy Bitcoin”): 58% dependency
• For geo-dependent queries (“cafe”, “delivery”): 0–25% dependency

This indicates strong dependency on TOP-10–30 for topics with clear recommendations, but independence for local queries.

English Results:

• Average dependency on TOP-100: ~24%
• Peak in TOP-30: 70–90% matches
• AI often focuses on review articles and popular sources (Wikipedia, NerdWallet, VeryWellMind) for categorical queries (“mental health app”, “online learning platform”): 31.7–63.6%
• In contrast, for local queries (“cafe”, “food delivery”): 0–19%

The absence of AI Overview in some queries (food delivery, SEO agencies) lowers average dependency. This suggests AI Overview is effective for structured topics but weaker for local queries due to generalization.

Key feature: AI Overview selectively chooses authoritative sources (binance.com, promodo.com, coursera.org, wikipedia.org), ignoring 60–70% of sites from TOP-30. For categorical queries, dependency is higher; for local queries — low or absent due to answer generalization.

ChatGPT

Ukrainian Results:

• Average dependency on TOP-100: ~10.7%
• Hit rate in TOP-30: 60–70%
• Higher dependency for crypto (“Buy Bitcoin”): 21%
• Lower for geo-dependent (“cafe”, “delivery”): 0–25%

ChatGPT relies more on internal knowledge, generating lists of companies/services without strong attachment to current search results. Logged-in accounts more frequently localize responses (Dnipro/Ukraine), with higher matches than incognito.

English Results:

• Average dependency: ~30%
• Hit rate in TOP-50: 80–90%
• Strong dependency for lists (“BTC”, “crypto wallets”, “food delivery”): 54.5–66.7%
• Logged-in accounts more frequently personalize for the US, with higher matches

General trends: Dependency is higher for general topics (SEO, crypto), where GPT/AI use popular sources. For local queries (cafes, delivery) — lower, due to personalization of responses without directly copying Google’s TOP. Average probability of appearing in TOP-30: ~60% for AI, ~50% for GPT. Recommendations: For accuracy, specify location in tests; aggregation shows GPT is less dependent on Google than AI.

Probability Table for Appearing in Results

Position Group	Google Positions	Average Probability for AI Overview	Average Probability for GPT
1	1–10	30% (0.3)	16% (0.16)
2	11–20	4% (0.04)	2% (0.02)
3	21–30	6% (0.06)	2% (0.02)
4	31–40	0% (0.0)	2% (0.02)
5	41–50	4% (0.04)	0% (0.0)
6	51–60	0% (0.0)	0% (0.0)
7	61–70	0% (0.0)	0% (0.0)
8	71–80	0% (0.0)	0% (0.0)
9	81–90	0% (0.0)	0% (0.0)
10	91–100	0% (0.0)	0% (0.0)

Probability of URL getting into AI Overview and ChatGPT (Ukrainian edition)

Probability of URL getting into AI Overview and ChatGPT (English edition)

Part 2: Dependency on Bing (English Results)

Based on query analysis, the following generalized conclusions can be drawn regarding the dependency between Bing search results and GPT citations:

Strong correlation with Bing TOP-10:

• 50–53% of GPT links fall in Bing TOP-10
• 60–70% of unique domains from TOP-10

This confirms that GPT (likely through RAG mechanics) actively uses top Bing search results, especially review articles (e.g., pcmag.com, forbes.com, clutch.co).

Sharp decline after TOP-10:

• Positions 11–20: 20–30% probability
• Positions 21–30: 10–30% probability
• Positions >30: 17–24% hit rate

This indicates GPT’s clear focus on high-ranking sources, although additional sources from lower positions (e.g., wikipedia.org, theverge.com) may appear through web_search or contextual relevance.

Account type impact:

• Logged-in responses: 60–70% of results from TOP-10 (authoritative reviews, e.g., verywellmind.com, coursera.org)
• Incognito mode: 40–45% of results, wider range of sources (>20 positions), such as local sites (st8.com, exodus.com)
• Focus difference: ~20–25%, which may indicate different data processing priorities depending on mode

Citation probability:

• Position 1 in Bing: 25–40% chance (coursera.org, clutch.co)
• TOP-10: 5–20% per position
• Positions >20: <15–20%
• Exceptions wikipedia.org or youtube.com: 25–40% regardless of position, have consistently high citation probability due to their authority and wide presence in web_search

Influence factors:

• Content type: Review articles dominate (forbes.com, techradar.com, coincodex.com)
• Localization: For regional queries, GPT includes local sources but focuses on national reviews
• Web_search: Additional sources (Reddit, YouTube) appear in incognito or through web_search
• Topics: For technical topics — specialized sources (coincodex, clutch), for general — mass reviews (pcmag, wired)

Differences by topic:

• Online learning platform: High concentration on top platforms (coursera, udemy) due to their popularity and frequent mentions in reviews
• SEO agencies in California: Greater emphasis on aggregators (clutch, designrush), but local agencies appear through web_search
• Crypto wallets for beginners: Mixed focus on reviews (coincodex) and hardware wallets (ledger, tangem), reflecting balance between informational and product sources
• Music streaming platform: Dominance of mass platforms (spotify, apple) with emphasis on reviews (pcmag, techradar), but with niche additions (qobuz) through web_search
• Mental health apps: Strong dependency on top reviews (verywellmind, healthline), but local/AI apps (rejoyn, woebot) appear through specialized sources

General Conclusion

Probability of a URL getting into ChatGPT by Bing positions

GPT demonstrates clear dependency on Bing TOP-10 results, especially for review and ranking sources. Logged-in responses are more oriented toward authoritative sources, while incognito mode expands coverage to less popular but relevant sites through web_search. Citation probability depends on Bing position, but exceptions (wikipedia, youtube) have stable influence. Query topics and localization affect source selection, but the general trend is priority for top, structured reviews from the US.

Google vs Bing Comparison

Probability of URL being included in ChatGPT response (%)

Probability of getting into ChatGPT answer by TOP-30 groups (Google vs Bing)

Detailed Comparison of Key Metrics

Metric	Google AI	Bing AI	Comment
Hit rate from TOP-30	65–85%	55–75%	Google shows higher results through broader data volume. Bing is stable but limited by search depth. For review queries (mental health, online learning) the difference is smaller: Google 80–90%, Bing 70–75%
Link frequency from TOP-30	70–90%	60–80%	Google covers more global and local sources, increasing citation frequency. For popular topics (crypto, music) Google reaches 80–90%, for local (cafe, food delivery) — 70–75%. Bing shows 60–70% for local
TOP-10	45–65%	48–60%	Google has higher rate through priority to authoritative sites (forbes.com, pcmag.com). Bing shows close results but stronger for aggregators (yelp.com, clutch.co) in local queries
TOP-11–20	15–25%	15–25%	Sharp decline after TOP-10 is the same for both systems. Citation probability drops 3x compared to TOP-10
TOP-21–30	10–20%	10–20%	Minimal citation probability. Bing adds ~15% results in this range compared to TOP-10, Google covers more sources including Reddit/YouTube
TOP-1 citation probability	30–50%	20–35%	Google shows higher probability for first position. For popular topics (crypto) Google metric reaches 70%, for local Bing shows 60% (food delivery)
Probability per position (TOP-10)	5–15%	5–18%	Average citation probability for each position in TOP-10. Bing has slightly wider range due to variability by query types
TOP-21–30 probability	<10%	<10%	Minimal probability for both systems in this range
Logged-in accounts (from TOP-30)	60–80%	55–75%	Logged-in responses gravitate toward authoritative sources from TOP-30. Google shows higher personalization. Difference between them and incognito is ~15–25%
Incognito (from TOP-30)	50–70%	45–65%	Incognito expands source coverage. Google adds ~25% outside TOP-30 from internal base, Bing — ~20% through web_search
Internal base / web_search	20–30% outside TOP-30	15–25% through web_search	Google uses internal base for local queries (NJ queries), especially for wikipedia.org, youtube.com. Bing more actively uses web_search in incognito, adding Reddit, YouTube, local sites (audimeadowlands.com)
Exceptions (Wikipedia, YouTube)	~30% regardless of position	~25% regardless of position	Authoritative sources have stable high citation probability regardless of exact position in organic search
Popular topics (crypto, music)	65–80% from TOP-30	55–70% from TOP-30	Google shows stronger correlation for global topics through larger data volume. Both systems focus on review articles (forbes.com, pcmag.com, coincodex.com)
Local queries (NJ, California)	70–75%	60–70%	For US local queries, Google correlates more strongly through broader source coverage. Bing actively uses aggregators (yelp.com, clutch.co). Both systems show lower dependency for geo-dependent queries without US context

Summary Conclusion

GPT demonstrates strong dependency on TOP-30 of both search engines, but Google dominates due to deeper US data coverage (65–85% vs 55–75% for Bing). The role of internal base and web_search makes dependency more flexible, allowing citation of sources outside TOP-30 (wikipedia.org, youtube.com, local sites).

Logged-in responses gravitate toward authoritative sources from TOP-30 (55–80%), incognito expands coverage to 45–70% through web_search. Popular topics (crypto, music) show 65–80% results from TOP-30 reviews, local queries with US context — 60–75% through aggregators.

Conclusions and Practical Significance

Our research confirms that AI systems are not completely independent of traditional search engines. There is a clear correlation between Google/Bing positions and citation probability in ChatGPT and AI Overview, especially for TOP-10 results.

Final Conclusions:

Metric	Google	Bing
Leader in TOP-30 hit rate	Yes (75.2%)	No (65.4%)
Leader for reviews (forbes, pcmag)	Yes	No
Leader for local aggregators (yelp, clutch)	No	Yes
Stronger correlation	-0.94	-0.89
Uses web_search more	No	Yes (incognito)
Better for US queries	Yes	No

However, the dependency is not absolute. LLMs add significant volume of their own knowledge (~20–75%) and selectively choose sources based on authority, content type, and relevance. This means SEO professionals need a hybrid strategy: achieving high organic search positions + building authority + creating content optimized for AI citation.

Strategic Conclusions for Client Promotion

Citation probability drops sharply after TOP-10:

• AI Overview: 30% for TOP-10 vs 4–6% for TOP-11–30
• ChatGPT: 16% for TOP-10 vs 2% for TOP-11–30

Therefore, focus on achieving TOP-10 positions for critically important queries.

Additionally, AI systems selectively choose sources, even from TOP-30. Wikipedia, major reviews (Forbes, PCMag), official sites (Binance, Coursera) have stable citation probability of 25–40% regardless of exact position. So it’s worth investing in building domain authority through quality content, E-E-A-T signals, mentions in authoritative sources.

Review articles with rankings, comparisons, “best of” lists significantly more often appear in AI responses than product or commercial pages. Therefore, create comprehensive reviews, comparisons, and guides for key topics in your niche.

Answers to Key Experiment Questions

1. Does site appearance in ChatGPT output depend on Google/Bing position?

Yes, it depends, but not completely. Sites from top positions (TOP-30) have higher probability of appearing in GPT responses — up to 70–90% for TOP-10. However, GPT adds ~75% content from its own knowledge base, so even sites from TOP-30 can be ignored if they don’t match the internal base.

2. Does appearance in AI Overview depend on Google position?

AI Overview has strong dependency on Google TOP-30 with appearance probability of 70–80% for top positions. Average dependency is ~27% (26% Ukrainian, 24% English), but it’s selective: AI Overview chooses authoritative sources, ignoring 60–70% of other sites from TOP-30. For categorical queries (SEO, crypto, education) dependency is higher — 50–100% matches in TOP-30, especially in English results. For local queries (cafe, delivery) dependency is low or absent (0–25%) due to generalization.