Short answer
In inurdemirel's Turkey AI answers study (November 5, 2026 – November 9, 2026), 320 Turkish queries across 8 industry groups were asked to five engines on 3 separate days: Google AI Overviews, Google AI Mode, ChatGPT, Gemini and Perplexity. Average source-domain similarity between engines for the same question is 11.4%. Most similar: Google AI Overviews and Google AI Mode (22.1%); least similar: Google AI Overviews and ChatGPT (5.2%). 37% of pages cited in AI answers have an FAQ section or FAQPage schema; among Google top-10 pages that are not cited it is 21.3% (1.7×).
What did the study measure?
The same 320 Turkish queries were asked to five engines and every source link shown in the answers was recorded. Queries span 40 industries and eight search intents: definition, how-to, comparison, “best”, price, local, purchase and trust. No query contains a brand name; brand mentions were measured from the answers themselves.
Of 4,800 planned requests, 4,794 responses were recorded; 4,623 contained an AI answer, with 26,059 source links in total. Cited pages and Google top-10 pages that were not cited were crawled (13,148 pages), so cited pages could be compared against a control group.
How often do AI answers appear for Turkish queries, and with how many sources?
Google AI Overviews appear for 98.7% of informational and 76.5% of commercial queries.
| Engine | Queries answered | Informational | Commercial | Sources per answer |
|---|---|---|---|---|
| Google AI Overviews | 82.1% | 98.7% | 76.5% | 6 |
| Google AI Mode | 100% | 100% | 100% | 4.1 |
| ChatGPT | 100% | 100% | 100% | 4.1 |
| Gemini | 100% | 100% | 100% | 4 |
| Perplexity | 100% | 100% | 100% | 10 |
Which sites are cited most in AI answers?
The top 10 of 4,626 cited domains take 8.8% of all citations. 148 domains appear in all five engines.
| Rank | Domain | Source type | Share of all citations | Engines |
|---|---|---|---|---|
| 1 | google.com | Brand / business site | 2.8% | 5 |
| 2 | instagram.com | Forums and social | 0.8% | 3 |
| 3 | eksisozluk.com | Forums and social | 0.7% | 3 |
| 4 | allianz.com.tr | Brand / business site | 0.7% | 5 |
| 5 | youtube.com | Video | 0.7% | 4 |
| 6 | donanimhaber.com | News and media | 0.6% | 5 |
| 7 | hesapkurdu.com | Marketplace and comparison | 0.6% | 5 |
| 8 | armut.com | Marketplace and comparison | 0.6% | 5 |
| 9 | vodafone.com.tr | Brand / business site | 0.6% | 5 |
| 10 | ticaret.gov.tr | Government | 0.6% | 5 |
| 11 | enuygun.com | Marketplace and comparison | 0.6% | 5 |
| 12 | trendyol.com | Marketplace and comparison | 0.5% | 5 |
| 13 | isbank.com.tr | Brand / business site | 0.5% | 5 |
| 14 | hangikredi.com | Marketplace and comparison | 0.5% | 4 |
| 15 | webtekno.com | News and media | 0.4% | 5 |
Which source types stand out?
The largest source share in each engine: brand / business site. Forums get the most space in Google AI Overviews (8.4%), marketplaces and comparison sites in Gemini (8.9%).
| Source type | Google AI Overviews | Google AI Mode | ChatGPT | Gemini | Perplexity |
|---|---|---|---|---|---|
| Brand / business site | 72% | 72.8% | 57.5% | 74.5% | 70.6% |
| Marketplace and comparison | 7.3% | 8.7% | 5.4% | 8.9% | 7.1% |
| News and media | 1.9% | 2.8% | 1.7% | 5.2% | 4.9% |
| Forums and social | 8.4% | 4.3% | 0.1% | 1.6% | 0.7% |
| Video | 1.5% | 0.3% | 0% | 0.5% | 0.7% |
| Wikipedia | 0.3% | 0.3% | 0% | 0.1% | 0.4% |
| Government | 1.1% | 1.9% | 13.9% | 0.6% | 2.6% |
| Non-Turkish site | 1.4% | 1.6% | 8.3% | 2% | 5.6% |
How much do AI answers overlap with the Google top 10?
For Turkish queries, 20.9% of Google AI Overviews sources come from domains in the same query's Google top 10; for ChatGPT it is 3%.
The share shows how many of the source links an engine displays also appear in the Google mobile top 10 for the same query.
| Engine | Same URL in Google top 10 | Same domain in Google top 10 | Domain in top 3 |
|---|---|---|---|
| Google AI Overviews | 17.6% | 20.9% | 16.3% |
| Google AI Mode | 4.7% | 8.4% | 5.7% |
| ChatGPT | 0.7% | 3% | 1.9% |
| Gemini | 2% | 6.1% | 2.6% |
| Perplexity | 0.9% | 4.2% | 2.2% |
Do engines show the same sources for the same question?
Average source-domain similarity between engines for the same question is 11.4%. Most similar: Google AI Overviews and Google AI Mode (22.1%); least similar: Google AI Overviews and ChatGPT (5.2%).
In 90.8% of queries answered with sources by all five engines, no single domain is shared by all five.
| Engine pair | Source-domain similarity | At least one shared domain |
|---|---|---|
| Google AI Overviews – Google AI Mode | 22.1% | 77.9% |
| Google AI Mode – Gemini | 19.4% | 79.5% |
| Google AI Overviews – Gemini | 16% | 78.9% |
| Gemini – Perplexity | 11.1% | 72.2% |
| Google AI Overviews – Perplexity | 9.1% | 65.1% |
| Google AI Mode – Perplexity | 9.1% | 58.6% |
| ChatGPT – Gemini | 7.7% | 47.3% |
| Google AI Mode – ChatGPT | 7% | 33.6% |
| ChatGPT – Perplexity | 6.8% | 42.6% |
| Google AI Overviews – ChatGPT | 5.2% | 29% |
How are brands mentioned in AI answers?
For "best …" questions, ChatGPT names 13.9 brands per answer on average. In Gemini, 80.7% of named brands are mentioned without their own site being cited.
| Engine | Brands per answer | In “best” questions | Answers naming a brand | Named without own site cited |
|---|---|---|---|---|
| Google AI Overviews | 7.5 | 10 | 99.8% | 73.4% |
| Google AI Mode | 8.4 | 11 | 99.9% | 75.8% |
| ChatGPT | 10.5 | 13.9 | 99.4% | 61.9% |
| Gemini | 9.1 | 12.1 | 99.2% | 80.7% |
| Perplexity | 5.3 | 8.6 | 94.7% | 66.8% |
Do the same sources come back every day?
Asked on 3 separate days, source-domain similarity from day to day is 66.5% in Perplexity and 35.1% in Google AI Overviews.
| Engine | Day-to-day source similarity | Same first source |
|---|---|---|
| Google AI Overviews | 35.1% | 21.7% |
| Google AI Mode | 36.7% | 19.4% |
| ChatGPT | 37.5% | 31% |
| Gemini | 39.5% | 33.9% |
| Perplexity | 66.5% | 55.6% |
What sets cited pages apart?
37% of pages cited in AI answers have an FAQ section or FAQPage schema; among Google top-10 pages that are not cited it is 21.3% (1.7×).
34.6% of pages cited in AI answers are longer than 1,500 words; among Google top-10 pages that are not cited it is 19.5% (1.8×).
29.1% of pages cited in AI answers contain at least one table; among Google top-10 pages that are not cited it is 14.8% (2×).
41.7% of pages cited in AI answers were updated in the last 12 months; among Google top-10 pages that are not cited it is 28% (1.5×).
An llms.txt file exists for 43.4% of cited and 30.2% of non-cited pages; the difference is statistically significant.
| Page feature | Cited pages | Non-cited top-10 pages | Ratio | Significant |
|---|---|---|---|---|
| FAQPage schema | 23.6% | 11.1% | 2.1 | Yes |
| At least one table | 29.1% | 14.8% | 2 | Yes |
| Year in title (2025/2026) | 19.1% | 10.4% | 1.8 | Yes |
| 1,500+ words | 34.6% | 19.5% | 1.8 | Yes |
| FAQ (schema or section) | 37% | 21.3% | 1.7 | Yes |
| Updated in last 12 months | 41.7% | 28% | 1.5 | Yes |
| llms.txt file | 43.4% | 30.2% | 1.4 | Yes |
| 3+ bulleted lists | 48.3% | 38.4% | 1.3 | Yes |
| 2+ question-style subheadings | 47.8% | 39.8% | 1.2 | Yes |
| BreadcrumbList schema | 61.4% | 52.4% | 1.2 | Yes |
| Short definition in first paragraph | 26.2% | 32.1% | 0.8 | Yes |
| Product schema | 3.8% | 2.6% | 1.4 | No |
| Organization/LocalBusiness schema | 58.3% | 55.6% | 1 | No |
| ClaudeBot allowed | 93.3% | 90.8% | 1 | No |
| GPTBot allowed | 92.8% | 90.4% | 1 | No |
| Google-Extended allowed | 94% | 92.4% | 1 | No |
| Date in schema/meta | 53.2% | 52.3% | 1 | No |
| OAI-SearchBot allowed | 98.8% | 98.3% | 1 | No |
| PerplexityBot allowed | 97.6% | 97.9% | 1 | No |
| TTFB under 600 ms | 74.4% | 75.5% | 1 | No |
| Article/BlogPosting schema | 42.1% | 44.3% | 1 | No |
| Author markup | 52.9% | 56.1% | 0.9 | No |
A data-based GEO checklist
Features that appear significantly more often on cited pages, ordered by ratio.
- FAQ (schema or section): 37% of cited pages vs 21.3% of non-cited pages.
- 1,500+ words: 34.6% of cited pages vs 19.5% of non-cited pages.
- At least one table: 29.1% of cited pages vs 14.8% of non-cited pages.
- Updated in last 12 months: 41.7% of cited pages vs 28% of non-cited pages.
- llms.txt file: 43.4% of cited pages vs 30.2% of non-cited pages.
- FAQPage schema: 23.6% of cited pages vs 11.1% of non-cited pages.
- 3+ bulleted lists: 48.3% of cited pages vs 38.4% of non-cited pages.
- BreadcrumbList schema: 61.4% of cited pages vs 52.4% of non-cited pages.
- Year in title (2025/2026): 19.1% of cited pages vs 10.4% of non-cited pages.
- 2+ question-style subheadings: 47.8% of cited pages vs 39.8% of non-cited pages.
No significant difference: Article/BlogPosting schema, Organization/LocalBusiness schema, Product schema, author markup, date in schema/meta, TTFB under 600 ms, GPTBot allowed, OAI-SearchBot allowed, PerplexityBot allowed, Google-Extended allowed, ClaudeBot allowed.
Method
- Dates and repetition: 3 rounds, November 5, 2026 – November 9, 2026. Each query was asked once per engine per round.
- Location and language: Turkey, Turkish. Google results on mobile (Android). No login or personalisation.
- Engines: Google AI Overviews and organic top 10 (SERP API), Google AI Mode, ChatGPT and Gemini user-interface answers, Perplexity Sonar. Data provider: DataForSEO.
- Source: a link shown in the answer; two links to the same domain in one answer count as one citation.
- Source type: rule-based classification.
- Brand mentions: dictionary match built from cited domains and ChatGPT brand entities.
- Page analysis: each page crawled once, robots.txt respected; control group = Google top-10 pages never cited by any engine; domain-clustered bootstrap confidence intervals.
- Limits: one country, one language, specific dates. Results show association, not causation.
Dataset
Queries, source links, domain summary and page features are published as CSV under CC BY 4.0 and may be reused with attribution. Answer texts are not shared.
Frequently asked questions
Which AI engines does the study cover?
Google AI Overviews, Google AI Mode, ChatGPT, Gemini and Perplexity; the same 320 Turkish queries on 3 separate days.
Do you need to rank in the Google top 10 to be cited in AI answers?
For Turkish queries, 20.9% of Google AI Overviews sources come from domains in the same query's Google top 10; for ChatGPT it is 3%. Most citations come from outside the Google top 10, and engines differ clearly.
Does an llms.txt file affect AI citations?
An llms.txt file exists for 43.4% of cited and 30.2% of non-cited pages; the difference is statistically significant.
Which page features go with AI citations?
The features that most separate cited from non-cited top-10 pages: FAQ (schema or section), 1,500+ words, at least one table, updated in last 12 months. These are associations, not guarantees.
How can the dataset be used?
The query, source, domain and page-feature tables are published as CSV under CC BY 4.0 and may be reused with attribution. Answer texts are not shared.
Sources
- Google Search Central – AI features and your website
- OpenAI – Overview of OpenAI crawlers
- Perplexity – Perplexity Crawlers
- The llms.txt proposal
- RFC 9309 – Robots Exclusion Protocol
This article was last reviewed on
