AI engines choose which pages to cite by running searches for a question and its sub-questions, reading the pages returned, and quoting the ones that answer a part of it clearly. Ranking first does not decide it: 38% of the pages cited in Google's AI Overviews rank in the top 10 for the query (Ahrefs, March 2026), and about 8% of ChatGPT's. Clear structure, recent updates and mentions of the brand elsewhere go with being cited. Length and schema markup do not.
A firm in Limassol can hold the first organic result for its main service and still be absent from every AI answer on the same subject. Google's results and the AI engines read the same web and reward different things. This article sets out what the published studies show about how the engines pick their sources, and what our own Search Console data shows.
Do AI engines cite the pages that rank first on Google?
Partly, and less than they did. Ahrefs found in July 2025 that 76% of pages cited in AI Overviews ranked in the top 10 for the query. By March 2026, across 863,000 results pages and 4 million cited URLs, the figure was 38%. Ahrefs notes that its parser now captures more citations, so the two numbers are not a like-for-like fall. seoClarity's study of 362,000 queries from October 2025 points the same way: 56% of AI Overview citations came from the top 20 organic results, and 44% from further down.
Outside Google the overlap is thinner. Ahrefs tested 15,000 prompts across six assistants in August 2025. On average 12% of cited URLs ranked in Google's top 10 for the same prompt.
| Engine | Where it looks | Cited pages that also rank in Google's top 10 |
|---|---|---|
| Google AI Overviews | Google's index, several searches per question | 38% (Ahrefs, March 2026) |
| Google AI Mode | Google's index, a wider set of searches | About 35% (Semrush, July 2025) |
| Perplexity | Its own crawler, PerplexityBot | 28.6% (Ahrefs, August 2025) |
| ChatGPT search | Third-party search providers, including Microsoft's Bing | About 8% (Ahrefs, August 2025) |
| Gemini and Copilot | Google and Bing respectively | About 8% each (Ahrefs, August 2025) |
For an owner, ranking still helps and is no longer enough. A page at position 14 can be the cited source if it answers the sub-question better than the page at position one.
What is query fan-out, and how does it pick sources?
Query fan-out is the engine splitting one question into several searches and building its answer from the results of each. Google's documentation for its AI features describes it: AI Overviews and AI Mode run related searches across subtopics and assemble one answer. A buyer who asks which accountant in Limassol to use for a new holding company triggers searches on company formation, tax residency, audit requirements and fees. Each of those searches has its own winners.
That is why narrow pages get cited. An article that settles one of those sub-questions in its first two sentences is a candidate for one branch of the fan-out, even when the firm's homepage ranks nowhere for the head term. It is also why this site publishes one question per article.
What makes a page more likely to be cited by AI?
Structure, consistently. Semrush compared more than 300,000 cited URLs with uncited ones in February 2026. Cited pages were 22.9% more likely to have strong section structure, 32.8% more likely to carry a clear summary or takeaways, and 25.5% more likely to use question-and-answer formatting. AirOps found in December 2025 that 87% of pages cited by ChatGPT use a single H1 and nearly 80% include a list.
Our own site shows the pattern in miniature. In the 90 days to 18 September 2026, Search Console recorded 134 impressions inside Google's AI features across 13 of our pages. One of our nine guides earned 16 of them. The other eight shared 5. The guide that earned them is the one whose headings sit closest to the way people phrase a question in full, such as how quickly AI citations decay, and it ranks between first and eighth for those long questions.
Does content length or schema markup affect AI citations?
Neither moves the needle. On length, Ahrefs measured 174,000 pages cited in AI Overviews in December 2025: the median ran to 1,115 words, and the correlation between length and citation was 0.04, close to none. Google's guide to generative AI search says the same in its own words: there is no ideal page length.
On structured data, Ahrefs ran a controlled test in May 2026 on 1,885 pages that added schema markup against 4,000 that did not, and citations in ChatGPT and AI Mode did not change in any measurable way. Schema still earns its place, because it tells Google what a business is and feeds rich results. It does not buy citations.
Does freshness matter for AI citations?
It matters for ChatGPT far more than for Google. Ahrefs studied 17 million cited URLs in July 2025. AI assistants cited pages that were on average 25.7% fresher than the organic results for the same topics. ChatGPT cited pages 458 days newer and listed the newest sources first. Google AI Overviews went the other way, citing pages 16 days older than the organic results.
A second Ahrefs study in October 2025 looked at the 1,000 pages ChatGPT cites more than any others: 76.4% of the dated ones had been updated within the previous six months. The practical reading is a quarterly refresh of the pages that matter, with the content changed before the date. We cover the method in our guide to content decay.
Do brand mentions matter more than backlinks for AI?
For AI visibility, the published correlation says yes. Ahrefs analysed 75,000 brands in May 2025. Mentions of the brand across the web correlated with visibility in AI Overviews at 0.664. The number of backlinks correlated at 0.218. Brands in the top quarter for mentions earned more than ten times the AI Overview mentions of the quarter below.
This is where Cyprus firms are thin. In September 2026 we checked 20 accounting, law and corporate-services websites on the island. One of the 20 firms has a Wikidata entity. None has a Wikipedia article about the firm. An engine asked to recommend a provider has little beyond the firm's own website to confirm that the firm exists, which is the problem entity SEO addresses.
What should a Limassol accountant or a Paphos hotel do with this?
Write for the sub-questions, keep the pages current, and make sure other sources confirm who you are.
- Write down the ten questions buyers ask before they contact you. Give each one its own page and answer it in the first two sentences.
- Structure every page the same way. One H1, question headings in order, a list or a table where it helps, and a short block of takeaways.
- Date the page and refresh it. Every quarter for the pages that bring enquiries, with a real change each time.
- Get mentioned where your sector is listed. The professional body's directory, the chamber, the trade press, LinkedIn. Use the same name, address and phone number everywhere.
- Check that the crawlers can read you. In our September check of 20 hotel websites, one served an empty page until JavaScript ran, and one allowed only Google's crawler and turned every other engine away. Our AI readability checker tests both on any address.
Steps one to three are writing. Steps four and five are what we handle for clients under SEO and AI search, alongside the technical work. The full playbook for a professional firm is in how to get your business cited by ChatGPT, and how the disciplines fit together is in GEO vs SEO.
Key takeaways
- 38% of pages cited in AI Overviews rank in Google's top 10. For ChatGPT the overlap is about 8%.
- Engines split one question into several searches, so a narrow page that answers one sub-question can be cited without ranking for the main term.
- Cited pages share structure: question headings, direct answers, lists, takeaways. Length and schema markup do not earn citations.
- ChatGPT favours recently updated pages far more than Google's AI Overviews do.
- Mentions of the brand across the web correlate with AI visibility three times more strongly than backlinks.
Frequently asked questions
Does ranking first on Google guarantee a citation in AI Overviews? No. Ahrefs found in March 2026 that 38% of pages cited in AI Overviews rank in the top 10 for the same query, down from 76% in mid-2025 on an earlier method. The first organic result is cited more than any other position, yet the larger share of citations goes to pages further down.
Does schema markup help a page get cited by ChatGPT? Not measurably. Ahrefs tested 1,885 pages that added schema against 4,000 that did not in May 2026 and found no meaningful change in ChatGPT or AI Mode citations. Schema remains worth having for Google's rich results and for describing the business clearly.
How long does it take for a new page to be cited by an AI engine? Semrush tracked 81 new pages on a strong domain in late 2025. In Google AI Mode, 36% were cited within 24 hours, with citations coming and going from day to day. ChatGPT cited 10% on day one and 42% within 30 days, and kept citing them once it had started. A small site should expect slower.
Which AI engine should a Cyprus business care about first? Google's AI features and ChatGPT. AI Overviews appear on the search engine your buyers already use, and Similarweb's mid-2025 analysis found ChatGPT behind more than 80% of AI referrals to the top 1,000 global websites. Because ChatGPT search draws on Bing, a page has to be indexed by Bing to be found there.
Do I need an llms.txt file to be cited? No. Google's guide to its generative AI features, updated in July 2026, states that Google Search does not use llms.txt, and no engine has confirmed that it uses the file for citations. It takes 20 minutes and does no harm, as our llms.txt guide explains. The work that earns citations is on the page.



