Short answer
ChatGPT answers most questions without searching the web. When it does search, it often queries a brand's own site for prices, specifications and official details. Ahrefs data suggests it is shifting from sites that summarise information towards the sources that produce it. A brand's core facts should sit on its own site as plain text.
Highlights
- Ahrefs studied 66.6 million ChatGPT answers from August and September; only 14.4% required a web search.
- 75.9% of the most common site: searches targeted a brand site.
- Reddit was cited in about 1 in 5 answers in March but 1 in 43 by late September.
2 min readAuthor: UNIT Journal EditorEditor-in-chief: Uğur Deniz İlhan
To understand how ChatGPT finds information, Ahrefs analysed 66.6 million free, logged-out answers from August and September and the 15 million fan-out queries behind them. A fan-out query is one of the smaller searches ChatGPT runs in the background to answer a single question. The results are published in Ahrefs' study.
Does ChatGPT search the web for every question?
No. According to Ahrefs, only 14.4% of answers required a web search. When one was run, it averaged about 1.6 fan-out queries. The study notes that free-tier models may lean more on training data to keep costs down. For many questions, then, whether your brand appears depends on how well ChatGPT already knows it.
Where does it look when it does search?
Some 6.8% of fan-out queries used the site: operator, which targets a single website. Of the most common site: searches, 75.9% pointed at a brand site and 11.9% at retailers such as Walmart, Costco and Amazon. Among the most frequent words in fan-outs, "official" rose from 10.1% in August to 13.1% in September, and "price" from 6.0% to 8.9%.
Comparison questions look different: 63.4% of citations for factual questions went to official sites, against 46.4% for "A versus B" questions.
Which sites are losing citations?
In Brand Radar data, Reddit was cited in roughly 1 in 5 ChatGPT answers in March but 1 in 43 by late September. Wikipedia fell from 1 in 7 to 1 in 35. The share moved to research bodies, regulators, brands and hands-on testers. Ahrefs reads this as ChatGPT moving from sites that summarise information to the sources that produce it.
What should companies in Turkey do?
The data covers English-language, global queries; the source has no equivalent measurement for Turkish queries. The logic still points to treating your own site as a source of record:
- Publish prices, plans, specifications, returns and support details on your own site as plain text, not buried in images or PDFs.
- Gather your official name, address, services and contact details on one clear page, consistent with your outside profiles.
- Invest in content only you can publish: original data, your own tests and customer case studies, rather than summaries of what others wrote.
- Ask ChatGPT the price, feature and comparison questions in your sector, note which sources it cites, and repeat every month.
Frequently asked
- What is a fan-out query?
- It is an extra search ChatGPT runs in the background to answer a single user question. Ahrefs found an average of about 1.6 per answer that searched.
Sources
- Ahrefs ·
Follow UNIT Journal
What's new in search, AI and technology, in your feed every day.
