ChatGPT Searched the Web in Only 14.4% of 66.6 Million Answers, Ahrefs Fan-Out Study Finds

Ras Al Khaimah, United Arab Emirates, 14 October 2026.

ChatGPT ran a web search for only a small share of the questions it answered in August and September 2026, but when it did search, its queries leaned increasingly toward official sources, prices and the current year. That is the headline from a study published by Ahrefs on October 12, 2026, written by Louise Linehan, which analysed 66.6 million free, logged-out ChatGPT responses on the GPT-5.6 Luna model and roughly 15 million fan-out queries.

Fan-out queries are the searches a model runs behind the scenes to ground an answer. For SEO teams they are the closest thing to keyword data for AI answers, which is why this dataset matters.

Key findings

• Search is the exception. Only 14.4% of the 66.6 million responses triggered a web search.
• Few queries per answer. Responses that searched ran about 1.6 fan-out queries on average.
• Site searches target brands. 6.8% of fan-outs used the site: operator, spread across about 116,000 domains, and 75.9% of the most common site: searches pointed at a brand site.
• "Official" is rising. The word appeared in 13.1% of English fan-outs in September, up from 10.1% in August.
• Official sources win factual questions. In a separate test of 20,000 US prompts, 84.4% of factual answers about pricing, features or company details cited at least one official source.
• Community citations fell. In Ahrefs' Brand Radar data, ChatGPT's Reddit citations dropped 88% between March and September 2026.

How fan-out vocabulary shifted in one month

Ahrefs compared the most common words in English fan-out queries between August and September. Terms tied to recency, official sources and commercial detail grew fastest.

Term in fan-out query	Share, August 2026	Share, September 2026	Change
2026	17.6%	22.9%	+5.2 pts
official	10.1%	13.1%	+3.0 pts
price	6.0%	8.9%	+2.8 pts
reviews	2.7%	3.6%	+0.9 pts
meaning	2.9%	2.3%	-0.6 pts

Ahrefs puts the move in "price" from 6.04% to 8.88% of fan-outs at a relative rise of about 47% in a single month, while definitional terms such as "meaning" declined. In other words, the searches ChatGPT runs are becoming more commercial and more time-bound.

Official sources and question type

The study also tested whether ChatGPT cites a brand's own site when asked about that brand. The answer depended on the question.

Question type	Answers citing at least one official source	Share of citations to official sites
Factual (pricing, features, company details)	84.4%	63.4%
Comparison ("A vs B", "which is cheaper")	68.3%	46.4%

For comparison queries, most citations went to sites other than the brands being compared. That is where reviews, publishers and comparison pages still earn visibility.

Citation shifts: Reddit, Wikipedia and primary sources

Using Brand Radar citation data for US ChatGPT answers from March to September 2026, Ahrefs reports that Reddit citations fell 88%, Wikipedia 80% and YouTube 27%. Over the same period PubMed climbed from 64th to 4th among the most-cited domains. Search Engine Journal's coverage framed the shift as a tilt toward primary sources and noted that citation ranks do not measure traffic.

How this compares with other fan-out studies

Fan-out counts vary widely by setup, and Ahrefs is explicit that its numbers are lower than several smaller studies.

Study	Setup	Fan-outs	site: share
Ahrefs	GPT-5.6 Luna, free, logged out, 66.6M responses	About 1.6 per searching response	6.8%
Chris Long (Nectiv)	GPT-5.6 Sol via API, about 4,000 prompts	7.61 per prompt	64%

Ahrefs attributes the gap mainly to scale and prompt type: its data covers broad prompts from free, logged-out sessions, which search less than paid or reasoning models, while many other studies used commercial "best X" prompts.

Methodology and limitations

This is our coverage of a third-party study; we did not replicate the data. As described by Ahrefs, the main sample was 66.6 million free, logged-out GPT-5.6 Luna responses collected in August and September 2026. Official-source behaviour was tested on 20,000 US prompts, and citation trends came from Brand Radar US data covering March to September 2026. Ahrefs lists its own limits: one model, two months, logged-out sessions only, and site: data that shows what ChatGPT looks up rather than what it cites. OpenAI changes retrieval often, so these are snapshots.

What this means for SEO teams

• Publish facts in plain text on crawlable pages. Pricing, specifications and support details on your own site are what "official" and "price" fan-outs look for.
• Keep pages dated and current. With "2026" in nearly a quarter of September fan-outs, visible update dates and current-year information matter.
• Win comparisons off-site. Fewer than half of citations for comparison questions went to official sites, so third-party reviews and comparison coverage still carry weight.
• Do not build a plan on Reddit alone. Its citation share fell sharply in this dataset.
• Re-check monthly. Ahrefs recommends re-running fan-out analysis regularly because the mix moved noticeably within a single month.

Conclusion

Ahrefs' data suggests that most ChatGPT answers in free, logged-out sessions never touch the live web, and that when they do, the model runs only a handful of searches that increasingly ask for official, priced and current information.

For brands, that puts the emphasis on accurate first-party pages for factual questions and on independent coverage for comparisons. The numbers come from one model and two months, so teams should treat them as direction, not a fixed rule.

Questions and answers

How often does ChatGPT search the web?

In Ahrefs' sample of 66.6 million free, logged-out GPT-5.6 Luna responses from August and September 2026, 14.4% triggered a web search.

How many fan-out queries does ChatGPT run?

Ahrefs found an average of about 1.6 fan-out queries for each response that searched. Nectiv, testing about 4,000 prompts through the API on a different model setting, reported 7.61 per prompt, so results depend heavily on setup.

Does ChatGPT prefer official sources?

For factual questions about pricing, features or company details, 84.4% of answers in Ahrefs' 20,000-prompt US test cited at least one official source, and 63.4% of citations went to official websites. For comparison questions the figures were 68.3% and 46.4%.

Is ChatGPT citing Reddit less?

Ahrefs reports that ChatGPT's Reddit citations dropped 88% between March and September 2026 in its Brand Radar US data, with Wikipedia down 80% and YouTube down 27%.

###