Inside ChatGPT’s search habits: brand sites, prices and partner citations

If your ecommerce pages hide prices, specs or official support details behind scripts or PDFs, ChatGPT is unlikely to pull them into answers. New data from Ahrefs shows the model searches the web rarely — and when it does, it targets brand sites and primary sources. That shift changes which pages earn AI-driven visibility and where marketing and SEO teams should focus their effort.

What the dataset found

Ahrefs analyzed 66.6 million free, logged-out GPT-5.6 Luna responses from August–September 2026 and extracted roughly 15 million “fan-out” queries — the extra lookups ChatGPT runs while composing a reply. Key measurements:

  • Only 14.4% of responses triggered a web search; when ChatGPT did search, it ran about 1.6 fan-out queries on average.
  • 6.8% of fan-out queries used a site: operator to search a single domain; those site: searches spanned roughly 116,000 domains.
  • About 75.9% of the most common site: searches targeted brand websites (for example apple.com, ikea.com, nike.com) rather than forums, review aggregators or retailers.
  • Language inside fan-outs shifted between August and September: the token “official” rose from 10.1% to 13.1% of English fan-outs, while “price” increased from 6.04% to 8.88% (a relative rise near 47%). “Review” queries also grew month‑over‑month.
  • Citation trends from March–September 2026 show large drops for some aggregators and platforms: Reddit citations fell roughly 88%, Wikipedia declined from about 1 in 7 answers to 1 in 35, and YouTube citations dropped from around 1 in 370 to 1 in 500. At the same time, government, regulator and original-research sites gained share.
  • OpenAI partner publications increased their citation share among the 200 most-cited domains: combined partner share rose from 7.5% in March to 15.8% in July and held near 12.5% in late September.

Why marketers and product teams should care

First, ChatGPT usually answers without a live search. That means an AI answer may reflect knowledge already encoded in the model rather than fresh retrieval. Brands need to build that knowledge in accessible, crawlable formats: clear product facts, structured data, and visible press or partner mentions help the model learn and repeat accurate details.

Second, when ChatGPT does perform lookups it favors first‑party sources. Three quarters of site: searches go to brand domains, and the fan-out language increasingly requests “official,” “price” and “review” pages. If your product pricing or specs live behind JavaScript, images or PDFs, they are effectively invisible during these lookups.

Third, citation mix is tilting toward primary sources. Content that simply aggregates or summarizes others’ work now faces lower odds of being cited directly. Original data, regulatory filings, test results and direct company statements are more likely to be surfaced and credited.

Finally, content licensing matters: partner publications already show higher citation presence. Coverage in those outlets can increase the chance your content is cited in AI responses as well as driving standard referral traffic.

Practical steps to improve AI visibility

  • Audit product and support pages for plain-text facts. Ensure prices, SKUs, specs and warranty information appear as readable text on crawlable pages; avoid burying them in images or dynamic widgets.
  • Publish original material only you can own: product tests, primary research, official statements, and regulator filings attract citations more reliably than summaries.
  • Align content templates to fan-out intent signals. Track which modifiers (price, review, official, specs) appear in your vertical’s fan-outs and surface those answers on dedicated pages.
  • Pursue placements in high-value partner publications. Ahrefs’ data show partner titles increasing citation share; earned coverage there helps both AI and human discovery.
  • Do not rely solely on social or aggregator exposure. Given the declines in Reddit, Wikipedia and YouTube citations, these channels may be weaker paths to appearing in AI answers than before.

What to watch next

Two variables will determine how these patterns evolve: how often OpenAI expands live retrieval for logged-in or paid tiers, and whether future model updates shift the balance between preloaded training data and live lookups. Either change would alter how often ChatGPT needs to run fan-out queries and which domains it targets. For now, brands that make their factual pages machine-readable, publish unique data and secure partner coverage will be best positioned to appear in ChatGPT-driven answers.