If you expect ChatGPT to hand out links evenly across every search topic, the numbers tell a very different story. AI discovery isn't a tide that lifts all boats. In May 2026, Similarweb published desktop analytics tracking U.S. user interactions with OpenAI's model. The headline finding is stark: overall, ChatGPT appended an outbound web link in just 6.8% of its answers.
That single baseline number hides massive industry swings. If you manage digital strategy for an online college or an ed-tech program, you're competing in a desert—only 4.8% of educational prompts returned a source link. But if you run marketing for a resort chain or flight aggregator, the picture looks completely different. Over 22.6% of travel-related prompts triggered at least one external link.
The Misconception of Universal AI Citations
Many search marketers assume generative AI models crawl the web uniformly whenever a user asks a question. That assumption is flat wrong.
According to data published by Search Engine Journal, citation rates in ChatGPT fluctuate wildly depending on what the user is trying to accomplish. When people ask static, informational questions, the model relies on its pre-trained weight parameters. It doesn't bother calling out to the live web.
The broad 6.8% average for May 2026 creates a false sense of consistency. In reality, generative search behavior is split into high-citation topics and low-citation topics. Understanding where your industry lands on that spectrum is the first step toward building a realistic AI optimization strategy—avoiding the common trap where AI and SEO promises collapse after deals close.
Where ChatGPT Outbound Links Actually Go
Similarweb ranked nine distinct market sectors based on the percentage of ChatGPT answers that included at least one outbound citation. Travel led the entire dataset by a massive margin, with 22.6% of responses citing web sources.
Retail took second place, with 13.5% of answers carrying web links. Sports came in third at 10.7%, followed by finance at 8.0%. All four of these categories outperformed the overall 6.8% desktop average.
Here is how the vertical citation distribution broke down across U.S. desktop users in May 2026:
- Travel: 22.6% citation rate
- Retail: 13.5% citation rate
- Sports: 10.7% citation rate
- Finance: 8.0% citation rate
- Overall Average: 6.8% citation rate
- Technology: Below 6.8% citation rate
- Health: Below 6.8% citation rate
- Media & Entertainment: Below 6.8% citation rate
- Education: 4.8% citation rate
On the bottom end of the spectrum, technology, health, media and entertainment, and education all fell short of the national benchmark. Education sat dead last at 4.8%.
If you're spending thousands of dollars trying to gain direct link visibility inside ChatGPT for textbook or curriculum topics, you're fighting for a tiny sliver of overall responses. The model simply doesn't feel the need to cite external sources when answering foundational educational queries.
Why Comparative Intent Triggers the Web
What drives these massive differences between verticals? The research points straight to user intent.
ChatGPT triggers live web retrieval most heavily when users engage in three specific actions: comparing products, planning commercial purchases, or asking about rapidly changing real-time information.
When someone asks for a comparison between two hotel properties or wants real-time flight options, static model weights can't deliver accurate answers. The model has to consult live web data. Conversely, when a user asks for an explanation of basic geometry principles, static weights handle the query effortlessly without touching an external server.
Beyond how often links appear, the type of websites cited changes dramatically across sectors. Across the top 10,000 recorded desktop citations in May 2026, Similarweb categorized the domain types:
- Reviews and User-Generated Content (UGC): 28.9% of all outbound links
- News and Publisher Sites: 26.0% of all outbound links
- Retail and E-commerce Sites: 14.1% of all outbound links
While reviews and publishers make up more than half of all citations overall, sector-specific habits completely flip those averages.
In the beauty sector, retail and e-commerce websites claim 54.7% of all citations. When users ask about skincare or cosmetics, ChatGPT links directly to digital storefronts.
In travel, reviews and UGC platforms capture 54.1% of all outbound citations. AI users looking for travel advice want real traveler feedback and authentic reviews—a trend reflected in Reddit's surge in AI search visibility—and the model aligns its citations accordingly.
In finance, specialized financial websites take 36.6% of citations, while news and publishing outlets claim 28.0%. Generic retail or UGC links play virtually no role in financial AI responses.
The Nine-Month Surge in External Linking
While citation behavior remains uneven across topics, ChatGPT's overall propensity to link out has increased substantially over time.
In June 2025, the overall citation rate across U.S. desktop queries stood at just 1.3%. By May 2026, that figure reached 6.8%—marking more than a fivefold increase in under a year.
However, this growth didn't happen in a smooth, straight line. The citation rate surged past 6.0% in October 2025, dropped back toward 4.5% by February 2026, and then rebounded during the spring.
This shifting trajectory aligns with findings from other industry trackers. For example, an analysis by Resoneo in April noted that ChatGPT cited roughly 20% fewer total websites per response following the rollout of the GPT-5.3 Instant update.
It's critical not to confuse these two metrics. Similarweb measures binary citation presence—whether a response contains at least one external link. Resoneo measured citation density—the total count of links crammed into a single answer. A model can easily increase the frequency with which it adds links across queries while simultaneously reducing the number of external links provided per response.
What Gets Cited Varies Completely by Niche
To evaluate these numbers accurately, you have to understand the study's scope and methodology constraints. Similarweb's figures reflect U.S. desktop web interactions based on algorithmic estimations and extrapolations from third-party data panels. The analysis explicitly excludes mobile apps, desktop software clients, and direct API traffic.
Even with those boundaries in mind, the strategic takeaway for brands is crystal clear: there is no universal AI search optimization playbook.
If you apply a generic content strategy across different industries, you'll waste time and budget. In travel, winning visibility means getting featured on user review platforms and active community forums, since UGC dominates 54.1% of travel citations. In beauty or consumer electronics, optimizing direct product feeds and structured e-commerce data is paramount, because store pages command the majority of links.
Brands must also ensure their technical product data is easily digestible by AI crawlers. As detailed in our guide on building enterprise knowledge graphs with schema, models favor structured, unambiguous inputs when choosing which sources to cite during product evaluations.
Tactical Takeaways for Optimization
How should search teams and brand marketers adjust their efforts based on these findings? Here are four practical adjustments to make immediately:
-
Know your vertical baseline. Stop comparing your AI visibility to generic industry averages. Benchmark your performance against your specific topic category. If you operate in finance, target an 8% potential link pool. If you operate in education, prepare for a environment where over 95% of model answers contain zero external links.
-
Optimize for comparative query formats. Because ChatGPT goes to the web primarily for product comparisons and real-time purchase planning, build content that directly answers comparative questions. Structured spec tables, transparent pricing comparisons, and clear pros-and-cons breakdowns make live retrieval easier for the model.
-
Target the dominant source types in your niche. Align your off-page PR and link-building efforts with the specific domain types ChatGPT prefers for your topic. In travel, invest in community and review management. In finance, prioritize coverage on authoritative financial portals and news sites.
-
Track citation frequency alongside position. Don't just track whether your brand ranks inside an AI response. Keep tabs on whether the model is actively fetching external web data for your high-intent keywords over time, as model updates routinely shift web retrieval frequency.
Understanding ChatGPT's selective citation behavior allows you to deploy resources intelligently—focusing your optimization work where AI models actually reach out to the web.
