Two brands can do identical work on AI visibility and get results that differ by a factor of four. Not because one executed better. Because they sell different things.
Similarweb's 2026 report on generative AI, covered this week by Search Engine Journal, measured how often ChatGPT includes citations in its answers. In May 2026, travel answers carried citations 22.6 percent of the time. Education answers carried them 4.8 percent of the time.
Same model. Same week. Nearly five times the difference, decided entirely by topic.
The Baseline Moved, Unevenly
The overall trend is worth holding onto first. ChatGPT's citation rate went from roughly 1.3 percent in June 2025 to 6.8 percent by May 2026. More than a fivefold increase in under a year.
That is the number people quote when they argue AI search will send traffic. It is real, and it is also an average that hides the thing you actually need to plan around.
The report found ChatGPT reaches for the web most when the question involves product comparisons, purchase planning, or anything that changes over time. It stays inside its own weights when the question has a stable, well-established answer.
That logic is not arbitrary. It is a reasonable model deciding when its training data is likely to be stale or thin. But the consequence for a marketing team is blunt: if you sell something the model considers settled knowledge, you are competing for a much smaller pool of outbound links, no matter how good your content is.
The source mix shifts by category too. In travel, 54.1 percent of citations went to reviews and user generated content. In beauty, 54.7 percent went to retail and ecommerce sites. In finance, 36.6 percent went to finance-specific publications.
Three verticals, three completely different answers to the question "what kind of page gets cited here."
Why the Generic GEO Advice Keeps Failing
Most of what gets sold as AI visibility strategy right now is a single checklist applied to every client. Structured data, clear headings, an FAQ block, an llms.txt file, a few authoritative mentions.
None of that is wrong. All of it is insufficient, because it optimizes the page while ignoring the category.
If you run a travel brand, this data says your competitive surface is review platforms and community content. Winning means being present and well-reviewed where real people write, not just polishing your own site. If you run a beauty brand, the surface is retail listings and product pages on marketplaces you do not control.
If you run an education business, the honest read is that ChatGPT will mostly answer without you. Your play is not citation share. It is being the source the model absorbed during training, which is a slower and more brand-driven game.
I made a related argument in AI mentions now move real traffic. The mention is worth something. What this new data adds is that the price of a mention is not the same in every market.
The Measurement Problem Underneath
Here is the operational reality most teams have not solved. You cannot tune for a citation pattern you have never measured in your own category.
Nobody in your analytics stack reports which sources ChatGPT pulled when answering a buyer's question about your product. Google Analytics shows you the visit that happened. It cannot show you the twenty answers where a competitor was cited and you were not.
This is exactly the gap we built GEOflux.ai for. It maps which sources the engines actually pull in your category, not just whether your name appeared, so the work goes into the surfaces that carry weight for your specific market rather than a generic checklist.
Whatever you use to measure it, the sequence matters. Measure your category's citation behavior first. Decide where the model looks for your kind of answer. Then invest there.
The alternative is the pattern I described in buying AI citations is the new link farm, where teams chase mention volume because it is the only number they can see, and it turns into the same low-quality spiral link building went through fifteen years ago.
The Practical Read
Three questions worth answering before your next content quarter.
First, does the model even go to the web for your category, or does it answer from memory? That single fact determines whether citation work or brand work is the better use of your budget.
Second, when it does cite, what type of source wins in your vertical? Reviews, retailers, publishers, and forums are not interchangeable, and the split is measurable.
Third, are the surfaces that get cited in your category ones you can influence at all? Sometimes the honest answer is that you need partnerships and reviews rather than another blog post.
The averages will keep rising and the headlines will keep celebrating them. Your business does not operate on the average. It operates in one category, with one citation rate, and that number was set before you started writing.