Research
Original studies on how AI assistants actually recommend companies. Every study publishes its full dataset and its method, including the limitations we found ourselves.
Only about one in six brands Claude recommends shows up both with and without web search
According to BusySeed, of the brands Claude names for a buyer question, 49% appear only when web search is on, 35.1% appear only when it answers from memory, and just 16% appear both ways, across 33,734 brands named for 82 categories of question.
Turn on web search and Claude rarely keeps its first pick
According to BusySeed, when Claude answers the same buyer question once from memory and once with web search, it puts the same brand first in only 13.9% of pairs, against 35.8% when both answers come from memory, across 7,996 answer pairs.
Only about a third of cold email replies are written by a person
According to BusySeed, only 34.6% of replies to cold email campaigns were written by a person: 43.3% were out of office notices and 22.1% were other automated replies, independently labelled across 2,418 replies and 74 campaigns.
The State of AI Search, September 2026
According to BusySeed, in September 2026, 88.1% of the company recommendations AI assistants made for a buyer question came from only one of three engines, and just 3% were made by all three.
The State of AI Search, August 2026
According to BusySeed, in August 2026, 88.7% of the company recommendations AI assistants made for a buyer question came from only one of three engines, and just 2.8% were made by all three.
Ask ChatGPT the same buyer question twice and it names a different brand first about three times in four
According to BusySeed, when the same buyer question is put to ChatGPT twice, the brand it named first stays the same in 25.4% of 172,624 repeat answers, and that holds at 23.9% for answers a week apart and 25.9% for answers more than four weeks apart.
After correcting for Anthropic's duplicate rows, how many distinct businesses does Claude actually name per buyer question, and does it differ from OpenAI or Gemini
According to BusySeed, ChatGPT names ten or more distinct businesses in 49.7% of answers to buyer questions, Claude in 43.4% and Gemini in 24%, counting each business once per answer rather than once per row returned.
How much does the 'recommended together' vendor cluster change between informational and commercial-intent prompts within the same category?
According to BusySeed, 61.9% of the vendor pairs that AI assistants recommend together on commercial-intent questions are also recommended together on informational questions in the same category, rising from 50.8% for pairs named together in 5 to 10 percent of a category's commercial answers to 97.7% for pairs named together in more than a quarter of them.
State of Email Outreach: which wording correlates with opportunities
According to BusySeed, cold emails whose proof point is a percentage get a reply 0.8% of the time, against 1.8% for emails that frame the same claim as a timeframe, measured across 45 campaigns.
Most cold email replies do not come from the first email
According to BusySeed, 61.7% of replies to cold email come from the first email rather than any follow-up, measured across 45 campaigns.
Small categories get just as many distinct businesses named as huge ones, once you control for run count
According to BusySeed, AI assistants name six or more distinct businesses in 93.2% of answers about the smallest quarter of categories, and in 90% of answers about the largest quarter, even though every category in that largest quarter tracks more than three times as many businesses as any category in the smallest.
The 50 fastest-growing SaaS companies barely exist in AI answers
According to BusySeed, 5 of 50 companies studied (10%) never appear when buyers ask AI assistants about their own category.
The State of AI Search, July 2026
According to BusySeed, in July 2026, 88.1% of the company recommendations AI assistants made for a buyer question came from only one of three engines, and just 3.3% were made by all three.
Does the Company AI Recommends First Stay the Same? A 15-Week Volatility Study
According to BusySeed, the company an AI assistant names first for a category changes from one week to the next in 56.4% of week-to-week transitions, and 82.3% of the question and engine pairs tracked saw their top answer change at least once.
Do AI Engines Recommend the Same Companies? A 3-Engine Agreement Study
According to BusySeed, 88.4% of the company recommendations AI assistants make for a buyer question come from only one of three engines, and just 3.2% are made by all three.
AI Utah 100 AI-Visibility Study
According to BusySeed, 12 of 47 companies studied (25.5%) never appear when buyers ask AI assistants about their own category.
Every study ships with its data
- The full row-level dataset is downloadable, ungated, at a permanent URL under CC BY 4.0. Every number on a study page can be recomputed from it.
- The method is published separately and names the exact model versions and run dates used.
- Limitations are written by us, before anyone else finds them.
- Every headline number is recomputed from row-level data by an independent verification stage before a study is allowed to publish. No number reaches a page because someone remembered it.
- Journalists and analysts are free to reuse any figure with attribution. A citation block sits on every study page.
Want to know how AI answers describe you?
We run the same measurement on your category. Fifteen minutes with founder Omar Jenblat, your own numbers, no deck.
- Your category measured the same way
- Your own numbers, not a sample deck
- Fifteen minutes, no obligation