Quick answer
Google searches from San Francisco for agent-tool phrases such as langchain, n8n and "ai agents" run at 9.5 times the city's share of the US population (median of 22 phrases we chose), and Seattle is similar at 8.6 times. This measures Google Ads search volume, not use, and no public data measures AI agents.
Key takeaways
- For 22 agent-tool phrases we chose, San Francisco's median is 9.5 times its population share and Seattle's is 8.6 times: similar, so we do not rank them.
- For the same phrases the medians are 4.8 times for Austin, 4.7 times for Boston, 2.4 times for New York and 2.1 times for Chicago.
- We could not detect a difference between agent-tool phrases and small-business-service phrases in San Francisco, and for 17 of 18 San Francisco service phrases the index is only a lower limit set by the data's floor of 10 searches a month, so this data cannot show whether the city searches them more than its population share.
- This is Google Ads search volume for phrases we picked. It is not use, purchase or adoption by businesses.
Where do people search most for AI agent tools?
Among the six cities we looked at, for the agent-tool phrases we chose, Google searches from San Francisco run at 9.5 times the city's share of the US population, and Seattle is similar at 8.6 times. Both figures are medians across 22 phrases. This measures Google Ads search volume for 44 phrases we picked, not how many businesses use AI agents, and no public dataset measures AI agents as such.
Read the figure as a ratio of two shares. If a city's share of US searches for a phrase equals its share of the US population, it scores 1 time. Here are the medians for the 22 agent-tool phrases (ai agents, n8n, langchain and similar), September 2025 to August 2026.
| City | Searches for agent-tool phrases, as a multiple of its population share (median of 22 phrases) |
|---|---|
| Austin | 4.8 times |
| Boston | 4.7 times |
| Chicago | 2.1 times |
| New York | 2.4 times |
| San Francisco | 9.5 times |
| Seattle | 8.6 times |
Part of San Francisco's 9.5 times may be pushed up by a reporting floor. The data we received never shows fewer than 10 searches a month, and for 9 of the 22 phrases that can push San Francisco's index up. On the other 13 of 22 phrases, San Francisco's median is 8.6 times its population share. That is a separate quantity from Seattle's 8.6 times, which covers all 22 phrases.
Treat San Francisco and Seattle as similar. A gap of one or two times between two cities is not something this data establishes, so we do not rank them.
Ten of the 22 phrases are product names, such as n8n, langchain, llamaindex and flowise. A search for a product name is a product lookup, not research into AI agents, and the list is our choice, not a sample of all searches. Without those ten, San Francisco's median is close to its all-phrase median.
Three more cautions apply to every figure in the table.
- There are no control phrases. We included no phrases unrelated to AI, so the figures cannot separate interest in AI from San Francisco searching more of many things. That fits a city effect as well as an AI effect. Check for such phrases before reading any multiple as interest in AI.
- The denominator is residents. Searchers include commuters, visitors and workers, and Google's city boundary may not match the census one. We cannot say how much that moves the figure.
- Each figure is a median of ratios of rounded numbers. The volumes we received are rounded bands, and we divided city by US figures month by month before taking the median.
Do small-business AI services get searched less than agent tools?
We could not detect a difference. We expected San Francisco to search agent-tool phrases much more than phrases for small-business AI services such as an AI receptionist or an AI answering service. The data did not confirm that, and it did not rule it out either.
We give no headline figure for the service phrases. For 17 of the 18 usable San Francisco service phrases, the index is only a lower limit set by the floor of 10 searches a month, so this data cannot show whether San Francisco searches those phrases more than its population share. The floor also pulls the tools-versus-services difference toward zero. When we left the floor phrases out, the difference was larger, but that check was exploratory, rests on only 8 service phrases and is not a finding. Without product names, the difference gets smaller.
We also could not detect that San Francisco's tools-versus-services gap differs from the other cities' gaps.
What did we measure, and what did we start from?
We measured Google Ads search volume for 44 phrases we chose, 22 for agent tools and builders and 22 for AI services sold to small businesses. We pulled monthly volumes for the US and for six city locations (San Francisco, New York, Seattle, Austin, Boston, Chicago), September 2025 to August 2026, through DataForSEO, a service that provides access to Google Ads keyword data. The numbers are Google's estimates. We divided each city's share of US searches by the city's share of US population, from the Census Bureau's 2024 American Community Survey.
We started from the assumption that AI agents are mainly a San Francisco developer story, with small-business AI services far less so. Here is what the data showed: for the agent-tool phrases we picked, search volume from San Francisco and Seattle is large relative to their populations, and we could not detect a difference between tool and service phrases. The direction was partly visible to us in a 12-phrase test before we froze the method, so the hypotheses were labelled confirmatory in form only. The method was frozen in a written protocol before analysis.
Technical detail (skippable): the tools-minus-services difference in San Francisco was 0.20 (95% CI -0.32 to 0.71), p = 0.412 (Holm-corrected for three tests, natural-log units). A confidence interval is the range the true figure plausibly sits in; this one is wide. The San Francisco gap compared with the other cities was 0.17 (95% CI -0.20 to 0.53), p = 0.412, with effective n = 108. Across n = 37 phrases, San Francisco's index correlates with the average of New York, Seattle and Austin at r = 0.86 (95% CI 0.77 to 0.92); in plain terms, phrases that stand out in those cities tend to stand out in San Francisco, but every city shares the same US figure and the phrase list is ours, so this is not evidence of local interest.
What can this data not show?
It cannot show use of AI agents, who is searching, or why. Beyond the cautions above, the limits are:
- The figures are searches, not use, people or businesses, and nothing here explains the pattern.
- English-language Google searches only. Chat assistants, other search engines and voice searches leave no trace here.
- Six cities only. Nothing here describes other cities, small towns or rural areas.
- September 2025 to August 2026 only, and 44 phrases that we chose.
- Phrases with fewer than 6 usable months in a city were dropped (four San Francisco service phrases), which favors higher-volume phrases.
- The floor of 10 searches a month in the data limits what the service phrases can show.
How can you check these numbers?
Compare against the public sources. The population denominators come from the Census Bureau's 2024 American Community Survey 1-year table B01003. The search volumes are Google Ads estimates; the DataForSEO documentation describes the endpoint, which returns approximate monthly searches for a keyword in the locations you target, and Google's Keyword Planner page describes the tool the figures come from. The raw search pulls are paid-account data, so no download is offered here.
For names rather than searches, see our count of San Francisco business registrations with AI in the name, which measures names in a public register.
What can you do with this?
Use it as a reading rule, not a scoreboard. When you see a chart of "AI interest by city", check three things before you rely on it.
- Ask whether it includes phrases unrelated to AI. Without them, a city that searches more of everything looks AI-heavy, and no single multiple tells you interest in AI.
- Do not read gaps of one or two times between two cities as a difference. Here that is the gap between San Francisco and Seattle.
- Ask whether the phrases are product names, which are lookups, and whether a low-volume phrase sits at the reporting floor.
This data cannot show whether searches for AI services for small businesses in your city differ from its population share, because most of those phrases sit at the reporting floor. For how many businesses report using AI by metro, see our Census-based look at AI use by metro, which measures the share of businesses reporting AI use.
Analysis by Deivy Hernandez, from Census data and Google Ads search estimates.
FAQ
Does a high search figure mean businesses in that city use AI agents?
No. Searching a phrase can be curiosity, study, a job task or a product lookup. Nothing in this data records what anyone did next, and no public dataset counts AI agents in use. For business use of AI in general, the Census Bureau survey covered in our metro article is the better source.
Why only six cities?
Each city needs its own paid pull of Google Ads data, and the cities were fixed in the written protocol before analysis. A city not on the list is not shown to be lower.
Can I run this for my own city?
Yes, with a Google Ads keyword tool or a data provider, using the same steps: take a city's share of US searches for a phrase and divide it by the city's share of US population. Add a few phrases that have nothing to do with AI, which this study lacked, so you can see whether your city simply searches more of everything.
Is the search figure the number of searches per person?
No. It compares two shares: the city's slice of all US searches for a phrase, and the city's slice of the US population. A value of 1 would mean searching in proportion to population.