How Chat Radar Local AI Search Trends works
Methodology: Michael Notbohm, Founder & CEO, Chat Radar · LinkedIn · ORCID 0009-0003-3320-5803
Michael Notbohm is the founder and CEO of Chat Radar, an AI visibility analytics platform for local businesses, and publishes Chat Radar Local AI Search Trends, a weekly dataset on how Google AI Overviews, ChatGPT and Perplexity cite and recommend local businesses. His career in marketing technology includes roles at Yellowpages.com and LexisNexis, and his work focuses on local search, answer engine optimization, and how generative AI surfaces small businesses.
Who runs this and why
Chat Radar Local AI Search Trends is published by Chat Radar, which sells AI visibility analytics and services to local businesses. We have a commercial interest in how AI search treats local businesses.
The index reports what the panel shows whether or not it helps that interest. Flat or unremarkable weeks are published the same way as dramatic ones, and the rules below are fixed before each week is collected.
The panel
A fixed panel of 116 US local businesses: 50 dental practices, 45 car dealerships and 21 other local businesses, including home services. Every business is one customers visit or one that serves a defined local area. Marketing, design and consulting firms, online-only suppliers and national lenders are not included.
The panel is frozen and versioned (this is panel version 1). It changes only by publishing a new version. We report aggregates only and never name a business.
The week
A week runs Monday to Sunday and is named by its Monday: the page for the week of September 28, 2026 is /trends/2026-09-28. Data is collected on the Monday and the week is published on the Tuesday, once it passes the checks below. A published week page never changes except through a logged correction.
What we collect
- Google AI Overviews. Each business has up to eight local questions a customer might ask, built from its category and market and never naming the business. We run them as Google searches from the business’s metro area (or its state when no metro applies) through a commercial search-results data provider, and record whether an AI Overview appeared, every source it listed, which sources it cited in the answer text, links written into the answer, and whether a local map pack appeared.
- ChatGPT and Perplexity. Four of those questions are asked through the platforms’ own search-enabled APIs. We record whether the answer names the business and which sources it cites.
- Business websites. Our identified crawler (ChatRadarTrendsBot) reads each site’s robots.txt, llms.txt and homepage once a week: which AI crawlers are blocked, whether the homepage answers at all, its structured data and whether it has a sitemap.
- Google Business Profiles. We read each business’s public profile through Google’s own API: name, phone, hours, website, category, rating and review count, and compare it with the week before and with the business’s own website.
Businesses in the same category and market share questions, so one answer can apply to several businesses. Each distinct answer is collected once. Measures of answers count it once; measures of businesses count it once for each business it applies to. Every definition below says which.
Metric definitions
Google AI Overviews
- Searches showing an AI Overview
aio_presence_rate - Share of panel searches where Google showed an AI Overview. Counted per search: a search asked for several businesses counts once.
- Sources per AI Overview
aio_avg_refs - Average number of distinct sources listed in an AI Overview’s source panel. Counted per AI Overview.
- Sources cited in the answer text
aio_inline_coverage - Of all sources listed in AI Overview source panels, the share also cited inside the answer text. Counted per source.
- Own website cited
aio_own_site_cited_rate - Share of AI Overviews shown on a business’s searches that cite that business’s own website. Counted per business: a shared AI Overview counts once for each business it applies to.
- Own Business Profile cited
aio_own_profile_cited_rate - Share of AI Overviews shown on a business’s searches that cite that business’s own Google Business Profile, matched by business name. Counted per business.
- Sources: directories and review sites
aio_ref_mix_directory_review - Share of AI Overview sources that are directories and review sites. Counted per source.
- Sources: publishers and media
aio_ref_mix_publisher_media - Share of AI Overview sources that are publishers and media. Counted per source.
- Sources: other Google pages
aio_ref_mix_google - Share of AI Overview sources that are other Google pages. Counted per source.
- Sources: Google Business Profiles
aio_ref_mix_google_business_profile - Share of AI Overview sources that are Google Business Profiles. Counted per source.
- Sources: links into Google AI Mode
aio_ref_mix_google_ai_mode - Share of AI Overview sources that are links into Google AI Mode. Counted per source.
- Sources: other businesses’ websites
aio_ref_mix_other_business - Share of AI Overview sources that are other businesses’ websites. Counted per source.
- Sources: other sites
aio_ref_mix_other - Share of AI Overview sources that are other sites. Counted per source.
- Searches showing a map pack
local_pack_presence_rate - Share of panel searches where Google showed a local map pack. Counted per search.
ChatGPT and Perplexity
- Business named in the answer
llm_brand_named_rate - Share of answers to a business’s local questions that name that business. The questions never name it. Counted per business: a shared answer counts once for each business it applies to.
- Citations per answer
llm_avg_citations - Average number of distinct sources an answer cites. Counted per answer.
- Own website cited
llm_own_site_cited_rate - Share of answers to a business’s local questions that cite that business’s own website. Counted per business.
- Citations: directories and review sites
llm_ref_mix_directory_review - Share of answer citations that are directories and review sites. Counted per citation.
- Citations: publishers and media
llm_ref_mix_publisher_media - Share of answer citations that are publishers and media. Counted per citation.
- Citations: other Google pages
llm_ref_mix_google - Share of answer citations that are other Google pages. Counted per citation.
- Citations: Google Business Profiles
llm_ref_mix_google_business_profile - Share of answer citations that are Google Business Profiles. Counted per citation.
- Citations: links into Google AI Mode
llm_ref_mix_google_ai_mode - Share of answer citations that are links into Google AI Mode. Counted per citation.
- Citations: other businesses’ websites
llm_ref_mix_other_business - Share of answer citations that are other businesses’ websites. Counted per citation.
- Citations: other sites
llm_ref_mix_other - Share of answer citations that are other sites. Counted per citation.
Business websites
- Sites blocking GPTBot
site_ai_bot_blocked_rate_gptbot - Share of business websites whose robots.txt disallows GPTBot from the whole site. Counted per business.
- Sites blocking ClaudeBot
site_ai_bot_blocked_rate_claudebot - Share of business websites whose robots.txt disallows ClaudeBot from the whole site. Counted per business.
- Sites blocking PerplexityBot
site_ai_bot_blocked_rate_perplexitybot - Share of business websites whose robots.txt disallows PerplexityBot from the whole site. Counted per business.
- Sites blocking Google-Extended
site_ai_bot_blocked_rate_google_extended - Share of business websites whose robots.txt disallows Google-Extended from the whole site. Counted per business.
- Sites blocking any AI crawler
site_any_ai_bot_blocked_rate - Share of business websites whose robots.txt fully blocks at least one of ten AI crawlers. Counted per business.
- Sites with llms.txt
site_llms_txt_rate - Share of business websites serving an llms.txt file. Counted per business.
- Sites with LocalBusiness markup
site_localbusiness_schema_rate - Share of business homepages carrying schema.org LocalBusiness-type structured data. Counted per business.
- Sites with a sitemap
site_sitemap_rate - Share of business websites with a sitemap listed in robots.txt or at /sitemap.xml. Counted per business.
- Homepages refusing our crawler
site_blocked_status_rate - Share of business homepages that answered our identified crawler with 403 or 503, which can mean bot blocking at the network level that robots.txt does not show. Counted per business.
Google Business Profiles
- Profile changes observed
gbp_profile_change_rate - Share of Business Profiles whose name, phone, hours, website or category changed since the previous week. We cannot tell who made a change. Counted per business.
- Profile phone differs from the website
gbp_phone_mismatch_rate - Share of Business Profiles whose phone number matches none published on the business’s own website. Counted per business.
- New reviews per profile (median)
gbp_median_new_reviews - Median change in a Business Profile’s review count since the previous week. Counted per business.
- Average rating
gbp_avg_rating - Average Google rating across panel Business Profiles. Counted per business.
When a number is not published
A figure is published only when at least 5 businesses and, for AI answers, at least 100 answers stand behind it. Smaller cells read “Not enough data”, and no count is published for them. Segments are all businesses, dental practices, car dealerships and other local businesses; no smaller cut is ever published.
Each data source must be complete enough to publish: 80% of AI Overview searches and 90% of everything else. A source that falls short is shown as not collected for that week, never published from partial data and never shown as zero. A week is delayed only if fewer than two sources are complete.
Before publishing, each figure is compared with the four weeks before it. A rate that moves more than 5 points, or a count that moves more than 2.5 standard deviations, holds the week for a person to review before it goes out.
What we can’t observe
- Who changed a Business Profile. We see that a profile changed, not whether the owner or Google made the change, so we report “profile changes observed”.
- Bot blocking at the network level. robots.txt shows a site’s stated rules; a firewall that refuses crawlers does not appear there. We report homepages that refuse our identified crawler separately, as a partial signal.
- Personalized results. Searches run without a signed-in user from a fixed location, so a real customer may see a different answer.
Corrections
If a published number turns out to be wrong, we correct it on its week page, keep the old value visible struck through beside the new one, and add a dated note. Every correction is listed in the corrections log.
License and citation
The data is licensed CC BY 4.0. Reuse it with attribution to “Chat Radar Local AI Search Trends” and a link to chatradar.ai/trends. Every week page has citations in APA, MLA, Chicago and BibTeX.
Changelog
Panel v1 and methodology v1 · October 2, 2026 · v1
The index measures a fixed panel of local businesses that customers visit or that serve a defined local area, all in the United States. Marketing, design and consulting firms, online-only suppliers and national lenders are not in the panel.
Searches run from each business’s metro area where one can be determined, otherwise from its state.
Businesses in the same category and market are asked the same questions. Each distinct answer is collected once. Measures of answers (such as how often an AI Overview appears) count it once; measures of businesses (such as how often a business’s own website is cited) count it once for each business it applies to.
A figure is published only when at least 5 businesses and, for AI answers, at least 100 answers stand behind it. A data source that is incomplete in a week is marked as not collected for that week rather than published from partial data.
Earlier series (v0): what it includes · October 2, 2026 · v0
The earlier series comes from a different, smaller sample of businesses whose searches were run from a national location, not from their own market. It is shown as a separate line and never combined with the panel’s weekly figures.
It begins with the week of August 17, 2026. An improvement to how those searches were collected took effect on August 16, and weeks before it are not comparable.
Only two measures are published for it: how often an AI Overview appeared, and an approximate mix of the sources AI Overviews cited. The source mix counts distinct sites per answer rather than individual sources, and cannot separate Google Business Profiles from other Google pages.
How often a business’s own website was cited is not published for the earlier series: with 7 to 28 businesses a week, it ranged from 24% to 59% over six weeks and moved by up to 28 points from one week to the next on the mix of businesses alone. The average number of sources per AI Overview is not published for it either.