--- title: "ARO Index | Public AI Recommendation Research on Local AI Recommendations" url: https://aroindex.com/ datePublished: 2026-03-21 dateModified: 2026-11-15 author: Therese Grittner publisher: ARO Index --- ARO Index research # The Independent Index of AI Business Recommendations. Measuring which businesses AI systems recommend, across models, markets, and time. Published March 21, 2026 · Updated November 15, 2026 24,600+ Audits completed, cumulative 5,100+ Businesses measured, cumulative 9 Markets published in the Index 4 AI platforms evaluated Totals as of November 15, 2026. Audit and business totals are cumulative research counts, rounded down to the nearest hundred, and refresh hourly. Not every measured business appears in a published market. Latest research report: Vol 3, data current as of September 14, 2026. Featured research ## The same 687 businesses, measured twice. Charleston, July 7 and September 10, 2026. Same business cohort, same 43 categories, same buyer questions. How many cohort businesses did each AI model recommend? Charleston cohort businesses selected, by model Selected by any of the 3 models: **89** of 687 in July, **130** of 687 in September Count of distinct cohort businesses each model selected in a cold buyer query. Bars share one scale. ChatGPT July: 19 of 687 Sept: 51 of 687 Claude July: 21 of 687 Sept: 84 of 687 Gemini Excluded from this comparison. July Gemini responses averaged 86 to 93 characters, a known incident. Gemini is included in the report’s controlled test. Perplexity July: 77 of 687 Sept: 77 of 687 ### Observed Cohort businesses selected rose from 89 to 130 of 687. 64 were selected both times, 25 dropped out, and 66 were new. Perplexity held at 77. ### What changed between waves The July run had no live web retrieval on ChatGPT, Claude, or Gemini. The September run had live retrieval on all four. Perplexity retrieved in both, and it is the one model that did not move. ### What this does not show This is not a trend, and it does not show cause. Retrieval, time, and model updates changed together. In the report’s controlled same-day test, selection fell slightly: Charleston 151 to 147. MeasuredJuly 7 and Sept 10, 2026 PublishedSeptember 23, 2026 Sample687 Charleston businesses, 43 categories ModelsChatGPT, Claude, Perplexity (Gemini excluded here) MethodCold-query observed selection. Retrieval condition differs by wave, see report Section 3 [Source: ARO Index Vol 3, The September 2026 Model Shift, Section 1 →](https://aroindex.com/research/model-shift-sep2026) Research standards ## How the evidence is produced. ### How businesses are evaluated Each model gets cold, single-turn buyer questions for a city and category. A model counts as recommending a business when it places the business in its top 3 in at least 2 of 3 phrasings. ### Which AI systems are measured ChatGPT, Claude, Gemini, and Perplexity, each queried independently. Census batches have recorded model versions since late July 2026, and panel changes are dated. ### Why measurements are repeated Recommendations move. In Vol 2, 33 of the 89 Charleston businesses selected in early July were not selected three weeks later. One reading is a snapshot, not a finding. ### How results are compared Published wave comparisons use the same business cohort and the same categories. Added categories are reported separately, not counted as change. ### How method changes are documented Method versions, model panel changes, and data corrections are dated on the [methodology page](https://aroindex.com/methodology). Corrections stay published. ### How commercial work is separated ARO Index is published by TaG Makes LLC, which also offers paid services. No client, partner, or data customer can pay to change a public ranking. Every score runs through the same audit process. ### What the Index measures Observed AI recommendations under documented research conditions. ### What it does not measure Actual customer conversions, all possible AI responses, guaranteed future recommendations, or direct influence over model training. [Full methodology](https://aroindex.com/methodology)·[Census method](https://aroindex.com/methodology#census-method)·[Corrections](https://aroindex.com/methodology#data-correction-oct2026)·[Research notes](https://aroindex.com/insights) Explore the Index ## Put the research to work. ### Search a business Find a business’s ARO Score and rank in its published market. ### Explore a market See which businesses AI recommends in a city, by industry. [Aiken](https://aroindex.com/aiken)[Atlanta](https://aroindex.com/atlanta)[Charleston](https://aroindex.com/charleston)[Greenville](https://aroindex.com/greenville)[Manhattan](https://aroindex.com/manhattan)[Miami](https://aroindex.com/miami)[Nashville](https://aroindex.com/nashville)[Non Local](https://aroindex.com/non-local)[Sarasota](https://aroindex.com/sarasota) ### Compare recommendations Published comparisons across models and measurement waves. - [By model and wave](https://aroindex.com/research/model-shift-sep2026): July vs September, Charleston and Nashville, plus a controlled same-day test (Vol 3). - [Over three weeks](https://aroindex.com/research/charleston-2026-vol2): who held, dropped, and entered on the same 688 businesses (Vol 2). ### Your market is not in the Index yet? Tell us the city and industry. We will email you when it is published. Research applications and data access ## Built for Research. Available for Integration. Independent AI recommendation data for market intelligence, competitive research, and platform integrations. - **AI recommendation benchmarking.** ARO Scores for tracked businesses across ChatGPT, Claude, Gemini, and Perplexity. - **Longitudinal market intelligence.** Repeated census waves on fixed business cohorts, published by market and category. - **Competitive research.** Which businesses each model recommends instead, category by category. - **Agency and platform reporting.** Agency Console tracking and white-label reports. - **Data licensing and API integration.** Rankings API and data feeds for qualified partners, scoped by use. Access is reviewed by use case. Commercial access does not influence public research findings or rankings. [Agency and data access options →](https://aroindex.com/agency) [Ask about data licensing →](mailto:therese@aroindex.com?subject=Data%2FAPI%20Partner%20Inquiry) [Research briefings for chambers, SBDCs, and associations →](https://aroindex.com/book) Research library ## Published research. ### [ARO Index Vol 3: The September 2026 Model Shift](https://aroindex.com/research/model-shift-sep2026) Published September 23, 2026 Measured Data current as of Sept 14, 2026 Sample Charleston cohort 687, Nashville cohort 481 Models ChatGPT, Claude, Gemini, Perplexity Method No version change. Model panel change dated Sept 16, 2026 Suggested citationARO Index. (2026). ARO Index Vol 3: The September 2026 Model Shift. Published September 23, 2026. aroindex.com/research/model-shift-sep2026 ### [State of AI Recommendations: Charleston, Vol 2](https://aroindex.com/research/charleston-2026-vol2) Published July 26, 2026 Measured Data current as of July 24, 2026 Sample 688 Charleston cohort, 43 categories. Also 1,444 expanded Charleston and 485 Nashville Models ChatGPT, Claude, Gemini, Perplexity Method Same conditions as Vol 1 Suggested citationARO Index. (2026). State of AI Recommendations: Charleston, Vol 2. Published July 26, 2026. aroindex.com/research/charleston-2026-vol2 ### [The State of AI Recommendations: Charleston 2026](https://aroindex.com/research/charleston-2026) Published July 8, 2026 Measured Data current as of July 8, 2026 Sample 688 Charleston businesses Models ChatGPT, Claude, Gemini, Perplexity Method Score version v2 Suggested citationARO Index. (2026). The State of AI Recommendations: Charleston 2026. Published July 8, 2026. aroindex.com/research/charleston-2026 Researchers, journalists, and educators may cite published findings with attribution. [All research and notes](https://aroindex.com/insights) Questions ## Common questions about the ARO Index. ### What is the ARO Index? The ARO Index is public research that measures which local businesses ChatGPT, Claude, Gemini, and Perplexity recommend, by city and industry. It reports what the models return, not what businesses advertise. ### How does the ARO Index decide that an AI model recommends a business? Each model is queried independently with cold, single-turn buyer-intent prompts for a city and category. A model counts as recommending a business when it places the business in its top 3, confirmed by a 2 of 3 majority across three phrasings of the query. Each business shows how many of the 4 models recommend it, for example 3 of 4. ### Which AI models does the ARO Index test? ChatGPT, Claude, Gemini, and Perplexity. Each model is queried independently, so a business can be recommended by 1, 2, 3, or all 4 of them. ### How current are the numbers? Homepage totals show an as-of date and refresh hourly. Each market and category page shows when its rankings were last refreshed. Methodology changes and model panel changes are logged on the methodology page. ### Can a business pay to change its ranking? No. Businesses cannot pay to change their ARO Index ranking. Every score is re-measured through the same audit process. ### How can a business check whether AI recommends it? Run the free ARO audit at aroindex.com/audit. It checks whether each of the 4 models recommends the business and which businesses it recommends instead. ARO Index is researched and published by Therese Grittner, founder of TaG Makes LLC, Charleston, SC. [LinkedIn](https://www.linkedin.com/in/therese-grittner/)