How we measure what AI recommends
Customers increasingly ask AI assistants, not search engines, who to hire. AI+ Local makes those answers measurable: who each assistant names, how strongly they converge, the sources a recommendation rests on, and how all of it moves over time.
We take no payment for placement and apply no editorial weighting to the results; what the page reports is what the systems returned, recorded and tracked rather than curated.
The Process
Every figure on the site comes from the same weekly pipeline, run step by step below. The terms that appear across every area, category, and report are defined in the Glossary.
1.0Fix a query set per area and category | A fixed set of consumer-phrased prompts (“Who's the best plumber in South Tampa?” with variants for price and speed). |
1.1Put identical prompts to all five assistants | The same wording goes to ChatGPT, Gemini, Claude, Perplexity, and Grok every week, so answers compare cleanly over time. |
1.2Repeat each prompt and average the runs | A model's answer drifts between calls, so each prompt runs between ten and twenty times; the average reflects the aggregate, not a single noisy response. |
2.0Query in live, web-connected mode | Each assistant's live, search-connected mode, the exact configuration a customer reaches, so the output mirrors real-world usage rather than a sandbox. |
3.0Resolve every name to one real business | “Hiller,” “Happy Hiller,” and “Hiller Plumbing LLC” map to one entity, so counts and trends never double-count variants. |
3.1Discard hallucinated names | Names that resolve to no verifiable entity, the model's hallucinations, are dropped; only confirmed businesses enter the record. |
4.0Record every page an assistant cites | Each citation is kept as primary data, the evidence behind a recommendation rather than a footnote. |
4.1Classify each source by type | Directories, review sites, local news, community threads, a business's own site. |
5.0Append each run, never overwrite | Every entry is immutable; nothing is edited or overwritten after the fact. |
5.1Derive every figure from the history | How often a business is named, how many assistants agree, how settled the answer is, and which way it is trending. |
The Public Report
Everything above is published openly. The public report covers every area, category, and business we track across a market: the combined ranking, the level of agreement between assistants, the sources behind each answer, and the movement week to week. It is free to read, for any business, with no account.
The Pro Report
The public report refreshes weekly and describes a whole market. A Pro Report narrows the same measurement to a single business and refreshes daily, sometimes more often, so an owner sees a change the day it happens. See the Pro Report for everything it includes.
What We Don't Do
- We don't sell placement. Rankings come from the assistants' outputs alone.
- We don't cherry-pick runs. Every snapshot enters the record, including the ones where the assistants disagree or change their minds.
- We don't hide thin data. Areas where the assistants haven't named enough distinct businesses are marked as early rather than padded.