A descriptive pilot across 20 local business verticals. What appears in Google does not necessarily appear in an AI answer. This edition measures the differences across a fixed question set.
Descriptive pilot, not a business league table. Five prompts per vertical, ten repetitions. Automated evidence checks. Not peer reviewed. The full Index and recommendation rates are not released.
Research notes / Reproducible counts
Methodology and scope
This collection pairs five Google queries with five natural-language AI prompts in each of the 20 verticals. Each pair is repeated ten times on six configurations. Google AI Overview observations are derived from the organic requests, not purchased or counted as another independent platform.
Observation-led discovery is used instead of a preselected city business panel. The three business examples are selected contrasts with disclosed name/domain pairs. No population-wide mention share or overall business score is calculated.
Organic
One hit per usable response when the exact domain or explicitly grouped source-domain family appears at organic rank_group 1 to 10. Raw responses returned 17 to 20 organic results. Absolute mixed-SERP positions are not organic ranks.
Name examples
A specified, disclosed alias must appear as complete normalized tokens in answer prose, excluding URLs and displayed hostnames. Two separately implemented lexical matchers must agree. This checks calculation consistency, not semantic ground truth or all possible aliases.
Answer links
A source must be linked inline, referenced by a used Perplexity citation marker, or supplied as a Claude answer citation. Retrieval-only sources are excluded. Domains are deduplicated within each response.
Repetition
Jaccard overlap is intersection divided by union of returned place-ID sets for Maps (domain or title fallback where no ID exists) and domain sets for organic. Average within each prompt, then average eligible prompts equally. Empty/empty pairs and flagged responses are excluded. Order within the set is not measured. Maps set keys are SHA-256 hashes in the download, preserving set equality without publishing branch identities.
Audit trail
Every finding has numerator and denominator observation IDs. The downloadable fact projection preserves selected artifact hashes, status exclusions, exact prompt wording and returned model labels. Raw answers, private paths, contacts and general unreviewed entity rankings stay outside the release.
Collection and configurations
2026-09-09T16:11:24.911Z through 2026-09-10T16:48:26.720Z, UTC. Submitted and retrieved timestamps are in the fact export.
| Surface | Planned / usable | Access | Recorded returned model |
|---|
| Google Organic | 1000/1000 | DataForSEO search results | Not an LLM |
| Google Maps | 1000/990 | DataForSEO search results | Not an LLM |
| ChatGPT Search | 1000/994 | DataForSEO scraped search experience | unknown: 981 / gpt-5-6: 19 |
| Gemini | 1000/1000 | DataForSEO scraped search experience | 3.5 Flash-Lite: 1000 |
| Claude | 1000/1000 | model API | claude-haiku-4-5-20251001: 1000 |
| Perplexity | 1000/1000 | model API | sonar: 1000 |
Recorded request controls
Unconfigured settings are unknown, not assumed defaults. These are recorded request controls, not guarantees about the consumer interface.
ChatGPT Search: request settings
- location
- United States
- language
- en
- device
- Not explicitly configured
- os
- Not explicitly configured
- resultDepth
- Not explicitly configured
- maxOutputTokens
- Not explicitly configured
- temperature
- Not explicitly configured
- forceWebSearch
- true
- searchToolVersion
- Not explicitly configured
- maxSearchUses
- Not explicitly configured
- searchContextSize
- Not explicitly configured
- loadAsyncAiOverview
- Not explicitly configured
Claude: request settings
- location
- Lubbock, Texas, US
- language
- Not explicitly configured
- device
- Not explicitly configured
- os
- Not explicitly configured
- resultDepth
- Not explicitly configured
- maxOutputTokens
- 2048
- temperature
- Not explicitly configured
- forceWebSearch
- Not explicitly configured
- searchToolVersion
- web_search_20250305
- maxSearchUses
- 1
- searchContextSize
- Not explicitly configured
- loadAsyncAiOverview
- Not explicitly configured
Gemini: request settings
- location
- Lubbock,Texas,United States
- language
- en
- device
- Not explicitly configured
- os
- Not explicitly configured
- resultDepth
- Not explicitly configured
- maxOutputTokens
- Not explicitly configured
- temperature
- Not explicitly configured
- forceWebSearch
- Not explicitly configured
- searchToolVersion
- Not explicitly configured
- maxSearchUses
- Not explicitly configured
- searchContextSize
- Not explicitly configured
- loadAsyncAiOverview
- Not explicitly configured
Google Maps: request settings
- location
- Lubbock,Texas,United States
- language
- en
- device
- Not explicitly configured
- os
- Not explicitly configured
- resultDepth
- 20
- maxOutputTokens
- Not explicitly configured
- temperature
- Not explicitly configured
- forceWebSearch
- Not explicitly configured
- searchToolVersion
- Not explicitly configured
- maxSearchUses
- Not explicitly configured
- searchContextSize
- Not explicitly configured
- loadAsyncAiOverview
- Not explicitly configured
Google Organic: request settings
- location
- Lubbock,Texas,United States
- language
- en
- device
- mobile
- os
- android
- resultDepth
- 20
- maxOutputTokens
- Not explicitly configured
- temperature
- Not explicitly configured
- forceWebSearch
- Not explicitly configured
- searchToolVersion
- Not explicitly configured
- maxSearchUses
- Not explicitly configured
- searchContextSize
- Not explicitly configured
- loadAsyncAiOverview
- true
Perplexity: request settings
- location
- Not explicitly configured; Lubbock in prompt
- language
- en
- device
- Not explicitly configured
- os
- Not explicitly configured
- resultDepth
- Not explicitly configured
- maxOutputTokens
- 2048
- temperature
- Not explicitly configured
- forceWebSearch
- Not explicitly configured
- searchToolVersion
- Not explicitly configured
- maxSearchUses
- Not explicitly configured
- searchContextSize
- low
- loadAsyncAiOverview
- Not explicitly configured
What automated validation establishes
All 6,000 normalized records pass structural and citation-anchor checks. Separately implemented name matchers agree on 599 usable responses for the three disclosed examples. A second set-overlap calculation reconciles the Maps and organic results. A 360-record computational check set excludes the earlier 120-record builder sample.
This establishes reproducible computation, not independent semantic accuracy. No human reviewer, peer review, precision score or recall score is claimed. Ambiguous company/branch identities, unreviewed recommendations and full-inventory leaderboards stay outside this edition.
Limitations
- Five deliberately selected prompt pairs per vertical and ten repetitions. This pilot is not a census, a representative sample of consumers or a statewide index.
- Collection window: September 9 to 10, 2026. Repetitions may share caches or other dependencies. No independence-based confidence intervals or significance claims are made.
- Scraped ChatGPT/Gemini experiences and Claude/Perplexity APIs are different configurations. ChatGPT returned no model label for 981 of 1,000 observations. Returned labels are provider-reported, not independently verified.
- Gemini supplied detectable answer links in 205 of 1,000 responses. Missing source metadata is unknown attribution, not evidence of no citations.
- Six ChatGPT location-clarification answers and ten city-only Maps responses are retained but excluded from usable-result denominators. The Maps exclusions remove one whole wrongful-death query group.
- Business examples are purposively selected contrasts. Matching specified names is not a validated full-inventory mention rate, recommendation or evidence of local service eligibility.
- Validation is automated structural and computational checking, not human annotation, peer review or measured semantic precision/recall. Complete recommendation rates, business winners and branch-level league tables are withheld.
- This descriptive pilot does not satisfy the existing provisional 15-prompt full vertical-release gate. That gate remains unchanged and the full index remains withheld.
Producer and conflicts
Monsoon commissioned and produced this research and sells SEO and AI-search visibility services. Examples were selected to illustrate observed contrasts, not randomly or by business quality. Client relationships across the discovered inventory have not been independently reconciled. No relationship or payment field is used in these counts. This is not an endorsement, paid placement ranking or independent academic study.
Future editions
A comparable new edition every 3 to 6 months, with separately authorized collection and a change log for prompts, configurations and methods. This edition cannot identify an up-and-coming business or a gain over time without a comparable earlier observation.