Visibility depends on the question.
We studied how selected AI model routes answered five homeowner scenarios for plumbing in Austin, Texas. Across three repetitions per question, 16 working routes produced 240 complete answers. The business reviewed was Rooter-Man of Austin.
In the native-search sample, the business appeared in 16 of 30 hidden-leak answers and one of 30 burst-pipe answers. That contrast is a useful starting point for investigation. It does not explain the cause or establish what a consumer will see in their next chat.
The business was mentioned.
The business was mentioned.
Each square is one reviewed answer. A mention can be a recommendation, a reference or a caution. These counts do not establish why the difference occurred.
Keep the search methods separate.
The study includes provider-native search and a separate group using model-directed external search. These groups contain different model panels and should not be treated as an experiment proving one search method performs better.
| Search method | Working routes | Client mentions |
|---|---|---|
| Native search | 10 | 43 of 150 answers |
| External search | 6 | 40 of 90 answers |
A mention may identify a provider, cite a useful page or describe a limitation. It is not automatically an endorsement. Both planned Google routes were unavailable and are excluded from these denominators.
Three findings worth acting on.
Customer intent changes the picture.
The questions covered urgent plumbing, recurring drain problems, water heaters, hidden leaks and sewer second opinions. A broad query about a company would miss the difference between these buying situations. A useful plan starts with the work the business actually wants to win.
Being named is only part of the evidence.
Some answers used the business as a service option. Others referenced practical advice or availability limitations. Some also confused Rooter-Man with Mr. Rooter. Reading the answer changes the meaning of the count.
Accuracy creates a concrete first task.
The report identified published business-detail variants, existing profile signals to review and specific editorial-list omissions. The next step is to confirm current business facts and assess the relevant pages. An uncited source is not proof of a missing listing, and an omission does not prove why a model selected another business.
What was included.
| Model family | Routes in the study | Search method | Complete answers |
|---|---|---|---|
| OpenAI | GPT-6.1 Sol, GPT-6 Sol, Luna, Astra | Native | 60 / 60 |
| Anthropic | Haiku 5.5, Sonnet 5.5, Opus 5.5, Fable 5.1 | Native | 60 / 60 |
| xAI | Grok 4.7 | Native | 15 / 15 |
| Meta | Muse Spark 1.3 | Native | 15 / 15 |
| Gemini 3.8 Flash, Gemini 4 Argon | Native | Unavailable | |
| Qwen | Qwen3.8-Max, Qwen3.7-Plus | External | 30 / 30 |
| Moonshot AI | Kimi K3, Kimi K2.6 | External | 30 / 30 |
| DeepSeek | V4 Pro, V4.1 Flash | External | 30 / 30 |
Dated study coverage, not a list of every current model. Native and external cohorts remain separate. External routes used model-directed Exa search. Eighteen routes were planned; sixteen completed. Two Google routes produced no usable answers after renewed requests.
These are configured API observations. Consumer products may use different systems, context and settings. The study does not measure market share, causal ranking effects or the probability of a future recommendation.
From a baseline to a working plan.
Confirm current locations, phone numbers and dispatch availability. Audit the relevant website answers and existing profiles. Prioritize justified improvements, publish them and repeat the same study design to look for change. Business inquiries and confirmed jobs remain separate measures.
That sequence connects research to delivery: the questions guide the pages, the pages supply approved content, and repeated observation helps the team decide where to work next.
This is a baseline study. No before-and-after result or improvement attributable to Found First is claimed here.
A public summary with a private evidence trail.
This article publishes selected aggregates and interpretation from the October 21 study. Found First retains the detailed conversations, source evidence and review records privately. The public summary excludes exact prompt wording, full source inventories, collection settings and raw exports.
Business reference: Rooter-Man of Austin. Research and analysis: Found First. The homeowner scenarios were drafted and reviewed for this study; they are not measured query-volume data.