What Claude finds when people ask it for a local business
Claude API with web search, 3 runs, 2026-11-03 21:47–21:52 UTC. 3 runs for each of 10 questions. A snapshot, not a ranking.
On November 3, 2026, we put 10 buyer questions to Claude through the Claude API with web search turned on, and asked each one 3 times. Nine asked for a business in Columbus, Ohio. One asked for the best CRM for a small business. Claude searched the web in 10 of the 30 runs, and for 6 of the 10 questions it never searched at all. This page covers all 10 questions. The 4 it searched each have a page of their own.
- Model and settings: claude-sonnet-5-5 through the Claude API, effort low, web search tool
web_search_20260209limited to 1 search per answer, dynamic filtering on (the default), max_tokens 4000, no system prompt. - Run times (UTC, 2026-11-03): three rounds of 10 questions, at 21:47–21:48, 21:48–21:50 and 21:50–21:52.
Does Claude search the web every time I ask it something?
No. With web search turned on and the same settings every time, Claude searched in 10 of 30 runs. Anthropic's web search documentation says "Claude determines when to search based on the prompt." Here, it searched every time for web designers, roofers and landscapers, once in three for plumbers, and never for the other six questions.
When it did not search, 19 of the 20 answers said so, with lines like "I haven't searched" or "from general knowledge". For local SEO agencies and estate planning attorneys, Claude named no agency or attorney in any run and pointed to directories, review sites and bar associations instead. For HVAC, it named companies in 1 run of 3. For CPA firms, architects and CRMs, it named businesses from general knowledge.
What did Claude find for each of the 10 questions?
Claude searched for 4 of the 10 questions, and its searches returned 8 to 12 domains per question. For the other 6 it named sources from general knowledge. The table gives each question verbatim, how many runs searched, and which names appeared in the answer.
| Question (asked 3 times each) | Searched (runs of 3) | Domains in the search results | Named in the answer (runs of 3) |
|---|---|---|---|
| Who is a good web design company for a small business in Columbus, Ohio? | 3 | 8, the same in every run | designrush.com 3, splitreef.com 3, marcy.com 3 |
| Who is a good roofing company in Columbus, Ohio? | 3 | 8, the same in every run | local.yahoo.com 3, angi.com 3 |
| Who are good landscaping companies in Columbus, Ohio? | 3 | 12, only 4 in all 3 runs | yelp.com 2, homeguide.com 2, thumbtack.com 1, houzz.com 1 |
| Who is a good plumber in Columbus, Ohio? | 1 | 8, from the one run that searched | local.yahoo.com 1, angi.com 1 |
| Who is a good local SEO company for a small business in Columbus, Ohio? | 0 | none | Clutch 3, UpCity 3, Google Maps 2, BBB 2. No agency named. |
| Who is a reliable HVAC company in Columbus, Ohio? | 0 | none | BBB, Yelp, Angi, HomeAdvisor, Nextdoor 3 each |
| What is a good CPA firm for a small business in Columbus, Ohio? | 0 | none | Accountancy Board of Ohio 3; accounting firms, named from general knowledge |
| Who is a good estate planning attorney in Columbus, Ohio? | 0 | none | Columbus Bar Association, Ohio State Bar Association, Avvo, Martindale-Hubbell, Supreme Court of Ohio 3 each. No attorney named. |
| Who is a good residential architect in Columbus, Ohio? | 0 | none | AIA Columbus 3, Houzz 3; architecture firms, named from general knowledge |
| What is the best CRM for a small business? | 0 | none | HubSpot, Zoho, Pipedrive, Salesforce, Freshsales 3 each, named from general knowledge |
"In the search results" means the URLs Claude's search returned. "Named in the answer" means the name appears in Claude's answer text. For the 6 questions without a search, every name in the last column came from the model's general knowledge, not from search. The API returned no citation data in any run, so no column measures citations.
Which websites does Claude use to find local businesses?
When it searched, mostly lists of businesses. Across the 4 questions it searched, the results held 42 URLs on 25 domains, mostly directories, review platforms and "best in Columbus" roundups. 35 of the 42 titles say "best" or "top", and 38 name Columbus, Ohio or OH.
Business websites appeared in only two ways. Two web design agencies' homepages came up, splitreef.com and marcy.com. Four businesses also showed up through their own "best/top companies in Columbus" posts: loungelizard.com, columbuswebseo.com, exaltodesign.com and 614exteriors.com. No roofer's, landscaper's or plumber's service page was in the results.
What did Claude say when it did not search?
For six questions, Claude answered from general knowledge. Each answer said it had not searched, could not verify current reviews or ratings, or was going from general knowledge, and told the reader where or what to check: directories, review sites, bar associations, licensing boards, or current vendor pricing.
- Local SEO agencies: no agency named in any run. Claude pointed to Clutch and UpCity in all 3 runs, and to Google Maps and the BBB in 2.
- HVAC: the BBB, Yelp, Angi, HomeAdvisor and Nextdoor in all 3 runs. Run 1 also named five companies from general knowledge. We do not list them.
- CPA firms: Claude named Columbus-area accounting firms from general knowledge in its answers, and pointed to the Accountancy Board of Ohio in all 3 runs. We do not list the firms, because they came from the model's knowledge, not from search.
- Estate planning attorneys: no attorney named. Claude pointed to the Columbus Bar Association, the Ohio State Bar Association, Avvo, Martindale-Hubbell and the Supreme Court of Ohio in all 3 runs.
- Residential architects: Claude named firms from general knowledge, and in run 3 it said some were names "I'm less sure about". It pointed to AIA Columbus and Houzz in all 3 runs. We do not list the firms.
- Small-business CRM: the only question that did not name a city. Claude named HubSpot, Zoho, Pipedrive, Salesforce and Freshsales in all 3 runs, from general knowledge.
What did the pages in the search results have in common?
Of the 20 result pages that served us HTML, 17 had JSON-LD and 14 had a heading phrased as a question. We fetched all 42 result URLs on 2026-11-03 between 21:53 and 21:56 UTC, from our server, with a Chrome user agent, and read the HTML as served before JavaScript ran. 20 served us the page. 22 answered our server with HTTP 403, 429 or 202 and are not measured.
- Structured data: 17 of the 20 had JSON-LD. The most common types were ItemList (9), LocalBusiness (9), FAQPage (8) and Article (7).
- Question headings: 14 of the 20 had at least one H2 or H3 phrased as a question.
- Dates in titles: 19 of the 42 titles carry a year or month.
- robots.txt for Claude-SearchBot: allowed on 36 URLs, disallowed on 4, not measured on 2. The 4 are Yelp search pages. yelp.com's robots.txt disallows / for Claude-SearchBot, with exceptions only for /article pages. Its search pages were in the results in all 3 landscaping runs and in the plumbing run; this record does not show how they got there.
- robots.txt for ClaudeBot: disallowed on 8 URLs, on expertise.com, local.yahoo.com and yelp.com. local.yahoo.com's roundups were in the search results, and named in the answer, for both roofing and plumbing.
How do I get cited by AI?
No one can promise it, and this study did not measure citations: the API returned no citation data. What it shows is narrower. When Claude searched these local questions, its results were mostly lists, and it credited most of the businesses it named to those lists. We did not test what causes a business to be named.
What you can check is whether your own site can be read and quoted. Are answer crawlers allowed in your robots.txt? Do your pages answer a buyer's question in the first sentence under a question heading? Does your structured data say who you are and where? The free audit below checks the first two directly.
Can AI crawlers read my website?
Your robots.txt tells each crawler what it may fetch. Open yourdomain.com/robots.txt and look for a group naming Claude-SearchBot. If no group names it, the User-agent: * group applies to it. Either one with Disallow: / tells it not to fetch your pages.
Anthropic says Claude-SearchBot "navigates the web to improve search result quality" and that disabling it "may reduce your site's visibility and accuracy in user search results". A rule for ClaudeBot is a different choice: Anthropic says a ClaudeBot restriction signals that your site's future materials should be excluded from its model training data.
The free Corpus Plan audit reads your robots.txt for you. Its AI-crawler rules separate answer crawlers (OAI-SearchBot, PerplexityBot, Claude-SearchBot, Applebot) from training crawlers (GPTBot, ClaudeBot, Google-Extended, CCBot and others).
What about ChatGPT?
This study tested Claude only. Here are two common ChatGPT questions, answered as far as this data allows.
How to rank in ChatGPT?
This study did not test ChatGPT, and no one can promise a place in its answers. Two things on your own site are worth checking: whether your robots.txt allows OpenAI's answer crawler, OAI-SearchBot, and whether your pages answer the question plainly. The audit checks OAI-SearchBot, one of the four answer crawlers named in our methodology. For a measured look at ChatGPT itself, see What ChatGPT and Google cite: three measured studies.
What sources does ChatGPT use?
We did not measure ChatGPT on these questions, so this page can't answer that for local businesses. Our October 2026 studies recorded the sources ChatGPT listed for software and small-business buyer questions: What ChatGPT and Google cite. To check one question for your own business, see Does my website show up in ChatGPT?
Check your own site
The free Corpus Plan audit covers five areas: SEO, AI citability, speed, trust and content. It reads your AI-crawler rules, answer crawlers and training crawlers separately. It checks whether your question headings, plus any FAQPage entries, open with a first sentence of 40 words or fewer. The checks never ask a model.
Free preview with no account · free account for the full report · up to 20 pages per audit.
Run the free audit on your site · How the audit works
What this does and does not show
- One model, one day, and one city for 9 of the 10 questions: 10 questions, 3 runs each, between 21:47 and 21:52 UTC on 2026-11-03. Ask again and the results may change.
- For 6 of the 10 questions there are no search results to report, only what Claude said from general knowledge.
- "In the search results" means the URLs the search returned. Anthropic's documentation says that with dynamic filtering, Claude's code filters the search results before they reach its context window. In all 10 runs that searched, the code tried to print 6 results, hit an error, and then printed the first 3,000 characters of the results, which held only the first 2 results. The sites named in the answers were those first 2 results, plus two web design agencies whose names were in the DesignRush text. The record does not show Claude reading the other URLs we list.
- The API returned no citation data, so nothing here measures citations.
- This is the Claude API. We did not test the Claude.ai app.
- We measured a state, not a cause. Nothing here shows that a list page, a JSON-LD type or a robots.txt rule makes a site more likely to be found.
- 22 of the 42 result URLs answered our server with HTTP 403, 429 or 202 and are not measured.
- It is not a ranking of businesses or of websites.
The four questions Claude searched: web designers · roofers · landscapers · plumbers
Related: What ChatGPT and Google cite: three measured studies · Does my website show up in ChatGPT?