Found Clearly sells AI-visibility audits and plans.
Questions and settings
We ran the study with Claude (Anthropic) in October 2026, outside the Claude app. Claude could search the web while it answered, and every question came with the same instructions. We wrote the protocol down before the first answer.
The questions were the 9 unbranded questions of our AI Visibility Benchmark, with the trade’s service and the city filled in:
- Who are the best {service} companies in {city}?
- I need {service} in {city}. Which local companies should I look at?
- Recommend three reliable {service} providers serving {city}.
- I need {service} in {city} this week. Give me three companies and one reason for each.
- Which {service} company in {city} would you call first, and why?
- Shortlist two {service} providers in {city} for a homeowner who cares about reliability over price.
- Which {service} companies in {city} offer same-day appointments?
- Who does {service} in {city} and is transparent about pricing?
- Which {service} providers near {city} are licensed and well reviewed?
The service was the trade’s own name, except for roofing (“roof replacement”), windows and doors (“window replacement”) and siding (“siding replacement”). The city was San Antonio, TX; Kansas City, MO; Indianapolis, IN; or Richmond, VA, written into the question and set as the buyer’s location for Claude’s searches.
That made 24 city and trade pairs and 2,748 answers in all, 2,376 of them in the main study, which put each question to Claude in more than one setup. Most of the numbers in the four articles come from the 864 answers of one setup: each question asked four times in each pair, 36 answers per pair. Where a number uses more of the main study, the article says so. The answers outside the main study tested other settings, locations and kinds of question; the 48 “near me” answers are among them.
How we counted companies
Claude read each answer and listed the organizations it names, with a type. Directories, manufacturers and media count as sources, not companies. Names were matched after removing case, LLC and Inc, punctuation and a trailing city name, and two spellings were joined only when they shared the company’s own website.
We checked that step by hand on a seeded sample of answers: 14 errors in 524 names (2.7%). An automatic check found 99 of 18,148 extracted names missing from the answer text (0.5%).
Overlap, links and list pages
Overlap is the Jaccard index of the companies two answers name; we report the median and the middle half of all pairs. A citation is one answer citing one page. Link groups come from a dictionary of known domains, and a page counts as a list page when it sits on a known list site or its address marks a list. Publishers of list pages were labeled by Claude and then checked by hand; the list-page figures use 6,471 citations because they also merge addresses that differ only by www, a trailing slash or tracking tags.
The website comparison
The two groups are described in What the websites of contractors Claude names have in common. The figures come from the homepage, read the same way in both groups. The rates are for each group as a whole; the last column compares the two groups inside each city and trade pair, so a pair with more sites of one group cannot tilt it. A profile counts only when its page came back in the search results.
Names: an amendment
The study was planned as counts only. After we collected the answers, we decided to publish the names Claude gave most often, with the checks below.
- Up to 10 names per city and trade, each named in at least 8 of the 36 answers behind the core-and-tail counts.
- Local contractors only, including local offices and franchises of national brands. Marketplaces, directories and lead-generation sites are sources, not companies.
- We checked each company on its own website or, where the site blocked our checker, on its BBB profile and other public listings: the name, city and trade match, and the business is open.
- We read every sentence that names a company for tone. A company Claude mentioned mainly as a caution is left out.
- The study’s other answers stay without names.
The lists are on who Claude names most often. They merge a few spellings the checks found to be one company, so their counts of different companies can sit slightly below the ones in the articles.
Corrections and removal
If a figure or a name is wrong, tell us through the contact page. We reply within 3 business days and fix a confirmed error within 7 days, and every change is dated in the log below.