Claude Recommends With Its Doubts Written Into the Answer
To get your business recommended by Claude, give it pages it can quote without a warning attached. That means a pricing page with current prices and a recent date, your service areas and phone number written out as plain text, a Yelp review count big enough to stand behind your rating, crawlers that aren't blocked, and no slogan doing the job a fact should do. That list comes from reading what Claude wrote. On November 11, 2026, we pulled Claude's answers to four customer questions on a Miami mobile car wash, taken from our local business case study, and every answer carried a caveat in Claude's own words. One page was flagged as old. One rival's claim to be the best was labeled as its own marketing. One rating was called low and based on only a few reviews. Each caveat points at a record somebody could have fixed.
Claude is still the small one by traffic, and it's growing fast. In April 2026, Statcounter put Claude's share of the visits AI chatbots send to websites worldwide at 2.91 percent for March, up from 0.30 percent in April 2025, measured across more than 3 billion page views a month. That counts clicks, not people asking, so read it as a trend. This guide belongs to our series on getting recommended, which opened with getting your business recommended by AI on any assistant and then took the assistants one by one. The stop before this one, getting recommended by Perplexity, sorted the sources listed under each answer. Claude needs a different reading, because the useful part isn't under the answer. It's in the sentences where Claude says how far it trusts what it found.
Four Answers From One Morning, Read for Their Warnings
The car wash's account asks Claude 23 customer questions, most of them daily, and saves each reply along with the pages it cited. We took four replies from November 11, a plain search, one with a neighborhood, one about price, and one asking who's best, each shortened in the table, and read them for two things, whether the car wash was named and what Claude said about its own confidence.
| Question | Car wash named | Pages cited | Whose pages | The caveat Claude wrote |
|---|---|---|---|---|
| mobile car wash miami | Yes, second of seven | 9 | All nine were company websites | Prices come from each company's own site, and one rival's page is older |
| mobile car wash aventura | Yes, second of six | 7 | Five company pages and two Yelp pages | The car wash's pricing page is about 438 days old, and one rival's Yelp rating is low on few reviews |
| car wash cost in Miami | No, quoted as one provider | 6 | Its pricing page first, then other washes, a guide, and Groupon | The ranges are a guide, not fixed rates |
| best mobile car wash Miami | No | 4 | Two Yelp search pages, a listings directory, and one rival site | No single best exists, and a rival's best claim is its own marketing |
The car wash was named on the two questions that asked for a company to hire, and left out of the two that asked for a number or a ranking. Claude didn't cite google.com in any of the four, and across the whole 30 days to November 11 it never cited google.com for this account, though it cited Yelp, the Better Business Bureau, Groupon, and Instagram. The caveats are the new part. In the answers we read from the other assistants to the same four questions that morning, none mentioned how old a page was.
One limit before the detail. These are our tracker's scheduled questions, and a person's own Claude app, with their location and whatever connectors they've added, can answer differently. Four answers show what Claude read and said on one morning. The fixes below are ordered by those sentences, not proven by them.
What Sits Behind a Claude Search, as Far as Outsiders Can Tell
Anthropic doesn't publish how Claude picks a local business, so this section is a map of what's documented and what's only reported. Claude's web search runs on someone else's index. In July 2026, the agency Xponent21 read Anthropic's list of data subprocessors and found Brave Search still listed for web search, where it has sat since March 2025, joined in May 2026 by a second search vendor called turbopuffer whose role nobody outside has pinned down. In November 2026, Search Engine Journal covered a New York Times lawsuit against Brave whose complaint names Anthropic among the companies Brave supplies with search results, and the writer added that he couldn't confirm Claude still uses Brave. So Brave is the best-supported answer, and it isn't a certain one.
That matters because of how Brave finds pages. Brave's own documentation says its crawler won't crawl a page Googlebot can't, announces no user agent of its own, and gets pointed at pages partly by people who browse with Brave and opt in to sharing what they visit. That document describes no owner console to sign in to. Our read is that a small site reaches Brave the way it reaches Google, by being crawlable and by having real visitors, with no extra form to fill in. Anthropic runs its own crawlers on top. It names three, ClaudeBot for training, Claude-User for fetching a page when a person asks, and Claude-SearchBot for search, each with its own robots.txt line, and Anthropic warns that blocking Claude-SearchBot may reduce how visible your site is in search answers. Our walkthrough for checking if AI recommends your business includes a robots.txt test that covers all three.
Listings are the murkiest layer. In May 2026, Search Engine World relayed a trade report that Claude uses Google Maps and the Google Places API for some local questions, hotels in particular. That's secondhand, and it isn't what our captures show for a service trade, where Claude read Yelp and company websites and never cited Google. The one first-party local source we could find is opt-in. Since September 2026, Claude's connector directory has carried a Yelp connector that returns ratings, reviews, and business details and can request quotes from service pros, for a person who chooses to switch it on. Our earlier post asking whether directory listings affect AI visibility carries a July 2026 measurement in which Claude read almost no ordinary web pages on local questions, close to the opposite of our four captures, so treat this layer as unsettled and cover both the listing and the site.
Caveat One, Claude Read the Date on Your Pricing Page
On the Aventura question, Claude listed the car wash second. It said the company is locally owned, said it lists Aventura among its service areas, printed its phone number, and quoted a premium wash from 45 dollars along with its subscription plans. Then it added that the car wash's pricing page is about 438 days old and told the reader to confirm current rates. On the plain Miami question it did the same to a rival, noting that the page was older and should be verified. The prices on the car wash's page may well be current. Claude couldn't tell, so it said so, and a recommendation that ends in a request to double-check is weaker than the rival's entry in the same answer that carried no warning.
Our read is that Claude's search results arrive with a date for each page, and Claude passes its doubt along when that date is far back. We can't see which date it used, whether a published date on the page, a modified date, or the last time the index fetched it. So cover all three. Put a visible line on your pricing page saying when the prices were last reviewed, keep the modified date honest by editing the page when you review it, and make sure the page sits in your sitemap with a current last-modified entry. Review prices on a schedule even when nothing changes, and move the date only when you've really checked them. A price list that was reviewed this quarter gives Claude nothing to warn about.
Caveat Two, a Slogan Got Labeled and a Review Count Got Quoted
Claude treats a company's description of itself as a claim and says whose claim it is. On the best question it listed a rival whose website calls it Miami's best mobile car wash, then wrote that this was the company's own marketing claim and that its listing showed a 4.8 rating from one review. On the Aventura question it wrote that one company describes itself a certain way, that another reports 4.9 stars from more than 300 Google reviews, and that a third's Yelp rating is low and based on only a few reviews. The verbs do the sorting. Describes itself and reports mark what a business said about itself. A star rating with a count beside it, read off Yelp, is stated flat.

Yelp was where those counts came from. The most-reviewed company in the best answer earned its line with a 4.5 rating across 191 Yelp reviews, and Claude's closing advice to the reader was to look for a higher review count over a perfect score. A wider test points the same way, with limits. In September 2026, 8ight.AI asked several assistants to name the best business for 39 trades in 49 places, and PPC Land's write-up says Claude cited Yelp in 202 of its 218 answers and named about 11 percent of the businesses tested. That ran through Claude's developer interface, not the app, the Claude sample is small, and 8ight sells AI visibility services, so take the direction and leave the decimals. For a local owner the direction is enough. How reviews shape AI recommendations covers earning them without a script, and for Claude the place to earn them includes Yelp, where a rating with twelve reviews behind it reads very differently from one with two hundred.
Caveat Three, Claude Wouldn't Name a Winner
Asked to pick the best of Miami's mobile car washes, Claude opened by saying there's no single consensus best, that the results were mostly self-promotional or review-platform listings, and that it couldn't name one definitively. Then it listed eight companies with their ratings and review counts and told the reader to get two or three quotes. The car wash wasn't among the eight, and none of its pages were among the four Claude cited, which were two Yelp search pages for mobile car washes in Miami, a listings directory, and one rival's website. ChatGPT, given the identical question that morning, opened with a single pick.
Two things follow. First, on a best question Claude built its list from the page a directory returns for the search, so a business has to appear in Yelp's results for its trade and city, with a count that holds up, before its own website matters at all. Second, there's no shortcut around that. In February 2026, Anthropic said Claude will stay ad-free, with no sponsored links beside its answers and no advertiser influence on what it says, and that it would be open about it if that ever changed. A place on a hedged list of eight is the most this question gives anyone. Being one of the eight, with the biggest honest review count on the page, is the first job.
Eighty-Four Answers Borrowed the Site and Left the Name Out
Asked what a car wash costs in Miami, Claude cited the car wash's pricing page first and lifted its four prices, basic, interior, premium, and full detail, into the mobile wash part of the answer. It introduced them with the words one provider. A second mobile company got the same treatment, while a tunnel chain and two hand-wash shops were named beside their prices. One answer can't say why. What it shows is that a page can do the work of answering and still leave the business out of the sentence.
That gap shows in the month's totals. Across the 30 days to November 11, 2026, Claude answered the car wash's tracked questions 508 times and named the business in 344 of them, or 68 percent. Of the 508, 342 named the car wash and cited its site, 84 cited the site and left the business unnamed, 2 named it with no citation, and 80 did neither. Counted separately, 86 answers named a rival and left the car wash out.

Why a footnote isn't a recommendation is the subject of citations vs. mentions in AI search, and on Claude the fix is small. Open your pricing page with one sentence naming your business, its trade, and its city, before the first number, so the price Claude quotes has a name attached to it.
The Fix List, Sorted by the Warning Each One Removes
Each fix names the sentence of Claude's that it answers.
- Date your pricing page and keep the date true. Removes the page-is-old warning. A visible last-reviewed line, an honest modified date, and a current sitemap entry, moved only when you've checked the prices.
- Write the service area, phone number, and starting prices as text. Earns the quote. Claude repeated the car wash's line about homes, offices, and two counties nearly word for word, and it can only repeat what's written out.
- Trade the slogan for a fact Claude can attribute. Removes the marketing-claim label. Best in Miami got flagged. A rival's statement that it has 4.9 stars from more than 300 Google reviews got repeated, with the word reports in front.
- Build the Yelp review count. Removes the few-reviews warning and puts you on the page Claude read for the best question. Claim the listing, fix the category and service area, and ask recent customers, since Claude printed the count next to every rating.
- Let Claude's crawlers and Googlebot in. Protects everything above. Allow Claude-SearchBot and Claude-User in robots.txt, and remember that a page Googlebot can't reach is one Brave says it won't crawl either.
- Name yourself beside your prices. Turns one provider into your name. Business, trade, and city in the first sentence of your pricing page.
Five of the six are edits to your own site, and the sixth is a Yelp listing. If you'd rather hand the list to someone, the team at Lighthouse Local for Business takes on the on-page work, content, and local citations, and nothing goes live until you approve it.
Collect the Caveats Claude Writes About Your Business
Set aside a quarter of an hour, and read for the warnings as well as for your name.
- Open Claude in a fresh chat with web search on. Use an incognito chat if you've talked to it about your business before, so memory doesn't tilt the answer.
- Ask for your trade and city, without your business name. Note where you're listed and whether any line about you ends in a warning.
- Ask again with a neighborhood you serve. Check whether Claude quotes your page for that area or falls back to your home page.
- Ask what your service costs where you work. See whether your prices show up, and whether your name is next to them.
- Ask who's best. Write down every site Claude cites, because that's the list you need to be on, and note the review counts it prints.
Copy any caveat about you word for word and date it. Then make the fix and ask again in a month, since our read is that a search index won't pick up an edit overnight. For a standing record, Lighthouse Local's AI visibility tracker asks Claude your customer questions on a schedule and keeps each answer with the pages it cited, so you can see the week a warning disappears or a rival's name turns up in your place. It tracks and reports. It can't make Claude name anyone.
Claude Shows Its Doubts, So Give It Fewer
Claude named the Miami car wash for the two questions that asked who to hire, quoted its prices without its name on the third, and left it off a best list built from Yelp. In one of the answers that did recommend the car wash, it also told the reader the prices might be out of date. None of that was hidden. Claude wrote each reservation into the answer, which is as close as any assistant comes to handing you its objections in plain sentences. For the site half of the list, run a free AI audit and the crawl-access and sitemap problems come back as a ranked fix list, and you don't need to sign up. The Yelp half starts with the count beside your rating. Read the warnings Claude attaches to you, then remove the reason for each one.
Frequently asked questions
Does Claude recommend local businesses?
Yes. With web search on, Claude answers a local question with a list of named businesses, quoting prices, service areas, and phone numbers from the pages it cites. In our November 2026 captures it listed a Miami mobile car wash second on a plain trade-and-city search and again on a neighborhood question, and attached a caveat to each list.
Which search engine does Claude use?
Anthropic's subprocessor list has named Brave Search for web search since March 2025, and a July 2026 reading of that list found a second vendor, turbopuffer, added in May 2026. A November 2026 report couldn't confirm Brave is still what Claude uses. Brave says its crawler only fetches pages Googlebot can reach, so a site open to Google is open to Brave.
Why did Claude say my page is old?
Our read is that Claude sees a date for each page in its search results and warns the reader when that date is far back. In our capture it called a pricing page about 438 days old and said to confirm current rates. A visible last-reviewed line, an honest modified date, and a current sitemap entry are the fixes.
Does Claude use Google reviews or Yelp?
In our captures, Claude took ratings and review counts from Yelp and from what companies reported on their own sites, and it never cited google.com for this account in 30 days. A September 2026 test through Claude's developer interface found Yelp cited in 202 of 218 answers. One trade report says Claude uses Google Places for some local queries, hotels especially, which we couldn't confirm.
Can I pay to be recommended by Claude?
No. In February 2026, Anthropic said Claude will stay ad-free, with no sponsored links and no advertiser influence on its answers, and that it would be open about any change to that. The pages Claude can read and quote are what decide a local answer.
How do I check whether Claude recommends my business?
Open a fresh Claude chat with web search on and ask four things, your trade and city, again with a neighborhood, what the service costs, and who's best. Note whether you're named, which sites it cites, and any warning it attaches to you. Repeat monthly with the same wording.



