How to get your business mentioned in Claude: 7 steps from new data

Anthropic has not published how Claude picks businesses, so most of what follows comes from one outside study. Searchable, a company that sells AI visibility tracking, ran about 3,000 prompts in Claude in July 2026 and logged roughly 95,000 search-result records. You can read it in their Claude visibility report (opens in a new tab).

Treat it as useful evidence from a vendor with something to sell, not as official fact. Its method is also partly inferred: Claude has not said which search provider it uses, so the study estimates it from overlap between what Claude read and what search engines return.

The main finding

Claude does two different things. For some questions it searches the web, reads pages and cites them. For others it answers from what it already learned in training. If you want to be mentioned, you need to know which kind of question you are trying to win.

The study found a big gap in how often Claude read pages. For "best CRM 2026" it read pages in about 9 in 10 runs. For "cheapest CRM" it was about 7 in 10, and for plain "best CRM" about 1 in 3. For comparisons ("Zendesk vs Intercom", about 1 in 7), "is it worth it" questions (about 1 in 11) and how-to questions (about 1 in 34), it mostly answered from memory.

What that means for your content

Words that signal freshness, price or a category ("2026", "cheapest", "best for small teams") push Claude to search, and a page can win those. Memory questions are mostly settled by what Claude already knows about you, which you can only change slowly, through the third-party coverage in steps 4 and 5.

Try this today

Write down the ten questions a buyer asks before choosing you. Mark each one as a search question or a memory question. Put your page-writing effort into the search questions first.

The study compared the pages Claude read with the top ten results from Brave and Google for the same queries on the same day. The overlap was 63% with Brave and 34% with Google. Claude has not confirmed which search provider it uses, so this is an inference, but it is a strong reason to check Brave.

A quick Brave check

1. Open Brave Search and search `site:yourdomain.com`. If few or no pages appear, Brave has not indexed you well.
2. Search your main service and your brand name and see whether you appear.
3. Submit your home page to Brave through its site submission form.
4. Check that your robots.txt and any firewall or bot-protection settings do not block crawlers. Some hosts block unfamiliar bots by default.
5. Keep your sitemap clean and your pages linked from each other, so crawlers can find everything.

Brave is a small search engine for most people, so this is a step many businesses have never taken. It costs very little.

Among pages Claude read, the share it went on to cite was similar across the top ten: 39.4% for ranks 1 to 3, 37.2% for ranks 4 to 6 and 36.8% for ranks 7 to 10. So being in the top ten matters more than being first.

How to find your page-two keywords

Open Google Search Console, go to Performance and filter by average position. Look for queries where you sit between 11 and 30 with real impressions. Those are your cheapest wins.

How to move them

• Add a direct answer in the first sentence or two under a question-style heading.
• Update old facts, prices and dates, and show an updated date.
• Add specifics that other pages lack: examples, numbers from your work, photos.
• Link to the page from related pages on your site.

Ordinary SEO still does the heavy lifting. Our guide to answer engine optimization tools covers the free reports that show where you stand.

The study compared how often each tool cited a page after reading it. On Claude, national press pages were cited 52.0% of the time (19.2% on ChatGPT), review sites 49.1% (44.8%), smaller comparison blogs 42.6% (28.8%) and directories 30.3% (29.3%). Reddit and forums were cited 8.3% of the time on Claude, against 12.7% on ChatGPT.

Another detail: no single site made up more than 9% of Claude's citations, while one site made up 22% of ChatGPT's. Claude spreads its sources out, so many small mentions can add up.

What to do about it

• Pitch a genuinely useful story or piece of data to publications your customers read.
• Get listed on the review and comparison sites in your category, and keep your profile there complete.
• Offer a real product, quote or data point to the bloggers who write comparison posts. Never pay for a fake review.
• Claim your listings on the directories that matter to your industry, and keep them accurate.

That gap is the most useful number in the report for small businesses. If Claude leans less on your own pages, a great blog post helps less than it would elsewhere, and what others say about you helps more.

A sensible split

Keep publishing on your own site: it is where the answers, prices and details live, and it feeds ordinary search. Then spend the next marginal hour on something third-party: a guest article, a comparison-site profile, a customer case study published on their site, a podcast appearance with a written summary.

Keep the facts consistent

Whatever others publish should match what your site says: name, services, prices and location. Conflicting facts make it harder for any tool to describe you with confidence.

An unbranded question is one that does not name you: "best accountant for an online shop in Austin", not "is Clarieo any good". When Claude names brands without being prompted to, they tend to stay in the answer. The study found that 86% of those self-named brands appeared in the final reply.

A simple tracking method

1. Write ten questions your customers would ask before they know you.
2. Run each one five times, each in a new chat, so earlier answers do not affect the next.
3. Count how many of the 50 answers mention you.
4. Write down who else was named and which sources Claude cited.
5. Repeat every three months and compare.

Answers vary from run to run, so look at the trend, not a single result. If you prefer a tool for this, see our roundup of answer engine optimization tools.

For local questions the study found Claude read about 0.1 web pages per query, and pulled from a listings service in 97.7% of prompts. In other words, your listing may matter more than your website.

What to check on every location

• Opening hours, including holidays.
• Primary and secondary categories.
• Description, services and photos.
• Address, phone number and service area.

One testing tip

When the study's prompts did not name a place, Claude defaulted to London. Test with an explicit town, city or postcode so the results reflect your area.

Our guide to AI search optimization for local businesses walks through the profile, consistency and review work in detail.