If you want to know whether Claude mentions your business, do not start with a rank tracker or a long audit. Start by asking Claude the same buying questions your customers ask, then log what it says. In 15 minutes you can usually tell whether your business is named at all, whether it appears for the right searches, and whether competitors are getting the mentions instead.
The key is to test properly. One vague prompt tells you very little. A small set of realistic prompts, in the right language and market, tells you much more. If you have been wondering how to check if Claude mentions your business, this is the practical way to do it.
Start with the quickest reliable check - how a busy owner can test in 15 minutes whether Claude names the business for real buying questions
Set aside 15 minutes. Open a spreadsheet or a simple document. You need three things before you begin:
- Your business name, written exactly as customers know it
- Five to ten competitor names
- A short list of real customer search intents
Do not test with vanity prompts such as “tell me about my company”. That only checks whether Claude can repeat a name it has just been given. You want buying questions.
A fast test looks like this:
- Pick one market
- Pick one language
- Write 8 to 12 prompts
- Ask Claude each prompt in a fresh chat where possible
- Record the result
If you serve more than one country, do not mix them in the same first pass. “Best payroll software in Ireland” and “best payroll software in Germany” are different tests. So are “dentist in Leeds” and “dentist in Vilnius”.
A useful first set is:
- 2 branded prompts
- 3 category prompts
- 2 local prompts
- 2 comparison prompts
That is enough to see a pattern without overcomplicating it.
For example, if you run a German bookkeeping firm, your test should not be “Who are the best accountants?” with no location and no language signal. It should be closer to what a buyer would ask, such as a German-language prompt about bookkeeping support for small companies in Hamburg, or a comparison between local providers and software-led options.
You are looking for four simple outcomes:
- Claude names you clearly
- Claude mentions you weakly, perhaps in a long list
- Claude does not mention you
- Claude gets confused about who you are
That last one matters. A wrong description, wrong location, or wrong category is still a visibility problem.
If you want a one-line answer to the question, “How do I check whether Claude mentions my business?”, it is this: test real commercial prompts, in the right language and market, and log exactly what Claude says.
Use prompts that match how customers actually ask - which prompt types to use for branded, category, local and comparison searches so the test reflects demand
Prompt choice is where most checks go wrong. Owners either ask prompts that are too broad, or prompts so narrow that only their own brand could appear.
Use four prompt types.
Branded prompts
These test whether Claude recognises your business as an entity and describes it correctly.
Examples:
- “What does [business name] do?”
- “Is [business name] a good option for [service]?”
- “Tell me about [business name] in [country]”
Branded prompts are not enough on their own. They mostly test recognition, not recommendation. But they are useful for spotting basic entity issues, such as Claude confusing you with another company or not understanding your service.
Category prompts
These are the most important. They test whether Claude names you when a buyer asks for providers in your category.
Examples:
- “What are good options for outsourced HR support for small businesses in Ireland?”
- “Which companies offer industrial cleaning services in Copenhagen?”
- “Best software for appointment booking for salons in Poland”
Keep them realistic. Buyers usually include the need, sometimes the business type, and often the country or city. They do not always say “best”. They may ask for “good”, “reliable”, “affordable”, or “suitable for a small company”.
Local prompts
These matter for service businesses, clinics, trades, agencies, legal firms, and any business where geography affects the decision.
Examples:
- “Who are reputable family lawyers in Cork?”
- “Which estate agents in Tallinn are good for first-time sellers?”
- “Recommended physiotherapy clinics in Malmö”
Local prompts should match your actual trading area. If you only serve one city, a country-level prompt may not be the right test. If you serve nationally, a city-only test may understate your visibility.
Comparison prompts
These show whether Claude groups you with the competitors buyers already know.
Examples:
- “[Your brand] vs [competitor] for payroll services”
- “Alternatives to [competitor] in Germany”
- “Compare [your category] providers for small manufacturers in the UK”
Comparison prompts are useful because buyers ask them late in the journey. If you appear here, it often means Claude sees you as a legitimate option, not just a name on the web.
How many prompts to use
For a basic manual check, use 8 to 12 prompts per market and language. That is enough to spot whether absence is general or limited to one prompt type.
A simple mix:
- 2 branded
- 4 category
- 3 local
- 3 comparison
If your business is not local, replace local prompts with more category and comparison prompts.
Prompt writing rules
Use these rules every time:
- Write prompts in the customer’s language
- Include the market if it matters
- Keep the prompt natural
- Do not stuff your own brand into non-branded prompts
- Do not ask leading questions such as “Why is [my company] the best choice?”
If you ask, “What are the best translation agencies in Riga?”, that is a fair test. If you ask, “Would you recommend my agency, [name], as one of the best translation agencies in Riga?”, you have biased the result.
Record results in a simple scorecard - what to log for each Claude answer including mention, position, wording, source patterns and competitor presence
Do not trust memory. Two days later you will only remember that “it mentioned us once”. That is not enough to improve anything.
Use a simple scorecard. One row per prompt. Log these columns:
- Date
- Market
- Language
- Prompt type
- Prompt text
- Mentioned, yes or no
- Position
- Exact wording used
- Description accuracy
- Competitors mentioned
- Source patterns
- Notes
Here is what each field means.
Mentioned, yes or no
This is binary. Either Claude named your business or it did not.
If it referred to you vaguely, such as “a local provider” without naming you, count that as no.
Position
If Claude gives a list, note where you appear.
Use a simple system:
- 1 for first mention
- 2 for second
- 3 for third
- 4+ for lower down
- N/A if there is no ranked list
Position matters because many buyers stop at the first few names.
Exact wording used
Copy the phrase Claude used. For example:
- “A well-known bookkeeping firm in Dublin”
- “A local clinic with multilingual staff”
- “An e-commerce platform”
- “A software provider for logistics teams”
This tells you whether Claude understands your category properly. If you are a managed service provider and Claude calls you an app, that is a clue.
Description accuracy
Mark it as:
- Accurate
- Partly accurate
- Inaccurate
Then note the problem. Wrong city. Wrong service. Wrong customer type. Wrong market.
Competitors mentioned
List every competitor named in the same answer. This is often more useful than your own mention status.
If the same three competitors appear repeatedly across prompts, they probably have stronger entity signals, more authority, better local proof, or broader topical coverage than you.
Source patterns
Claude does not always show sources in the same way as search engines or answer engines that cite heavily. But you can still inspect patterns in what it seems to rely on.
Look for signs such as:
- It repeats wording close to your homepage
- It refers to review themes
- It uses directory-style descriptions
- It emphasises editorial or comparison language
- It reflects local business profile information
- It echoes third-party articles or lists
You are not trying to prove a direct source with certainty. You are trying to spot what type of web evidence may be shaping the answer.
Notes
Use this for anything unusual:
- Claude asked follow-up questions before answering
- Claude refused to recommend specific providers
- The answer changed significantly on a second run
- The market was interpreted wrongly
- The language drifted into English
A basic scorecard is enough for manual checks. If you need repeated monitoring across Claude, ChatGPT, Gemini, Perplexity and Grok, a dedicated platform can save time. Seonis is built for that kind of work, especially for businesses with non-English websites, and explains results in the owner’s own language through Seonis: SEO and AI visibility in your website’s language.
Test in the right language and market - how to check mentions properly for businesses with non-English sites, multiple countries or two-language shops
This is where many EU businesses get misread.
If your site is in Polish, Lithuanian, German, Swedish or another local language, test in that language first. An English prompt may produce a different answer, because the model may rely on a different body of evidence or infer that the buyer is international.
For non-English sites
If your website, reviews, service pages and business listings are mainly in one language, your first test must use that language.
For example:
- A Latvian legal practice should be tested in Latvian for local consumer demand
- A German B2B manufacturer should be tested in German for domestic buyers
- A Finnish clinic should be tested in Finnish, and possibly Swedish if that reflects real patient demand
Do not assume English is a neutral benchmark. For many local businesses it is the wrong benchmark.
For multiple countries
Run separate tests for each market. Keep the prompts market-specific.
Examples:
- One set for the UK
- One set for Ireland
- One set for Germany
- One set for Poland
If you trade in the EU generally but have country-specific pages, local references and local customers, Claude may mention you in one market and ignore you in another. That is normal. Visibility is rarely uniform.
Where rules differ, say so in your own site content too. For example, a financial service may need separate pages for UK and EU treatment because the practical details differ. Claude is more likely to mention a business when the business itself is clear about scope.
For two-language shops
If your business genuinely serves two languages, test both.
This is common in:
- Ireland, English and Irish in some contexts, though English usually drives business demand
- Finland, Finnish and Swedish
- Belgium, depending on region
- The Baltics, where local language and Russian may both matter in some sectors
- Tourist areas across the EU, where local language and English both bring customers
Do not merge the scorecards. Keep one per language. You may find that Claude recognises your business strongly in one language and barely at all in the other.
Check the market framing inside the prompt
A good prompt says enough to anchor the market without overloading it.
Examples:
- “Good payroll providers for small businesses in Ireland”
- “Recommended tax advisers for freelancers in Berlin”
- “Best wedding venues in South Wales”
A weak prompt leaves too much room for drift:
- “Best payroll providers”
- “Tax advisers near me”
- “Good wedding venues”
Claude can only answer the market you imply. If your test prompt is vague, your result will be vague too.
Work out why Claude does or does not mention you - the common reasons behind visibility or absence, from weak entity signals to thin local proof and poor topical coverage
Once you have a scorecard, patterns start to show. Most visibility problems fall into a few common buckets.
Weak entity signals
Claude may not be confident that your business is a distinct, well-defined entity.
Common causes:
- Inconsistent business name across site and listings
- No clear About page
- No legal or company details where appropriate
- Different phone numbers or addresses in different places
- Confusing brand architecture
- Thin mentions outside your own website
If your business is called one thing on the homepage, another in the footer, and a third in directories, you make the model’s job harder.
Thin local proof
This affects local service businesses most.
Common gaps:
- Few reviews on credible platforms
- No location-specific service pages
- Weak local backlinks
- Sparse mentions in local media, chambers, associations or directories
- No evidence of local projects, clients or case studies
If competitors are named for local prompts and you are not, local proof is often the reason.
Poor topical coverage
Claude is more likely to mention businesses that are clearly associated with the buyer’s topic.
If you offer six services but only one thin page covers each, you may not be strongly associated with any of them. A competitor with better service pages, examples and supporting articles may be easier to recommend.
This is especially common with non-English websites that have relied on translated filler content. Buyers and models both notice when content is generic.
Weak comparison context
If you never appear in lists, round-ups, alternatives pages or discussions that place you beside competitors, Claude may not learn that you belong in the set.
This does not mean you need spammy list placements. It means your business should appear in the normal web context of your category.
Market mismatch
Sometimes Claude does not mention you because your site does not clearly say where you serve.
Examples:
- A Danish agency with English-only pages but no clear Denmark signals
- A Polish software firm serving Germany but with no German-market pages
- A UK company using generic “Europe-wide” messaging without country detail
If the model cannot tell where you are relevant, it may choose brands with stronger market evidence.
Technical and content neglect
Basic site issues still matter indirectly:
- Important pages not indexed in Google
- Weak internal linking
- Old service pages with little substance
- No schema where appropriate
- Thin title tags and headings
- Slow or broken mobile pages
Claude is not Google, but models are influenced by the web ecosystem that search also reflects. If your site is hard to understand, your brand is harder to mention.
Improve your chances of being named next time - the practical fixes to make on your site and across the web, plus when a tracking tool such as Seonis can save time
You cannot force Claude to mention you. You can make your business easier to recognise, trust and retrieve.
Start with the fixes that usually move the needle.
Make your business identity explicit
Check these first:
- One consistent business name everywhere
- Clear About page
- Clear service pages
- Clear country and city coverage
- Contact details that match across site and listings
- Legal details where relevant for your market
If you operate in regulated sectors, be specific about registrations, memberships and scope. In the UK and Ireland that often means naming the relevant body where appropriate. In EU markets, the exact proof differs by sector and country, so use the real local standard.
Build proper service and location pages
If you want to be named for a service in a place, create a page that proves it.
Good pages include:
- What you do
- Who it is for
- Where you do it
- What makes the service suitable
- Real examples
- Local proof
- Clear next steps
Do not publish ten thin city pages with the place name swapped out. That rarely helps.
Improve topical depth
Add supporting content around the questions buyers ask before they choose.
Examples:
- Process guides
- Cost factors
- Comparison pages
- Common mistakes
- Local regulation explainers
- Case studies
- Industry-specific pages
If your site is in a local language, write natively in that language. Do not translate English articles and hope for the best. That is one reason Seonis focuses on native-language research and publication rather than translation-driven output.
Strengthen internal linking
A surprising number of businesses have useful pages that are barely linked from anywhere important. Fix that.
Link related service, location and advice pages together in ways that make sense for users. If you need a practical guide, this article on internal linking for local language SEO pages is worth a read.
Earn the right third-party mentions
Focus on sources that make sense in your market:
- Local business directories
- Professional associations
- Chambers of commerce
- Trade publications
- Local news
- Partner pages
- Relevant suppliers and memberships
Do not chase random links in the wrong language or country. Relevance matters more than volume.
Add comparison and proof content
If buyers compare providers, help them do it.
Useful formats:
- “Who we are best for”
- “When to choose a local provider”
- “Comparison of service models”
- “Alternatives for small businesses”
- “Case studies by sector”
Be factual. Overclaiming tends to backfire.
Re-test after changes
Do not expect instant movement. Make a batch of improvements, wait, then run the same prompt set again.
Use the same:
- Language
- Market
- Prompt types
- Competitor list
- Scorecard fields
That is how you see whether mention rate, position and wording improve.
When manual checking stops being practical
Manual checks are fine for one business, one market, and a small prompt set. They become tedious when you need to track:
- Multiple languages
- Multiple countries
- Several competitors
- Several AI systems
- Repeated checks over time
That is where a tracking platform earns its keep. Seonis monitors whether ChatGPT, Gemini, Perplexity, Claude and Grok name your business when buyers ask, and it is designed for SMEs with websites in their own language rather than English-first brands. If you run an agency, the Partner programme for agencies may be the more relevant route.
The main thing is not the tool. It is the method. Use real prompts. Use the right language. Log the answers. Fix what the pattern shows. Then test again.
That is the honest answer to how to check whether Claude mentions your business. It is not mysterious. It is just easy to do badly if you test the wrong prompts, in the wrong language, and without a scorecard.