The source of the mismatch
Claude cited older opening hours from an external business directory. In a separate test without web search, models attributed a service to the practice that it does not offer.
VISIBILITY IN RECOMMENDATIONS GEO / AEO
I measure whether ChatGPT and other services recommend your business.
The report shows what they say about you, which sources they use and what is worth changing.
For local and national services in the UK and US.
Geotify is a service by L9 Studios. I measure whether ChatGPT, Gemini and Claude recommend your business and suggest what to change.
Your report
Who recommended you and who they named instead. Which sources they used and whether your details were correct. Reported separately for each service.
Fictional business · coach hire in London
4of 30
Named in a recommendation
30 questions, each asked once
Green marks an answer that named the business
In this example, the service found the website, linked to it and gave the correct phone number. Those are three separate findings.
18of 60
Named in a recommendation
30 questions, each asked twice
Green marks an answer that named the business
In this example, the service names the business most often, but gives no website link and the wrong street number. Frequent mentions do not mean accurate details.
2of 30
Named in a recommendation
30 questions, each asked once
Green marks an answer that named the business
In this example, the website did not appear in the recorded sources. A single measurement cannot show that the service does not know it.
WHAT YOU RECEIVE
We agree the scope, price and delivery date in writing before you order. Measurement and implementation are separate services.
See which answers named you, what they said and who they named instead. The report explains the uncertainty and separates incomplete results.
A PDF report and a CSV file with the recorded answers and sources. You can trace each finding back to the answer behind it.
Specific proposed changes to your website, profiles or information elsewhere, ordered by priority. You can implement them with your own team or ask me to quote the work.
A real measurement / MySmile, Prague
After launching the new MySmile website, we checked how AI answers describe the orthodontic practice. They often mentioned it, but some opening hours still came from outdated directories.
Measured 1–2 October 2026 · Initial snapshot
Visit the MySmile websiteanswers mentioned the practice
30 Czech questions. Answers collected through provider APIs with web search. Gemini repeated on a second day. A mention can include a reservation.
Same question, three services„My child’s teeth are growing crooked. Where should we go for orthodontics in Prague 5?“
Claude cited older opening hours from an external business directory. In a separate test without web search, models attributed a service to the practice that it does not offer.
Unify the name and opening hours in directories, link verified profiles to the website and distinguish the practice from unrelated products with the same name. Profile changes require access or approval from the owner.
Record each completed change, then repeat the same questions under comparable conditions. This is an initial measurement. We have not yet measured an uplift after the repairs or an increase in enquiries.
All 120 planned main answers were available and manually reviewed. The 24 answers to brand questions and 45 follow-up answers are separate and do not enter the 88/120 result. This measures the practice’s appearance in the recorded answers, not treatment quality or the likelihood that a patient will choose it. The practice currently states that it is not accepting new patients; correct capacity notices were not treated as criticism.
FREE QUICK TEST
I ask ChatGPT, Gemini and Claude one question a customer might ask. Usually within a minute, you can see whether they recommend you and who else they name.
The test uses developer APIs. The apps your customers use may answer differently.
LET’S START WITH YOUR WEBSITE
Tell me what you offer and who you want to reach.
Include your website in the message.
This message will open
Hi, I would like to check whether our company is recommended.
Add your website above or in your message.
You send the message yourself from your own app.
FOR AGENCIES
I handle the measurement, report and repairs under my own name. You receive a 20% referral commission on the client’s paid measurement and repair package with a repeat measurement.
GEO GUIDES
How to read your results, what to fix on your website and how to track visits from ChatGPT and other services. Practical guides you can follow.
How I measure
I ask ChatGPT, Gemini and Claude the questions your customers would ask. 30 questions without your name, with web search enabled. I then review each answer: whether it names you, who it names instead and which sources it uses.
I follow the same procedure so the results can be compared with a later measurement. Your website is enough to start; no logins or passwords. You decide on repairs after seeing the results.
Preparation
Send your website, the service you offer and the area you serve. I select a question set for local or national services in the UK or US, in the customer’s market.

Preparation
I ask as a customer would: who is best, a top three, comparisons, urgent needs, prices and reviews. Your name is left out so the answer cannot simply repeat it. If you name competitors your customers mention, I add “an alternative to…” questions and count them separately. I leave out questions comparing medical providers for medical practices and publicly funded care.

“Who is best for roof repairs in Manchester?”
“What are the top three roof repair companies in Manchester? Rank them.”
“I need an urgent roof repair in Manchester. Who can help quickly?”
Measurement
I ask with web search enabled: ChatGPT and Claude once, Gemini twice. Gemini is repeated because its answers vary more between runs. I include today’s date and, for local businesses, an approximate location where supported. For the first 15 questions, I also ask a follow-up in the same conversation in each service. Google checks are optional.

30×(1 + 2 + 1)=120recommendation answers. These make up the main metric.
Optional checks
Measurement
What the services know about you, how to contact you, your reviews and customer complaints. I ask with and without web search to separate stored knowledge from information they look up. I check phone numbers, email addresses and postal addresses against your actual details, and opening hours if you provide them. When a detail is wrong, I look for its source.

24 additional answers. I review descriptions of your services manually.
Analysis
Did your name come up? Did the service recommend you, mention you or advise against you? Who did it name instead? Did it search, which websites did it find and which did it cite? I review name variants and uncertain cases manually, and count them only after confirming them.

Found≠cited
Analysis
I check access for the crawlers used by ChatGPT, Gemini, Claude and Copilot. I inspect the homepage without JavaScript, machine-readable business details and contact information. I also check for conflicting details on your website and broken links cited in the answers. Separately, I show how easily an AI agent that searches or orders for a customer can load and use your website. That is not the same as being recommended in the answers.

Results
How many answers named you, out of the total, with an uncertainty range. Which details were wrong and which websites or directories supplied them. Which sources feature your competitors but not you, and which customer questions lack a useful page on your website. I prioritise repairs by their likely impact, with a specific finding behind each one. You receive all the answers in a spreadsheet.

Optional next step
Repairs and a repeat measurement are separate orders, with scope and price agreed in advance. I repeat the same questions and report the change and its uncertainty. If the data does not support a change, I say: no clear difference found.

What this measurement cannot tell you. I ask ChatGPT, Gemini and Claude through developer APIs, without a customer account or its history. A customer’s app may answer differently. Optional Google checks use one answer per question, so I treat them as a snapshot of that day. I can check whether a Google AI Overview cites your website, but cannot reliably count its business mentions because the tool does not return complete lists. This measurement does not include website traffic.
What I record with the results. The model and settings for each service. If they change between measurements, I show the results side by side without treating the difference as a measured improvement.
BEFORE YOU CONTACT ME
I can promise the work.
The services decide who to recommend.
No. I can document what the measurement found, make the agreed repairs and compare a later measurement. I cannot control the answer or your position in it.
One answer is one sample. I normally ask 30 questions that customers would ask. Gemini is repeated because its answers vary more between runs. This helps distinguish regular mentions from occasional ones. I assess each service separately.
Often, changes to your existing content or information elsewhere are enough. If I find a technical limitation, I explain it and propose a specific fix.
The English service supports local and national service businesses in the UK and US. I confirm the customer market, question set and scope before an order. Online shops and product brands are outside this English pilot.
Not exactly. I measure ChatGPT, Gemini and Claude through their developer APIs, without a customer account or its history. I include today’s date and, for local businesses, an approximate location where supported. A customer’s app may use their account, precise location or earlier messages. I explain these differences in the report. The market, language and scope of any optional Google checks are agreed separately.
We agree on an interval based on the changes made. A repeat measurement usually makes sense after several weeks. A better result is not guaranteed, even after that time.
Yes, as an optional extension. I check AI Mode, whether AI Overviews cite your website, search results for short “service + location” queries and your exact business name. The market, language and available checks are agreed before the measurement. I cannot reliably count how often an AI Overview names you because the tool does not return complete business lists. An overview will not appear for every question; I record that too.
Automated advisor. Answers are generated by an OpenAI model.
Hi, I am GEOs.
I can explain how measurement works and what to look at on your website. I will include the sources behind my answer.
I am an automated advisor. An OpenAI model answers using sources checked by Lukáš. I do not measure your website or check actual recommendations.
Prefer to write to Lukáš? Email Lukáš · WhatsApp