How to check if ChatGPT, Gemini and Claude mention your dietitian practice (free, under an hour)

You can check whether ChatGPT, Gemini and Claude mention your practice for free, in under an hour. Ask each assistant the questions a client would ask, in a private session with web search turned on, and write down three things: whether you were named, which websites the assistant cited, and which other dietitians it named. Do it once a month. One check is a snapshot, not a score. Over a few months, though, it shows you where you stand and what to fix first.

All you need is a list of 8 to 12 prompts, a blank spreadsheet and 30 to 45 minutes. Writing the prompt list takes a little longer the first time. After that you reuse it every month, so the check itself is just running prompts and filling in rows. This post is only about checking. For the background, start with how AI assistants choose dietitians to mention.

Step 1: Write 8 to 12 prompts the way clients ask

Write full questions in a client's voice, not keywords, and cover five kinds of question: specialty, city, telehealth state, insurance and "near me". People don't type into an AI assistant the way they type into a search box. In a SparkToro study, volunteers who all wanted the same thing wrote prompts that barely resembled each other. Google's own AI Mode help page suggests asking several versions of a question. So vary the wording on purpose.

The examples below are for a made-up practice: an IBS and PCOS dietitian in Austin, Texas, who sees telehealth clients in Texas and Colorado and is in network with Aetna. Swap in your own specialties, city, states and insurers.

Specialty

1. "I have IBS and I'm tired of guessing what to eat. Can you recommend a registered dietitian who specializes in IBS?"
2. "Who are some good dietitians for PCOS that I could work with?"

City

3. "I'm looking for a registered dietitian in Austin, TX who works with IBS. Who would you recommend?"
4. "Best PCOS dietitians in Austin"

Telehealth state

5. "I live in Colorado Springs. Is there a dietitian who does virtual sessions for gut issues and can see clients in Colorado?"
6. "Online dietitian for PCOS who sees clients in Texas"

Insurance

7. "Does any dietitian in Austin take Aetna? I need help with IBS."
8. "How do I find an IBS dietitian covered by my insurance in Texas?"

Near me

9. "Registered dietitian near me for IBS"
10. "Dietitian near me who takes insurance"

Your name (optional, checks accuracy rather than discovery)

11. "What can you tell me about [Your Name], RDN, in Austin?"
12. "Tell me about [Your Practice Name]."

Three rules for the list. Never include anything from a real client, even paraphrased. Keep the list the same every month so the results are comparable. And if you add prompts later, add them as new rows rather than rewriting old ones.

Step 2: Turn off memory and personalization in each assistant

Use a signed-out, temporary or incognito session in every assistant, so your own history doesn't shape the answer. This matters more than it sounds. If you've ever told ChatGPT that you're a dietitian in Austin, OpenAI says ChatGPT may use saved memories when rewriting a search query. Your future client's ChatGPT doesn't know any of that.

ChatGPT

The simplest option is to open chatgpt.com in a private browser window without signing in. OpenAI's help page says people who are not signed in can also use web search.

If you'd rather stay signed in, open a new chat, select Temporary, and choose Unpersonalized before you send your first message. OpenAI says the unpersonalized option does not use memory, custom instructions or plugins, and you can't change the choice once the chat starts. Your regular memory settings live under Settings, then Personalization, then Memory (OpenAI's memory help).

Gemini

At gemini.google.com, click Temporary chat next to New chat. Google's help page says a temporary chat can't give personalized responses. You can also turn off memory of past chats for your whole account: Menu, then Settings & help, then Personal Intelligence, then turn Memory off (Gemini memory help). Google also says you can generally ask informational questions while signed out.

Claude

Start a new chat and click the ghost icon in the upper right to open an incognito chat. Anthropic says incognito chats aren't saved to Claude's memory or your chat history and are available on every plan. You can also pause memory under Settings, then Memory.

Optional: Google's AI Mode

Many clients still start on Google, so add Google's AI Mode (google.com/ai) as a fourth column if you have time. Use a private window. Google says AI Mode can personalize answers for adults who have search history and personalized recommendations turned on.

Step 3: Run every prompt with web search on, and record the results

Only count answers where the assistant searched the web. Without search, it answers from what it learned in training, and nothing on your current website can change that answer.

• ChatGPT may search on its own. To make sure, select View all tools, then Search, or type / in the message box and choose Search (OpenAI's search help). A searched answer shows citations and often a Sources button.
• Claude: if you see a Web search option under the + button, turn it on. The newer Claude app has no toggle and searches when it helps, so start the prompt with "Search the web:" and only count answers that show citations. Anthropic's help page says responses that use it include citations, and web search is available on all Claude plans.
• Gemini: look for the Sources button under the answer. Google says that if there's no Sources button, Gemini didn't provide any links for that response. Record that as "no sources".

Then work through the list. Open a fresh chat for every prompt, because earlier questions in a conversation shape later answers. Paste the prompt exactly as written. Don't follow up with "what about [your name]?", since a real client wouldn't. With 10 prompts and three assistants, that's 30 answers, which fits in about 20 minutes.

Record each answer as you go, in a simple sheet. Make one row per prompt per assistant in Google Sheets or Excel. Use these columns:

• Date
• Assistant (ChatGPT, Gemini, Claude, AI Mode)
• Prompt number from your list
• Searched the web? yes or no
• Mentioned you? yes or no
• Accurate? If you were named, were your name, credential, city and services right?
• Sources cited: the website names, such as your site, a directory, an insurer's lookup page or a news site
• Other dietitians named
• Notes: anything wrong, odd or surprising

Take a screenshot of any answer that mentions you or gets something wrong. You probably won't see the exact same answer again, so the screenshot is your record.

Step 4: Repeat it monthly, not daily

Run the same prompts, the same way, once a month. Checking every day mostly measures randomness. In the SparkToro study, 600 volunteers ran 12 prompts through ChatGPT, Claude and Google's AI a combined 2,961 times. The researchers found there was less than a 1 in 100 chance that ChatGPT or Google's AI, asked the same question 100 times, would give the same list of brands in any two of those answers.

So a mention that vanishes tomorrow doesn't mean anything changed. The useful question is the trend. Are you named in more of your 30 answers than three months ago? Are the cited sources shifting toward your own site?

The honest limits of a do-it-yourself check

This check shows what a handful of answers looked like on one day, from one place. Keep these limits in mind:

• Answers vary from run to run. Ask the same question twice and you may get a different list. That's how these tools work, not a sign that something broke.
• Answers vary by location. ChatGPT uses an approximate location from your IP address, and OpenAI notes a VPN can change it. Claude may use a location inferred from your IP address for local results. Google says results differ by location, language and device. Your "near me" test from your office shows what a client near your office might see, not a client across the state. For telehealth states, name a city in the prompt instead of relying on location.
• One check is a snapshot, not a score. Thirty answers is a small sample. Being named in 3 answers one month and 5 the next is probably noise, not progress.
• Real clients phrase things in ways you won't guess. Your list is a sample of possible questions, not all of them.

Why paid "AI visibility scores" can mislead

A single number hides how much the answers move, and some scores measure things that aren't stable enough to measure. Some specific problems:

• Rank positions are close to random. The same SparkToro study found you'd need something like 1,000 runs before seeing two lists in the same order, and concluded that tracking your position inside an AI answer isn't meaningful. The researchers did find that how often a name appears across many prompts, each run many times, is a more reasonable measure. One of the two researchers works at an AI tracking company, which the post discloses.
• The vendor chooses the prompts. If the prompts don't match how your clients ask, the score describes someone else's questions.
• A score doesn't tell you why. Your sheet shows which sources the assistant cited. That's the part you can act on.
• No one can promise placement. OpenAI's own help page says: "Placement is not guaranteed." Be wary of anyone who implies otherwise.

If you do consider a paid tool, ask which prompts it runs, how many times each, from which location, whether it's signed in, and whether it publishes its method.

How to read your results, and what to fix first

Read the sources column first. It tells you where each assistant gets its information about dietitians like you. Then look for these patterns.

It named you, but got something wrong

Fix this first, because accuracy matters more than visibility. Open the source it cited and correct that page if you control it. Then make your name and credentials match everywhere. Our guide to showing your credentials clearly on your website covers the wording.

It cited directories, but not your site

This means the assistant found what it needed for that question on directory or marketplace pages, and your own site wasn't one of the pages it chose. Being named through a directory still counts as a mention, so record it. Then do two things. Make sure every cited directory profile is complete and matches your site, since those pages are clearly being read. And give your site a page that answers that exact question: a specialty page that names the specialty, your city, your telehealth states and how payment works. OpenAI also says a site must allow its OAI-SearchBot crawler to be eligible for ChatGPT search results.

It named other dietitians

Open the pages it cited for them. Look at what those pages state plainly that yours doesn't, such as the city, the specialty, the states served or the insurers accepted. Don't copy their words. Write your own facts just as clearly. And don't read the list as a verdict on who is better, since the order changes from run to run.

It never named you at all

This is common for solo practices, and it's fixable. If you didn't show up for "near me" or city prompts, check that your Google Business Profile is complete and accurate. For a step-by-step diagnosis of the usual causes, from search indexes to blocked crawlers, see why ChatGPT doesn't recommend your practice.

What to fix first

1. Wrong facts about you, wherever they came from.
2. Your own website: it exists, it's in Google and Bing, and it states your credential, specialties, city, telehealth states and payment in plain sentences. If you don't have one yet, you can draft a free preview of your site.
3. One page for each main specialty.
4. Directory profiles and your Google Business Profile, all matching your site.

Then run the same check next month. These fixes improve your odds. They don't guarantee a mention.

Common mistakes

• Checking while signed in with memory on. You'll see an answer shaped by your own history, not a client's.
• Counting answers that didn't search the web. They say nothing about your current site.
• Asking leading follow-ups like "what about [your name]?"
• Changing the prompts every month. Then you can't compare months.
• Reacting to a single answer. One missing mention is noise. Look at the trend.
• Rewriting pages to repeat a prompt word for word. It makes pages harder for clients to read. Google says there are no special optimizations or AI files needed to appear in AI Overviews or AI Mode, as long as the site hasn't been excluded in Search Console. Clear, accurate pages are the work.