How often should I run this check?

Once a month is enough for a manual check, and the same questions each time so the results are comparable. If you change pages or listings, run it again after the engines have had time to recrawl, which can take days to weeks.

Should I use my own logged-in account?

Use a fresh chat and, where you can, a signed-out or private window. A logged-in account can carry memory and location from earlier chats, which can change who is named. Write down the conditions you used so you repeat them.

Does my position in the list matter?

Less than whether you are named at all. The SparkToro and Gumshoe.ai study found rank positions varied widely from run to run, while how often a brand appeared was more stable. Record position if you like, but base decisions on the named percentage.

What if the AI gets my facts wrong?

You can't edit the assistant, but you can fix what it reads. Check the sources it cited, then correct your website, your Google Business Profile and the directories that carry the wrong fact.

Is AI recommending your business? A 30-minute check

To check whether AI recommends your business, write five questions a buyer would ask, run each one three times in ChatGPT, Gemini and Perplexity, and record whether you are named, who is named instead and which sources are cited. Your visibility is the share of runs that name you. The whole check is 45 runs and takes about 30 minutes.

One run tells you almost nothing, because AI answers vary. That is the reason for the repeats, and it is also why the result is a percentage, not a yes or no.

Step 1: write five buyer questions

Use the words a buyer would type, with a service and a place. Avoid your brand name in the question, because the point is to see whether you are named without being asked about.

Pattern	Example
Best service in area	Which clinic is best for dental implants in Hackney?
Who to call for a problem	Who should I call for a cracked tooth in Islington today?
Cost	How much do dental implants cost in London?
Service for a type of buyer	Dentist in Camden that treats anxious patients
Comparison	What should I look for when choosing an implant dentist in north London?

Pick the services that earn you the most and the areas you most want to win. Five is enough to start.

Step 2: set the conditions and write them down

Use a new chat for every run, so earlier answers don’t carry over.
Turn web search on where the assistant offers a switch.
Sign out or use a private window where you can, and note your location setting.
Use the exact same wording for the three runs of one question.

Write the conditions at the top of your sheet. Next month you will repeat them.

Step 3: run and log every answer

Make a sheet with one row per run and these columns:

Column	What to enter
Date and assistant	For example 5 Nov, ChatGPT
Question and run number	Q2, run 3
Named?	Yes or no
Who else is named	The competitors in the answer
Sources cited	The domains the answer links to
Facts correct?	Hours, address, services, price

For Google, search the same questions in a private window and note whether an AI Overview appears at all. Google says AI Overviews show only when its systems judge they add value beyond classic results, so they often don’t appear.

Step 4: calculate your visibility

Visibility is named runs divided by total runs, per assistant and overall.

An example with the fictional Lakeside Dental Studio. In ChatGPT it ran 5 questions three times each, which is 15 runs, and was named in 4. That is 4 divided by 15, or 27%.

Treat small samples with care. A 95% Wilson confidence interval for 4 of 15 runs runs from roughly 11% to 52%. That is wide, so don’t read a move from 27% to 33% as progress. Compare larger gaps: a competitor named in 12 of 15 runs, or a question where you are never named.

There is no official benchmark for a good percentage. The useful comparisons are your competitors’ share on the same questions and your own result next month.

Step 5: read the patterns

Sort your log and look for:

Questions where you are never named. These are your first pages to write or fix.
Competitors who keep coming back. Open their pages and compare them with yours.
Domains that appear again and again as sources. These are the third-party pages worth getting onto. See why AI recommends your competitor instead of you.
Wrong facts. Each one is a source to correct.

Why you must repeat each question

In the SparkToro and Gumshoe.ai study, ChatGPT and Google’s AI returned the same brand list in fewer than 1% of repeated runs of the same prompt. The authors, who published their methodology and raw data, concluded that visibility across many runs is a more stable measure than rank position. The study ran in late 2025, was not peer reviewed and was partly sponsored by an AI tracking company, so read it as strong evidence about variability, not a final word.

What a manual check can’t do

It can’t see what a different buyer in a different place gets.
Fifteen runs is a small sample, as the interval above shows.
It is slow to repeat every month for every service and area.
Consumer apps and the APIs behind them can answer differently, and a manual check only covers the consumer apps you test.

RankWeb automates this: it asks the questions you choose, repeats them, reports who is named instead and which sources are cited, and states the limits in every report, including that API answers can differ from consumer apps. If you only need a one-off look, the steps above are enough.

What to do with the result

Pick one question where you are never named, find the page or listing that should answer it, and fix that first. Then use the five-step method to work through the rest.