Write down 15 to 20 questions your buyers ask, put each one to ChatGPT, Claude, Gemini and Perplexity three times, and record whether any answer names your business and which websites it cites. Then repeat the same questions, with the same number of asks, every week, because one answer on one day is mostly noise and only the comparison across weeks tells you where you stand.
That is the whole method. The rest of this page covers how to do it by hand, which questions to use, why repetition matters, what a useful report contains and what it costs to automate.
How do I check by hand?
You need a spreadsheet and a free afternoon.
- Open each assistant in a fresh session: logged out or in a private window, with no saved memory or custom instructions. An assistant that already knows where you work is not a fair test.
- Make sure the assistant can search the web. Answers about vendors, providers and prices depend heavily on what it retrieves at that moment.
- Paste each question word for word, the same way every time.
- Start a new chat for every ask, so the previous answer does not shape the next one.
- Record one row per answer: date, assistant, question, ask number (1, 2 or 3), whether your business is named, which competitors are named, and which URLs are cited.
Twenty questions across four assistants with three asks each is 240 answers. By hand that is tedious, which is why most people do it once and stop. Once is better than never, but read it as a snapshot, not a verdict.
Which questions should I ask?
Ask your buyers' questions, not questions about yourself. 'What is [your company]?' tests whether the model has heard of you. It says nothing about whether you come up when a stranger with a problem asks for help, which is the moment that brings business.
Good test questions come from sales calls, support tickets and the phrases customers use in emails. A useful set mixes three kinds:
- Choice questions: 'Which agency should I hire for...', 'What is the best tool for...', 'Who does X in [city]?'
- Cost and process questions: 'How much does X cost?', 'How long does X take?'
- Problem questions: 'Why does X happen and how do I fix it?'
Expect names mostly on the first kind. In a test on 23 September 2026 with 20 German business questions, the model answered 10 without searching at all and cited nobody. Where it did search, it cited primary sources such as laws, ministries, the OECD and large accounting firms on 9 of 10. Firms were named only on 'who should I hire' and 'how to choose' questions. For an orthopaedic practice in Bonn on 20 October 2026 the pattern was the same: 'who to choose' questions named practices, while questions about costs, risks and treatments named none.
So cost and problem questions are rarely where you get named. They are where you learn which sources the assistants trust, and that is the second thing you need to know.
Why is one ask not enough?
Because the same question asked twice often gets two different answers. In the 23 September test, two runs of the same question named two different sets of firms. Ask once and see your name, and you may be celebrating a coin toss. Ask once and miss it, and you may be worrying over one.
The rules we use in our weekly measurement:
- Three asks per question per assistant. A business counts as named for that question if any of the three asks names it.
- The same questions every week. Swap questions and you are measuring something else; the numbers no longer compare.
- The same number of asks every week. Three asks one week and one the next will make you look worse for no real reason.
Hold those three things fixed and a change in the count means something. Change any of them and it does not. In the figures below, one 'answer' is one question put to one assistant, counted as named if any of its three asks names the business.
What should an AI visibility report show?
A single percentage is hard to act on. A useful weekly report answers four questions:
| What to report | Why it matters | Example from our measurements |
|---|---|---|
| Named answers out of total, per assistant | Shows where you stand and which assistant lags | Bonn practice, 20 October 2026: named in 12 of 80 answers; Claude 5 of 20, Gemini 4, ChatGPT 2, Perplexity 1 |
| Which competitors are named instead | Shows who the assistants treat as the answer | 'Which GEO tool' questions, 21 October 2026, English answers: Profound in 12, Semrush 9, Otterly.AI 8, Peec AI 8, Ahrefs 7 |
| Which websites are cited | Shows the pages the assistants rely on, so you know where you need to be present | Bonn practice: the directory jameda.de was cited in 12 answers; competing practices' own websites were the most cited sources |
| Whether your own pages are cited | Often the first sign that your content is being used | On 20 October 2026 Perplexity cited an article we had published on a client's site, a German company-formation advisory |
Our own starting point shows why the cited-sites column matters. On 21 October 2026 we put 20 English and 20 German questions about getting a business into AI answers to all four assistants, three times each. We were named in 0 of 160 answers and none of our pages was cited; the most cited sources were the blogs of GEO agencies and tool vendors. The full numbers are in We Checked Our Own AI Visibility and Were Named in 0 of 160 Answers.
Citations are easy to collect programmatically. OpenAI's web search documentation states that when the web search tool is used, the response includes inline citations as url_citation annotations with the URL, title and location of each cited source.
Can I see AI mentions in my analytics instead?
Mostly not. In about two weeks of logs from three of our course websites, roughly 1.2 million requests in the two busiest logs, human visitors arriving from AI assistants numbered: Copilot 1; ChatGPT, Perplexity, Gemini, Claude, Meta AI and others 0. Assistants usually answer in place rather than send a click, so referral traffic tells you almost nothing about whether you are named.
Server logs show something more useful: live retrievals. OpenAI documents separate crawlers, including GPTBot for training, OAI-SearchBot for ChatGPT's search index and ChatGPT-User for fetches triggered by a user. Perplexity works similarly; an analysis of its crawler documentation describes Perplexity-User as fetching pages live only in response to a specific query. In one fortnight an edtech platform we work with received 1,918 such fetches on content pages, and a client blog of a few dozen articles received 56 across 24 articles. A live fetch means a page was read while someone was asking; it does not tell you whether you were named. Use it alongside the asking method, not instead of it.
How much does it cost to automate with APIs?
The API fees are small. With prices of October 2026, one question asked once with web search costs us about:
| Assistant | Cost per ask with web search | 20 questions x 3 asks |
|---|---|---|
| Claude | about $0.069 | about $4.14 |
| ChatGPT | about $0.010 | about $0.60 |
| Gemini | about $0.009 | about $0.54 |
| Perplexity | about $0.0013 | about $0.08 |
A full report of 20 questions x 4 assistants x 3 asks comes to about $5.40, so a weekly report runs a little over $20 a month in fees, most of it Claude.
The real cost is the work around the calls: writing and maintaining the scripts, matching your business under every name variant (legal name, brand, domain, common misspellings), pulling cited domains out of each response, storing results so weeks line up, and keeping the question set frozen. Treat API answers as a consistent proxy; what a person sees in the consumer app may differ in detail.
What should I do if I am not named?
Check access before anything else. On our own site, a robots.txt rule meant to hide one redirect page (/b) was written as a prefix, so from 11 August to 21 October 2026 it also closed /blog, /business and every business article to all crawlers. Bing Webmaster Tools listed our business page as 'discovered but not crawled' the whole time. We changed the rule to /b$ on 21 October 2026. Is My Website Blocking GPTBot, ClaudeBot and PerplexityBot? walks through the checks. For Google, its AI features documentation says the same Googlebot and robots.txt controls as normal Search apply, and the Google-Extended token does not affect inclusion in Search or its AI features.
Next, do not assume Google indexing is enough. On 19 October 2026 we asked assistants for the exact titles of 18 articles, each older than three weeks, on three client sites. ChatGPT found 1 of 18 and Claude 7 of 18, while Google had indexed 29 of 30 articles of one of those sites. Does ChatGPT Use Bing? covers the Bing and IndexNow side.
Then build surface. A site with complete markup but only 75 URLs was absent from answers on its own article's exact topic, while a site with about 190,000 crawlable URLs was crawled constantly. Pages that answer your buyers' actual questions are what get read. Why Doesn't ChatGPT Mention My Company but Names My Competitors? goes into the reasons.
To see a first result for your own site, the free check at itgrows.ai generates five buyer questions from your homepage and shows ChatGPT's and Perplexity's answers in 25 to 40 seconds. ItGrows runs the full version every week, your buyer questions to all four assistants three times each, while publishing articles on your domain and fixing crawler access and indexing, for $499 a month (plus a one-time $499 setup, free with 3 months or a year). No one can guarantee that an assistant will name a business, and we do not.
Frequently Asked Questions
Can I just ask ChatGPT whether it knows my company?
You can, but it tests the wrong thing. A buyer who does not know you yet asks about their problem or about which provider to choose, and that is the question you need to appear in. Use buyer questions and keep your brand name out of them.
How often should I check?
Weekly is a practical rhythm: frequent enough to see a trend, cheap enough to sustain at about $5.40 per full report in API fees. What matters more than frequency is keeping the same questions and the same three asks each time, otherwise weeks cannot be compared.
Does being named by an assistant actually bring customers?
Clicks from assistants are rare, so the effect shows up as buyers who arrive already knowing your name. Clients report it that way: in October 2026 FreightSizer reported 2 new B2B customers in one month from ChatGPT, MagiScan reported hundreds of installs a month from ChatGPT, Claude and Gemini, and appslift reported 2 to 5 leads a month from AI assistants.
Can anyone guarantee that an assistant will name my business?
No. Answers vary from one ask to the next and the assistants decide what to retrieve and whom to name. What you can control is access, indexing and the pages you publish, and what you can measure is how often you are named over time.