Most founders check their AI visibility the same way: open ChatGPT, type their own brand name, watch it find the website and describe the product, and relax.
That test proves almost nothing. Your buyers don't ask "what is [your brand]?". They ask "what's the best tool for…", "which brand should I buy for…", "X vs Y" — and the real question is whether you are named in that answer.
It matters more every month. ChatGPT passed roughly 900 million weekly users in February 2026. Google says AI Mode passed 1 billion monthly users. And when Google shows an AI summary, Pew Research found people clicked a regular result in 8% of visits, against 15% without one. If the answer doesn't name you, the click often never happens.
Here's the method I use. It works by hand in a spreadsheet, and it's exactly what AI visibility tools automate.
TL;DR — Ask 20–30 real buyer questions (no brand names) in a clean session with web search on, three times each, on every engine you care about. Log whether you're named, your position, who's named instead and which URLs are cited. Visibility = answers naming you ÷ all answers. Share of voice = answers naming you ÷ answers naming any brand you track. Re-run weekly with the same questions.
Step 1: Build the question set
Write 20–30 questions a buyer asks before they know your name. Mix three types:
| Type | B2B SaaS example | Consumer brand example |
|---|---|---|
| Category | "best invoicing software for freelancers" | "best protein powder for beginners" |
| Problem | "how do I stop late client payments" | "protein powder that doesn't cause bloating" |
| Comparison | "FreshBooks vs Wave" | "whey vs plant protein" |
Leave out every question that contains your own brand name. You'll "win" those almost every time, and they inflate every number downstream. Track them separately if you want to see how AI describes you.
Step 2: Use a clean session, with web search on
- Log out, or use a temporary chat with memory off. Personalisation and chat history change the answer.
- Make sure the answer uses web search. For current products, ChatGPT, Claude, Gemini and Perplexity search the web, read a handful of pages, and summarise them. An answer from model memory can't reflect a page you published last week — and it isn't what most buyers see.
- If you sell in several countries, ask from each one (or put the country in the question). Answers and sources differ by market.
Step 3: Ask each question at least three times
AI answers aren't deterministic. The same question can name four brands in the morning and a different three in the afternoon. One run is an anecdote; three runs on different days is a measurement.
Step 4: Log every answer
A flat CSV is enough:
date,engine,country,question,named,position,rivals_named,cited_urls
2026-11-08,chatgpt,US,best invoicing software for freelancers,N,,"FreshBooks; Wave; Zoho Invoice","https://example-review-site.com/best-invoicing; https://www.reddit.com/r/freelance/..."
2026-11-08,chatgpt,US,how do I stop late client payments,Y,3,"FreshBooks; Bonsai","https://example-blog.com/late-payments"
Count each brand once per answer, however many times the answer repeats it.
Step 5: Do the math
Three numbers matter:
- Visibility (mention rate) = answers that name you ÷ all answers
- Share of voice = answers that name you ÷ answers that name any brand you track
- Position = where you appear when you are named
Worked example: 20 questions × 3 runs = 60 answers. You're named in 9 of them. Across all 60 answers, the brands you track are named 50 times in total (counting each brand once per answer).
| Measure | Calculation | Result |
|---|---|---|
| Visibility | 9 ÷ 60 | 15% |
| Share of voice | 9 ÷ 50 | 18% |
| Position | average of your "position" column | e.g. 2.7 |
In Google Sheets, with the log above in columns A–H:
Visibility: =COUNTIF(E2:E,"Y") / COUNTA(D2:D)
Average position: =AVERAGEIF(E2:E,"Y",F2:F)
ChatGPT only: =COUNTIFS(B2:B,"chatgpt",E2:E,"Y") / COUNTIF(B2:B,"chatgpt")
Pool 7–14 days of answers before you compare weeks. A single day is too noisy.
Step 6: Read the citations — that's where the fix is
The cited_urls column is the most useful thing you'll collect. It shows which pages the AI trusts for your category: usually "best X" lists, review sites, comparison articles, Reddit threads and YouTube videos. Several 2026 studies found third-party comparison content among the most-cited sources for commercial questions.
So the to-do list writes itself:
- Sort the cited domains by how often they appear.
- For the top ones, get included: pitch the article's author, claim your review-site profile, answer the Reddit thread honestly (and say who you are).
- For questions where no page is cited at all, publish one page on your own site that answers the exact question in its first paragraph.
Every engine answers differently
Don't stop at ChatGPT. The same question regularly gets different brands from ChatGPT, Claude, Gemini and Google's AI Overviews, because each one searches differently and cites different pages. Track each engine separately and report the number per engine — a decent overall score can hide an engine where you're invisible.
Automating it
By hand, 30 questions × 3 runs × 4 engines is 360 answers a week. That's why AI visibility tools exist. The main ones, with published prices (November 2026):
| Tool | Starts at | Engines |
|---|---|---|
| Otterly.ai | $29/mo · 15 prompts | ChatGPT, Perplexity, AI Overviews, Copilot; Gemini, AI Mode, Claude as add-ons |
| Peec AI | $95/mo · 50 prompts | 3 models per plan; more on Enterprise |
| Semrush AI Visibility Toolkit | $99/mo per domain | inside Semrush |
| Aurvikon | $149/mo · 25 questions, daily | ChatGPT + AI Overviews on every plan; Gemini and Claude by plan |
| Ahrefs Brand Radar AI | from $199/mo + an Ahrefs plan | $699/mo for all platforms |
| Profound | free trial, then Enterprise by quote | up to 9 engines |
Disclosure: I build Aurvikon, so I'm not neutral. Prices change often — check each vendor's pricing page.
Whatever you use, check three things: it collects answers with live web search, it leaves out questions containing your brand name, and it shows the cited URLs. A tool that can't do those can hand you a flattering number you can't act on.
Common mistakes
- Checking once and treating it as the truth.
- Counting questions that contain your brand name.
- Only checking ChatGPT.
- Reading the answer but ignoring the sources.
- Changing the question list every week — then nothing is comparable.
If you want a quick look before you build the spreadsheet, our free check asks ChatGPT five buyer questions for your category with live web search and shows who gets named: aurvikon.com/?check=1. There's also a longer version of this method on our site: How to check whether ChatGPT recommends your brand.
How are you tracking this today — by hand, with a tool, or not yet? I'm happy to answer questions in the comments.
Top comments (0)