The short answer
To check AI visibility, run a fixed list of buyer questions several times in each AI engine, record whether you are mentioned, cited and described correctly, and report the share of runs. Then add Google's generative AI report, Bing's AI Performance report and ChatGPT referrals for the data those companies publish.
What AI visibility measures
AI visibility is how often, and how accurately, AI engines present your brand when people ask questions you should appear for. If the term is new, start with what GEO is. It is not one number. Record four separate things, because each has a different cause and a different fix.
| Measure | The question | What to write down |
|---|---|---|
| Mention | Does the answer name you? | Yes or no for each run |
| Citation | Does it link to one of your pages? | The URL it cited |
| Accuracy | Are the facts about you right? | Each wrong or outdated fact |
| Recommendation | Are you suggested to the buyer? | Yes or no, and who is named beside you |
Why one check tells you almost nothing
AI answers change from one run to the next. SparkToro and Gumshoe had 600 volunteers run 12 prompts through ChatGPT, Claude and Google's AI results, 2,961 runs in all. They found less than a 1 in 100 chance of ChatGPT or Google's AI giving the same brand list in any two of 100 runs. The same list in the same order came up less than once in 1,000 runs.
Whether a brand shows up at all was much steadier. One cancer hospital, City of Hope, appeared in 69 of 71 ChatGPT answers but came first in only 25. That is why the authors treat the share of runs mentioning a brand as a reasonable measure, though not a precise one, and why position inside an answer is not worth tracking.
The authors point out that they are not professional researchers and that the work was not peer-reviewed. It ran in November and December 2025, so newer models may behave differently.
One brand across 71 ChatGPT answers
City of Hope appeared in 69 of 71 ChatGPT answers and came first in 25.
Showing up was steady. The order was not. Across the study's 2,961 runs, the chance of the same list twice in 100 runs was under 1 in 100.
SparkToro and Gumshoe, late 2025
What the official reports can and cannot show
Google and Microsoft now publish real data about their AI results. Use it first, because it is the only data here that comes straight from the engine. It covers only those companies' own products.
| Source | Covers | Shows | Limits |
|---|---|---|---|
| Google Search Console: generative AI report | AI Overviews and AI Mode | How many times links to your site were shown, by page, country, date and device | Usually hidden until a site has enough impressions. It shows that links were displayed, not what the answer said |
| Google Search Console: standard Performance report | All web search, including AI features | Clicks and impressions | AI features are combined with normal results |
| Bing Webmaster Tools: AI Performance | Copilot, AI summaries in Bing and select partners | Citations, cited pages and the phrases used to retrieve them. Citation Share is in preview | Citations are not clicks or placement, and the query data is a sample |
| ChatGPT Business Hub: insights | ChatGPT answers, for a claimed business | Mentions that reference your website, clicks on links to it and discovery topics, for the last 28 days | Beta, and only for businesses that can claim. Low counts are withheld, and no prompts are shared |
| Your analytics: ChatGPT referrals | ChatGPT search | Visits, which ChatGPT tags with utm_source=chatgpt.com | Only visits. A mention nobody clicks leaves no trace |
OpenAI added the ChatGPT row in November 2026, when it opened Business Hub; our guide on how to get your business on ChatGPT covers who can claim. Even so, no official report tells you what ChatGPT, Claude or Perplexity actually said about you. For that you need your own test.
How to run your own AI visibility check
- 1Write 20 to 30 questions your buyers really ask, across five types: the best provider, a specific need, your location, one company against another, and what it costs.
- 2Add three questions about your brand by name, to test what engines believe about you.
- 3Pick the engines your buyers use, and test each one the way a buyer would, with web search on where the product offers it.
- 4Run every question at least five times per engine, in a new chat each time, signed out or with memory off.
- 5Record the four measures from the table above for every run, in one spreadsheet.
- 6Keep the question list frozen. Add new questions to a separate list, so your trend stays comparable.
- 7Repeat on the same schedule, and compare the share of runs instead of single answers.
Nobody has established how many runs are enough; SparkToro lists it as an open question. Five per question is our practical minimum, not a statistical standard. What matters most is using the same number every time.
How to score the results
For each engine, divide the runs that mention you by the total runs. Do the same for citations and recommendations, and keep a plain count of wrong facts. Then run the same sums for your two or three closest competitors, because a percentage only means something next to theirs.
| Engine | Runs | Mentioned | Cited | Wrong facts |
|---|---|---|---|---|
| Engine A | 120 | 48 (40%) | 18 (15%) | 2 |
| Engine B | 120 | 30 (25%) | 24 (20%) | 0 |
| Engine C | 120 | 66 (55%) | 6 (5%) | 4 |
Buyer prompt
ChatGPT
Perplexity
AI Mode
Copilot
Best CRM integrator in Dubai
Zoho and WhatsApp setup help
Who automates real estate leads
Google Search Console
AI Overviews and AI Mode counted in the Performance report
Bing AI Performance
How often your pages are cited in Copilot answers
What to fix first
| What you find | Likely cause | First move |
|---|---|---|
| No Google AI impressions | Pages not indexed or not eligible to show with a snippet | Check indexing in Search Console and your generative AI setting |
| Missing from ChatGPT search | OAI-SearchBot may be blocked | Check robots.txt. OpenAI says opted-out sites are not shown in ChatGPT search answers |
| Named, but not cited | Other pages answer the question better | Publish a direct, specific answer on your own page |
| Wrong facts | Old or conflicting information | Correct it at the source: your site, profiles and directories |
| Competitors named instead | They appear on the lists engines read | Earn genuine mentions on those same sites |
| Unknown to engines by name | Too little consistent information about you | Make your name, category and location the same everywhere |
AI search
Allow to be cited
Training
Your choice
OpenAI
OAI-SearchBot
GPTBot
Anthropic
Claude-SearchBot
ClaudeBot
Perplexity
PerplexityBot
None listed
Googlebot
Google-Extended
OpenAI: sites that block OAI-SearchBot are not shown in ChatGPT search answers.
Free tools, paid trackers or your own sheet
| Option | Good for | Limit |
|---|---|---|
| A free check, such as our AI Visibility Check | A quick first look at what a language model knows about your brand | Ours uses five fixed prompts against one model, without live web search |
| A paid AI tracking tool | Many prompts, many engines and automatic history | SparkToro lists whether API results match what users see as an open question. Ask each vendor how it collects data |
| Your own spreadsheet | Full control, at no software cost, and it shows what buyers really see | It takes hours each month |
Whichever you use, be skeptical of precise claims. Google says no third-party tool has access to its internal ranking or AI systems.
How often to check
Monthly is enough for most businesses. AI answers move day to day, but the things you can change, such as your pages, facts and mentions, take weeks to show up. Check the official reports weekly if you enjoy it, and run the full prompt test once a month. If you would rather not run it yourself, our generative engine optimization service includes monthly tracking.