AI visibility is how often, and how accurately, AI answer systems such as ChatGPT, Perplexity, Gemini and Google AI Overviews mention or cite your brand when people ask relevant questions. It is harder to measure than a ranking because answers vary by run, wording and user. This guide gives a method that holds up, and it flags where the numbers are weak. It sits alongside our generative engine optimization services work and our content audit for AI search guide.
Why a single rank number does not exist here
In classic search you can report position 4 for a keyword. An AI answer has no stable position. The same prompt can return different sources on different runs, and results can differ by country, account state and model version. Any AI visibility figure is therefore a sample, and it should be reported with its sample size and date.
The metrics worth tracking
Mention rate
Out of a fixed set of prompts, the percentage of answers that name your brand. Count a mention only when the context is accurate. A negative or wrong mention is a finding, not a win.
Citation rate and cited URLs
The percentage of answers that link to your domain, and which pages are linked. This shows which content AI systems actually use. Perplexity and ChatGPT search show sources, and Google AI Overviews show supporting links.
Share of answer
Your mentions or citations compared with competitors for the same prompts. It is the closest equivalent to share of voice and is useful because it is relative, so it survives some of the run-to-run noise.
Accuracy and sentiment
Whether facts such as services, locations and pricing model are described correctly. Errors here are often fixable by clarifying your own pages.
Referral traffic and conversions
Visits from AI platforms in your analytics, and what those visitors do. This is the hardest evidence of business value, though it undercounts, as explained below.
Step 1: build a prompt set
Write 30 to 60 prompts that mirror real buyer questions: category questions, comparisons, problem-led questions and branded questions. Pull wording from sales calls, support tickets and Search Console queries. Keep the list fixed so results are comparable over time, and tag each prompt by funnel stage.
Step 2: sample consistently
Run each prompt on each engine you care about, ideally several times, in a logged-out or clean session. Record the date, engine, the full answer, mentioned brands and cited URLs in a spreadsheet. A manual pass is enough to start and is free. Tools can automate the same loop at scale. Be cautious about any vendor score that blends engines into one number, because a gain in one engine can hide a loss in another. Ask to see each engine separately.
Step 3: check what Google reports
Google documents that traffic from AI Overviews and AI Mode is included in the Web search type in the Search Console Performance report (Google Search Central). It is not broken out separately, so you cannot isolate it directly. Watch impressions and clicks for the informational queries that trigger AI features and annotate changes.
Step 4: measure referral traffic
In Google Analytics 4, build a channel or segment from session source matching AI platform domains such as chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com. Google explains how channel groups work in its Analytics help. Expect undercounting: some apps strip referrers, and a person may read the answer and later search your brand name instead of clicking. Treat branded search growth as supporting evidence, not proof.
Step 5: check access and eligibility
Visibility cannot start if crawlers are blocked. Confirm robots.txt and any CDN or firewall rules allow the bots you want. OpenAI documents OAI-SearchBot for search, and you can check your server logs for these user agents. Our free AI visibility audit runs a first pass on access and on-page signals, and the llms.txt checker validates that file if you use one.
Step 6: review monthly, act quarterly
Run the same set monthly and compare. Look for pages that earn citations and ask why: usually a clear answer, a table, or original data. Look for prompts where competitors appear and you do not, and decide whether you have a content gap, an authority gap or an access problem. Changes in one month are often noise, so wait for a trend before reorganising a strategy.
Limits to state in every report
- Samples are small and answers are non-deterministic.
- Tools simulate prompts and may not match what real users see.
- Personalisation and location can change results.
- Correlation with traffic or sales is not causation.
If you want a baseline and repeatable reporting set up for your brand, our team can help through GEO services and semantic SEO. You can also contact us.
Frequently asked questions
What is AI visibility?
It is how often and how accurately AI answer systems mention or cite your brand when users ask relevant questions.
Can I track AI visibility for free?
Yes. Build a fixed prompt list, run it on each engine on a schedule, and log mentions and cited URLs in a spreadsheet. Add Search Console and GA4 referral data.
Does Search Console show AI Overview traffic?
Google documents that traffic from AI features is included in the Web search type in the Performance report, but it is not reported as a separate segment.
Why do my results change every time I run a prompt?
AI answers are generated, and sources can vary by run, model version, location and session. This is why you sample repeatedly and report ranges.
How often should I measure?
Monthly is enough for most sites. Weekly checks mostly capture noise unless you are testing a specific change.
