AI visibility is not a Google ranking position. ChatGPT, Gemini and Google AI may name different companies and sources for the same question. Define what you measure before judging whether visibility improved.
Build a small set of real questions
Choose 10 to 20 questions from actual sales conversations. Group them by intent: learning (“What is AEO?”), comparison (“SEO or GEO for a B2B company?”) and buying (“Who provides an AI visibility audit in Serbia?”). Keep the original wording unchanged for later comparisons.
Do not test only prompts containing your brand name. That measures name recognition, not discovery by a buyer who has never heard of you.
Record the context and answer
- Context: platform, displayed model, date, language, location and signed-in state.
- Result: whether the brand appears, its role, and whether the service and contact details are accurate.
- Sources: exact cited URLs and whether they are owned, third-party or absent.
- Usefulness: whether the answer helps the buyer decide what to do next.
- Evidence: the full answer and screenshot alongside the original prompt.
A VPN and private window change context; they do not create a perfectly controlled lab. Document the settings and repeat the same protocol several times.
Four practical measures
One row represents one question on one platform at one time. Twelve prompts across three systems produce 36 rows per cycle. Compare the next cycle using the same prompts, then calculate:
- Mention rate: answers naming your brand divided by all answers.
- Citation rate: answers citing your URL divided by all answers.
- Description accuracy: how often a mention describes the service and location correctly.
- Next-step quality: whether the answer offers a useful path to your site or contact point.
Do not collapse these into one “AI rank”. A frequent but inaccurate mention may be less useful than a precise citation. Track Search Console impressions and clicks separately: an impression is not an article read.
Turn a finding into a change
If AI misunderstands the service, first improve its explanation on the relevant page. If it knows the service but cannot substantiate expertise, add a verifiable method and work example. If it cites an old page, check canonical URLs and internal links. Then rerun the same questions. Attribute improvement cautiously: AI systems and their sources can change without your intervention.
Start with SEO, AEO and GEO explained, then see the AEO worked example and our AI visibility audit.
Scaling the test? Use the European multilingual playbook for country and language decisions, or the global benchmark method for a multi-region comparison.
