AI assistants do not give a fixed answer. Ask the same question twice and the list of recommended businesses can change, and ask a different assistant and it often changes a lot. In our Lisbon padel report, the top club was named in 39 of 42 Gemini answers but only 10 of 42 ChatGPT answers. That is why a single manual check tells you very little, and why AI visibility has to be measured as a share across many answers.
TL;DR
- The same 14 buying questions, asked 3 times on each of 3 engines, produced 126 answers per city.
- In Lisbon, Clube de Padel was named in 93% of Gemini answers but 24% of ChatGPT answers.
- Even on one engine with one question, the top club appeared in all 3 runs only 40% of the time.
- In New York, Padel Haus was named in 121 of 126 answers (96%), almost identically on every engine.
- Consistency across engines and runs is the real sign of a category leader. Measure share of voice, not a single screenshot.
Why does ChatGPT give different answers to the same question?
Language models generate answers word by word, choosing among likely options, so the same prompt can lead to a different path each time. When an assistant searches the web, the pages it retrieves can also differ between runs. Each engine also has its own search index and its own preferences for sources, as our citation data from 961 sites shows. The result is variation on two levels: between runs on the same engine, and between engines.
How did we test it?
We used two market reports from Padel Visible, our padel industry brand, built with the same method:
- 14 buying questions per city, such as "What are the top padel clubs in Lisbon right now?" and "Which padel clubs in New York City have the best coaching programs?"
- Each question asked 3 times on ChatGPT, Perplexity and Gemini, all with web search enabled.
- 126 answers per city in total (42 per engine), with every club name extracted and matched to a single canonical name.
The Lisbon data was captured on October 12, 2026 and the New York data in August 2026. Both reports are published: best padel clubs in Lisbon and best padel clubs in New York City.
How much did answers differ between AI engines?
In Lisbon, a lot. Here is how many of the 42 answers on each engine named the top five clubs:
| Club | ChatGPT | Perplexity | Gemini | All 126 answers |
|---|---|---|---|---|
| Clube de Padel | 10 | 19 | 39 | 54% |
| Lisboa Racket Centre | 21 | 20 | 20 | 48% |
| Rackets Pro | 19 | 10 | 31 | 48% |
| W Padel Country Club | 15 | 18 | 13 | 37% |
| Padel Campo Grande | 7 | 10 | 26 | 34% |
Clube de Padel led overall because Gemini named it in almost every answer. On ChatGPT and Perplexity, Lisboa Racket Centre was named more often. A business owner who checked only ChatGPT would think Lisboa Racket Centre leads the city; one who checked only Gemini would think Clube de Padel does. Both would be partly right.
The engines also differed in how widely they spread recommendations. Across its 42 answers, ChatGPT named 62 different clubs, while Perplexity and Gemini each named 33. Gemini listed the most clubs per answer (7.0 on average, against about 4.5 for the others) but kept returning to the same ones.
How much did answers change between runs?
Even on one engine, asking one question, results moved. For each of the 42 combinations of engine and question, we checked whether a club appeared in all 3 runs, in some, or in none:
| Club | All 3 runs | 1 or 2 runs | No runs |
|---|---|---|---|
| Clube de Padel | 17 | 12 | 13 |
| Rackets Pro | 17 | 8 | 17 |
| Lisboa Racket Centre | 15 | 10 | 17 |
| W Padel Country Club | 11 | 9 | 22 |
| Padel Campo Grande | 10 | 9 | 23 |
The leading club appeared in every run only 17 times out of 42 (40%). In 12 cases it appeared in some runs and not others. If you ask once and do not see your business, that does not mean the assistant never names you. If you ask once and do see it, that does not mean it always does.
What does a real category leader look like?
New York looked completely different. Using the same method, Padel Haus was named in 37 of 42 ChatGPT answers, 42 of 42 Perplexity answers and 42 of 42 Gemini answers: 121 of 126 in total, or 96%. The next club, Reserve Padel, was named in 97.
That is what an established leader looks like in AI answers: named by every engine, on almost every question, on almost every run. It happens when a business is described clearly and consistently across many independent sources, so every engine and every run arrives at the same conclusion. Lisbon's market is more fragmented, and the engines disagree because the web does.
How should you measure your own AI visibility?
Three rules follow from this data:
- Ask many questions, not one. Buyers phrase things differently, and each phrasing can produce a different list.
- Repeat each question. Three runs per question is a practical minimum to see whether a mention is stable.
- Check several engines. The gap between engines can be larger than the gap between you and your competitors.
Then report a share: the percentage of answers that name you, and how it compares with competitors. That number is stable enough to track month to month. Our guide to auditing your own AI visibility in 10 minutes shows how to do a small version by hand, and a VisibAI audit runs the full version across several engines with averaged results.
Frequently asked questions
Why does ChatGPT give a different answer every time?
Language models generate text by choosing among likely next words, so the same question can produce different answers. When web search is on, the pages retrieved can also change between runs. In our Lisbon data, the leading club appeared in all 3 runs of the same question only 40% of the time.
Do ChatGPT, Perplexity and Gemini recommend the same businesses?
Sometimes, but not reliably. In Lisbon, one club was named in 39 of 42 Gemini answers and 10 of 42 ChatGPT answers. In New York, the leading club was named in nearly every answer on all three engines. Agreement tends to be high only when a business clearly dominates the sources in its category.
How many times should I ask an AI assistant to check my visibility?
Ask several different buyer-style questions, at least three times each, on at least two or three engines. Then count the share of answers that name you. A single answer is too unstable to judge from.
What is AI share of voice?
AI share of voice is the percentage of AI answers to category questions that name a given business. It turns many varying answers into one comparable figure, so you can track it over time and compare against competitors.
Can a business be consistent across all AI engines?
Yes. In our New York report, Padel Haus was named in 96% of 126 answers across ChatGPT, Perplexity and Gemini. That level of consistency comes from clear, repeated coverage across many independent sources.
Meet us at Web Summit next week
We will be at Web Summit in Lisbon from November 9 to 12 as part of the ALPHA startup programme, running live AI visibility audits. If you want to see how consistently AI names your business, run a free audit before then and bring the result, or read why we are going in our Web Summit announcement.