Notes on how AI assistants answer casino questions
Notebook opened March 2026, last entry 26 August 2026
This site is a working notebook, not a magazine. Every month I put the same set of casino questions to the major AI assistants, record what comes back, and write down what changed. The questions cover recommendations, safety, licensing and brand confusion, the four areas where an assistant's answer can send a player somewhere they should not go.1
The panel is small on purpose. Five assistants, forty fixed prompts, three markets, run in clean sessions with no account history. A fixed panel makes drift visible. When an assistant that refused casino recommendations in April starts naming operators in July, that is a finding, and it goes in the log with the date and the exact wording.
Why this is worth logging
Search results leave a public record. Anyone can screenshot a results page and argue about it later. Assistant answers mostly do not. They are generated per session, they vary with phrasing, and they are gone when the window closes. If nobody writes them down, nobody can say how casino advice from these systems is shifting, and the shift through 2026 has been fast. Assistants that once answered from memory now browse, quote review sites, and check regulator registers before naming an operator.
The log so far
- Entry 15, August 2026AI answers on casino payments. What assistants tell players about deposits, withdrawal times, fees and stuck payouts, from the new payments prompt block.
- Entry 14, August 2026Tracking AI answer visibility. A repeatable method for measuring which brands assistants name, and where the vendor comparisons come from.
- Entry 13, August 2026How AI assistants choose casino sources. Which pages browsing assistants read and cite, and the selection pattern the citation trail shows.
- Entry 12, July 2026AI license checks. What happens when an assistant verifies a gambling license against a regulator register, and where the process breaks.
- Entry 9, June 2026ChatGPT and casino recommendations. Six months of one assistant's answers to the same recommendation prompts.
- Entry 5, April 2026How AI assistants rank casinos. The inputs that decide which operators get named, as far as testing can show.
Method in one paragraph
Each prompt runs three times per assistant per month. I record the named entities, the cited sources, any license claim, and any refusal. Wording is preserved verbatim in the raw log. An observation only appears in a published entry when it held across at least two runs and two phrasings. Single odd answers are noted but never used as evidence of a pattern.2
1. Prompt sets are frozen per quarter. Q3 2026 uses the set frozen on 1 July 2026.
2. Assistants are stochastic. One answer proves nothing, which is the whole reason this notebook repeats itself on a schedule.