In 2024, most operators were right to watch and wait. Traffic from ChatGPT and Gemini was interesting, but conversion rates were low.
By 2025, we’d learnt to trust AI recommendations.
Across more than 100 million site visits we’ve analysed, AI-referred visitors now convert at rates that outperform paid channels. Ofcom reported ChatGPT use in the UK increased fivefold last year. AI can’t continue growing at that pace. This is now a mature market.
When ChatGPT recommends a brand, it carries real weight. Closer to a recommendation from a knowledgeable friend than simply another voice competing for attention online.
There’s still a problem with measuring it. Ask ChatGPT which online casino to join. You might get Bet365, Coral, and SkyBet. Ask again, and you might get Betfair, Ladbrokes, and Paddy Power.
We like certainty, don’t we? Paid media gives you a cost per acquisition to two decimal places. SEO gives you rankings, traffic, click-through rates, and return on investment – clean numbers on clear dashboards. AI doesn’t work like that.
And a widely cited study from SparkToro appears to confirm the worst fear: they tested 2,961 AI prompts and found a less than 1-in-100 chance of getting the same recommendation list twice. If you take that at face value, AI visibility can’t be measured. And if it can’t be measured, it can’t be optimised.
We’ve spent six months tracking how AI perceives iGaming and lottery brands across four major LLMs. The lists do vary. Our own data can shift by a couple of percentage points month to month. But the brands that appear, and how often they appear, follow consistent patterns. The data isn’t as clean as the paid media data. It isn’t as established as SEO. But if half your potential players are using AI as much as they’re using Google, you need to be visible there. And you can measure it. You just have to do it the right way.
SparkToro’s research – and what it shows
SparkToro’s study is the most cited piece of research on this topic. It deserves a proper look.
I should say upfront: I’m a Rand Fishkin fan. I learned my craft watching his Whiteboard Friday sessions at Moz. So, when SparkToro published research on AI brand recommendations, I paid attention.
His team ran 12 prompts across ChatGPT, Claude, and Google AI. Each prompt was repeated 60 to 100 times, giving them 2,961 total responses. They tested categories like chef’s knives and headphones, asking AI to recommend brands.
Their headline finding: less than a 1% chance of getting identical recommendation lists from the same prompt.
That grabbed the industry’s attention. But it’s worth reading past the headline.
To his credit, Rand says exactly this in the piece. He acknowledges that “visibility % across dozens to hundreds of prompts run multiple times is a reasonable metric.” The frequent appearance of top brands isn’t a footnote in his research. It’s a finding.
So, there’s more agreement here than the headline suggests. Rand is essentially saying – ‘be careful how you measure this. Don’t trust single queries. Don’t trust ranking positions. And be sceptical of any vendor who can’t explain their methodology.’
We’d say the same thing. Where we’d go further is on what the data actually means for operators.
AI typically recommends four or five brands per query. Google gives you 10 blue links plus ads – a page of choice.
We’re moving from a world of choice to a world of recommendations. In a world of 10 links, being on page one means you’re one of many options. In a world of four recommendations, you’re either in the consideration set or you’re not.
For operators and lottery brands, this means the market is consolidating around AI visibility. Brands that AI consistently recommends will capture a larger share of the next generation of players. And that, as we’ll see, is something you can measure and influence.
The polling analogy
There’s a useful comparison with opinion polling.
No pollster expects every respondent to give the same answer. That would be strange. The value is across all responses. Ask 1,000 people which party they’ll vote for, and you won’t get 1,000 identical answers. But you’ll see Labour at 38% and the Conservatives at 27%. That difference is real, measurable, and worth acting on.
AI visibility works the same way.
Ask ChatGPT 100 times which online casino to join. You won’t get 100 identical lists. But you’ll see Bet365 mentioned 70 times and a competitor mentioned 30 times. That gap tells you something important about how AI perceives those brands. And it’s consistent enough to track month to month.
Pollsters don’t claim perfect precision. They publish margins of error. They’re transparent about sample sizes. And nobody dismisses polling as unreliable because two polls give slightly different numbers. The value is in the pattern, not from any single result.
The same standard should apply here. AI visibility data has a margin of error. Our own tracking can shift by a couple of percentage points between months. That’s normal. It doesn’t make the data unreliable. It makes it data.
The question is whether the patterns are consistent enough to base decisions on. The answer is – they are.
The technical evidence for consistency
The variability is real. But it’s mostly on the surface.
Gumshoe, a research firm specialising in AI visibility measurement, ran repeated experiments with identical prompts across LLMs. They found that while the wording changed every time, the underlying meaning stayed stable. Their similarity scores – a measure of how closely two responses match in meaning – frequently came out between 0.7 and 0.9 out of 1.0.
In plain terms: AI says the same thing in different words. It recommends the same brands, describes them in similar ways, and reaches similar conclusions. It just phrases it differently each time.
Three factors make the data more reliable than the headlines suggest.
The category effect.
Broad, open-ended questions scatter widely. “Recommend a good book” could go anywhere. But narrow, specific questions converge quickly. “Which online casinos are licensed in the UK?” has only so many possible answers. So does “which US states offer iLottery?” or “what’s the best betting app for football?” In regulated industries like iGaming and lottery, licensing limits the field. There are only so many operators AI can recommend. Fewer possible answers mean the data settles faster.
The sample size effect
One query tells you nothing. Ten queries start to show a pattern. Gumshoe’s research suggests you need to run the same question around 100 times before the result is accurate to within 10 percentage points. Around 400 times to get within 5 percentage points. In practice: dozens of repetitions per prompt, not one or two. A well-run tracking programme can reach those numbers comfortably.
The frequency effect
Across Rand’s data and ours, the same pattern appears: a small group of brands dominates, a middle tier appears intermittently, and the rest barely register. That distribution is consistent month after month. The individual lists vary. The distribution doesn’t.
What drives the consistency
So why do some brands appear in 70% of AI responses while others barely register?
Kevin Indig analysed 1.2 million ChatGPT responses to find out. Around 30 domains control roughly two-thirds of AI citations across any given topic. The top 10 domains alone capture 46% of citations. A small group of brands dominates. The patterns are clear.
Those 30 domains are not operator websites, of course.
They’re Reddit threads where players discuss which apps are worth using. They’re Trustpilot pages with hundreds of real reviews. They’re Wikipedia entries, gambling comparison sites, editorial coverage in national newspapers, and industry publications. They’re the places where the internet talks about your brand – not the places where you talk about yourself.
That’s why the narrative you build across the web matters as much as anything on your own site. It’s something Receptional’s specialist digital content team works on with operators directly – shaping and building the kind of presence across independent sources that AI can draw on.
This changes the strategic picture. Your own website is one input. What the rest of the internet says about you is where the competition is actually happening.
The signals that drove Google visibility for the last 20 years – backlinks, keyword optimisation, technical SEO – don’t reliably translate into AI recommendations, despite still being crucial foundations. A brand can hold the number one position on Google for its primary keyword and still be largely invisible to ChatGPT. We’ve seen this in our own data.
What matters is whether you have a consistent, credible story across the sources AI reads. Reddit discussions, review platforms, editorial mentions, Wikipedia, comparison sites. The brands that appear consistently in AI recommendations have built that presence – deliberately or otherwise. The brands that don’t are waiting to be told their visibility gap exists.
Two AI channels, not one
Google AI Overviews now appear in around one in three searches. Your customers aren’t choosing to engage with AI recommendations. They’re seeing them by default, every time they search.
That makes Google AI visibility essential. If you rank well and Google’s AI surfaces your brand in Overviews, that’s the largest AI visibility channel available. And because Google AI Overviews draw heavily from Google’s own index, strong organic rankings do carry weight here.
ChatGPT is a different channel, with different signals. OpenAI runs its own crawler, independent of Google. The content it surfaces and the brands it recommends work differently. And it’s not a niche audience. Research from Bain, published in September 2025, found that under-45s use LLMs for research as much as they use Google. For iGaming and lottery operators trying to reach younger players, ChatGPT, Claude, and Perplexity are already where those players are looking.
Your customers are using both. The question is whether you know how you appear in each, and whether you’re measuring them separately.
Google AI Overviews and ChatGPT recommendations are related but distinct signals. What drives visibility in one doesn’t guarantee visibility in the other. You need to track both, with enough data to see what’s real.
What we’ve seen
We’ve been tracking AI brand visibility in the UK iGaming and US lottery markets since November 2025. We run the same prompts repeatedly across ChatGPT, Claude, Gemini, and Perplexity every month. By the time this article publishes, that’s around six months of data.
A few things stand out.
The first is concentration. In a UK market of around 50 licensed operators, six brands consistently appear in the top tier across almost every relevant prompt. The gap is hard to miss. The top six are measurably, consistently better represented than the brands below them.
The second is how to read the variation. Month-to-month figures do shift, and those shifts do affect positions within our tracking tables. So, rather than focusing on whether a brand ranks second or third each month, we group brands into bands. The top tier stays the top tier. A brand moving from second to fourth within that band is less meaningful than a brand moving from the top tier into the middle tier. The bands are stable even when individual positions aren’t.
The third is that the four platforms we track don’t always agree. ChatGPT, Claude, Gemini, and Perplexity each surface different brands in response to the same prompts. That variation is informative.
How to measure AI visibility correctly
The starting point is asking the right questions. There are two types.
Brand questions – “What do you think of Bet365?” – tell you how AI perceives a specific operator: sentiment, accuracy, how it positions you against competitors. Useful for understanding your own brand and theirs.
Non-brand questions are where the commercial picture emerges. “What’s the best football betting app?”, “Which online casino has the best welcome offer?”, “What’s the best iLottery app in Michigan?” These are the questions potential customers are asking the moment their decision-making happens. The brands that appear in those responses are the ones capturing the next generation of players.
In SEO terms, it’s the difference between brand and non-brand keywords. You need both. Brand queries tell you how you’re perceived. Non-brand queries tell you whether you’re in the consideration set when it matters.
Ask each question many times, not once
A single prompt tells you nothing reliable. Run the same question dozens of times across each platform before drawing any conclusions. The pattern emerges from repetition.
Track across platforms
ChatGPT, Claude, Gemini, and Perplexity behave differently. A brand that appears consistently in ChatGPT may appear far less often in Claude. If you only track one platform, you’re seeing a partial picture.
Use bands, not rankings
Month-to-month figures shift, and those shifts affect positions within tracking tables. What matters is whether a brand is in the top tier, the middle tier, or the bottom tier – and whether that band position changes over time. A brand moving from second to fourth within the top tier is less significant than a brand dropping from the top tier to the middle tier.
Track longitudinally
A single snapshot tells you where you are. Six months of data tells you whether you’re moving. The brands gaining ground in AI recommendations today are building a position that will be hard to dislodge.
Be honest about the margin of error
The data has natural variation. Presenting AI visibility figures to two decimal places implies a precision that isn’t there. Show ranges, acknowledge variation, and focus on the direction of travel rather than absolute positions.
The brands building AI visibility today are creating a compounding position. Every month they appear consistently in AI recommendations is a month their competitors don’t. That gap widens. And the channel is only growing.
We’ve been tracking UK iGaming AI visibility since November 2025 — six months of data across ChatGPT, Claude, Gemini, and Perplexity. Named brands. Real numbers.
This April and May, we’re sharing the detailed findings for the first time. If you want to be among the first to understand this data and know where your brand stands, this is a session you need to attend.
Register for the AI Visibility Webinar to be first to see the data
Receptional’s AI Opportunity Analysis shows you exactly where you stand.
How often are you recommended versus your competitors? Which content and technical factors are holding you back? What are the specific changes that move your brand from being occasionally mentioned to consistently recommended?
The assessment will highlight even the smallest tweaks you can make to get you in the spotlight.
