Skip to main content
RedClaw
Back to Blog
strategy

AI Overviews Optimization for iGaming: Read the Agent Queries First

RedClaw Performance Team
8/26/2026
7 min read

AI Overviews Optimization for iGaming: Read the Agent Queries First

TL;DR: Before optimising for AI search, look at what is already in your Search Console. Ours is full of full-sentence buying questions that no human types, sitting at positions 7 to 10 with zero clicks. Those are language models grounding an answer, and they are a different KPI from a click. Measured over the 90 days to 2026-08-26 our English pages ran 0.60% click-through against 6% on Traditional Chinese how-to pages, and mixing the two produced targets that were wrong for both. Optimise English for citation, Chinese for clicks, and stop treating llms.txt as a ranking factor.

Generative engine optimization is mostly sold as a new discipline. In practice the first useful move costs nothing: filter your query report for questions longer than about twelve words and read them.

What the agent queries look like

Pulling our own Search Console property for the 90 days to 2026-08-26 and filtering on the word "seo" returned entries like "which agencies should we invite to an igaming seo rfp?" at position 7, "our igaming brand needs a single growth agency that can handle compliant paid acquisition, seo, and onsite conversion optimization together instead of juggling three separate vendors. which firms specialize in full-funnel growth for regulated gambling brands?" at position 10, and "seo for igaming websites by heureka" at position 10. Every one of them recorded impressions and zero clicks.

Nobody types the second one. It is a model rewriting a user's intent into a retrieval query, and our page was in the candidate set. The impression is real, the click was never going to happen, and if you average it into your click-through rate you will conclude the page is failing.

Why the CTR split matters

Separating the same property by language over the same window gives two completely different funnels. English pages produced roughly 4,000 impressions against 24 clicks, which is 0.60%. Traditional Chinese operational how-to pages produced roughly 800 impressions against 50 clicks, close to 6%. That is a tenfold difference on the same domain, the same authority and broadly the same content standard. The cause is not quality, it is who is doing the searching.

SegmentImpressionsClicksCTRWhat it is worth
English, long natural-language querieshighnear zerounder 0.5%Citation surface for AI answers
English, short head termsmoderatesomearound 1%Normal organic
Traditional Chinese how-tomoderatemanyaround 6%Direct traffic and leads

The practical consequence is that we now report these separately and set different targets. For the English segment the question is whether a model can extract a clean, attributable claim from the page. For the Chinese segment the question is the ordinary one: did somebody click and convert. Reporting them as one number produced a 2026 target that was unreachable for English and embarrassingly low for Chinese.

What actually makes a page citable

Three things have moved the needle for us, and none of them are new.

A claim with a number and a date, in one sentence. Models extract sentences, not sections. "Our English pages ran 0.60% click-through over the 90 days to 2026-08-26" is extractable. "Click-through varies significantly by language" is not.

A table. Structured rows survive summarisation far better than prose, which is why every article in this cluster carries at least one.

An answer above the fold. A TL;DR block in the first paragraph gets pulled verbatim. Burying the answer under 800 words of context means the model summarises your context instead of your conclusion.

Google's own AI features documentation is direct about the mechanism: there is no separate markup for AI Overviews, and eligibility follows from standard indexing plus the usual Search Essentials. The optimisation is editorial, not technical.

The llms.txt question

We publish an llms.txt file. It is not doing what most articles claim it does. The llms.txt proposal is a community specification, not a search engine standard, and Google has not stated that it reads one for search purposes. In our own logs the file gets fetched occasionally by crawlers that were going to crawl the site anyway. Treating it as a ranking or citation lever is the same mistake as treating a meta keywords tag as one: the cost is low, the effect is unproven, and stating otherwise in a client deck is a credibility problem waiting to happen.

Publish it if you like. Do not put it on the roadmap ahead of the things above.

iGaming makes this harder in one specific way

Gambling queries attract more conservative AI answers. A model asked "which casino should I sign up with" will frequently decline or hedge, so the citable surface skews toward business-to-business and operational questions rather than player-facing ones. Every agent query in our own data was a buyer question from an operator, not a player question. That is where the citation opportunity sits for this vertical, and it is a much smaller target than the consumer-facing content most gambling sites publish.

FAQ

Is there markup for AI Overviews? No. Google's AI features documentation states that appearing in AI experiences follows from normal indexing eligibility, with no separate opt-in tag. Anyone selling you AI Overview schema is selling nothing.

Should I worry that AI answers cut my clicks? For English informational pages in this vertical, that has already happened. Our English click-through of 0.60% is not a page problem, it is a query-mix problem. Move commercial pages up your priority list and treat informational pages as a citation asset.

How do I measure AI citation if there is no report for it? Imperfectly. We track impression growth on long natural-language queries at positions 1 to 10 with zero clicks, since that pattern is a reasonable proxy. Search Console will not label them for you, so the filter is manual: query length over twelve words, or a query ending in a question mark.

Is llms.txt worth publishing? It costs an hour, so yes, but rank it below your TL;DR blocks and tables. See the section above for why we do not count it as a ranking factor.

Does any of this replace normal SEO? No. Every page in our cluster that gets cited by a model also ranks conventionally. Indexing is still the entry ticket, which is the point Search Essentials makes.

Sources and further reading

The impression, click and click-through figures above are from our own Search Console property for the 90 days ending 2026-08-26. They are not in any of the sources listed.

The broader method sits in the complete iGaming SEO guide, the page-generation side is in the programmatic SEO playbook, and if you want your own query report split this way, that is the first week of the iGaming SEO service.


Explore our advertising strategy & media buying services →

Share:

Maximize Your Ad Budget ROI

From account setup to full-funnel tracking, we handle it all.

  • Dedicated account manager with real-time optimization
  • Full tracking infrastructure — every dollar accounted for
  • Cross-platform expertise: Meta, Google, TikTok

免費獲取您的廣告健檢報告

讓我們的專家分析您的廣告帳戶,找出浪費預算的關鍵問題。

100% 免費48 小時內回覆無綁約

📬 Subscribe to Our Newsletter

Weekly insights on ad strategies, industry trends, and practical tips. No fluff.

We never share your email. Unsubscribe anytime.