Google Shows an AI Overview for 74% of Our Queries. For Gambling, It's 1.4%.

We measure whether AI answer engines cite the sites we and our clients operate. Doing that across enough verticals for long enough turns up things nobody was looking for. This is one of them.

Between 23/06 and 08/07/2026 we recorded 2,942 clean Google AI Overview runs across 36 sites. On gambling and betting queries, Google returned an AI Overview 11 times out of 799 — 1.4%. On everything else, it returned one 1,577 times out of 2,143 — 73.6%.

That is not a small difference. It is a different product.

Disclosure, up front. Lead Media operates gambling comparison sites, including verdikt.bet. This finding is commercially inconvenient for us: it says a channel we would like to win is shut. The measurement engine behind this post, GetCited, is our own. We sell nothing here. We are not a neutral party — read the method and judge for yourself.

The number

An "AI Overview shown" run means Google returned an ai_overview block in the SERP. A run with no overview means Google returned a normal results page without one — not an empty overview, not a failed request. Errored runs are excluded and counted separately.

| vertical | sites | AIO runs | overview shown | rate | |---|---:|---:|---:|---:| | trading / finance | 8 | 927 | 835 | 90.1% | | software / consumer tech | 2 | 240 | 205 | 85.4% | | other | 3 | 96 | 61 | 63.5% | | local services | 12 | 880 | 476 | 54.1% | | gambling / betting | 11 | 799 | 11 | 1.4% |

Look at the first row before the last one. Trading and finance is YMYL. It is heavily regulated. In many of the same countries, forex marketing is restricted harder than casino marketing. It returns an AI Overview nine times in ten.

So this is not Google being cautious about money, or about regulated industries, or about advice that could hurt someone. Whatever this is, it is specific to gambling.

It gets sharper

Strip out the one sports-betting property and look only at the online-casino sites: 727 runs, 4 overviews.

All four were legal or regulatory questions — "Which operators have lost their Coljuegos licence in Colombia, and why?" and "What is Polymarket, and is it legal to use in Mexico?"

Not one commercial casino query returned an AI Overview. Not "best online casinos". Not "best casino bonuses". Not "best slots". Not in Toronto, Madrid, Mexico City, Bogotá, São Paulo, Rome, or the United States. Not in English, not in Spanish. Not on any day in a two-and-a-half week window.

Of the 11 gambling overviews across the whole corpus, 8 were legal or regulatory questions:

| prompt | overviews | type | |---|---:|---| | Is online sports betting legal in New York? | 4 | legal/regulatory | | What are the best sports betting sites in the US? | 3 | commercial (US sports betting) | | ¿Qué operadores han perdido su licencia de Coljuegos en Colombia y por qué? | 2 | legal/regulatory | | ¿Qué es Polymarket y es legal usarlo en México? | 2 | legal/regulatory |

(Prompt text only. We withhold which site each scan belongs to — some are client and prospect properties.)

The exception has a shape. Google will render an overview about gambling law. It will not render one about gambling products.

We spent the day trying to prove ourselves wrong

A 1.4%-versus-73.6% split is exactly the kind of result that turns out to be your own bug. Three rival explanations were live, and two of them were suggested by our own data. We tested each with live API calls before publishing.

Rival 1: the runner was broken during the gambling scans.

Every one of the 97 scans ran on runner 0.2.0 — a single version across the whole window. Better: on 23/06/2026, gambling ran 400 AIO runs at 100% no-overview while non-gambling ran 440 AIO runs at 18.9% no-overview. Same day. Same runner. Same code path. Killed.

Rival 2: the location string.

This one nearly caught us. Our gambling sites were configured with the location Mexico City,Mexico; our forex site with Mexico City,Mexico City,Mexico. Those are not typos — they are two different DataForSEO locations for the same city at different administrative granularity (location_code 20703, state-level CDMX; and 1010043, city-level). In the corpus, the first showed 99.2% no-overview and the second 9.0% — and because each location was used by exactly one topic, location and topic were perfectly confounded. No amount of staring at the existing data could separate them.

So we ran the factorial live, on 10/07/2026:

| topic | location string | language | AI Overview? | |---|---|---|---| | gambling | Mexico City,Mexico | en | no | | gambling | Mexico City,Mexico City,Mexico | en | no | | forex | Mexico City,Mexico | en | yes | | forex | Mexico City,Mexico City,Mexico | en | yes |

Across two genuinely different location codes, the topic split holds identically. The location changes nothing. The topic changes everything. Killed.

(We first assumed Mexico City,Mexico was a malformed string. We checked it against DataForSEO's locations endpoint before writing this paragraph. It isn't malformed — it is the state-level entry for Mexico City. The assumption was wrong and would have made the post wrong.)

Rival 3: the language code.

Our runner hardcoded language_code=en while asking Spanish and Italian prompts — a genuine defect we found while writing this. Maybe Google withholds overviews on a language mismatch. We re-ran both topics with language_code=es at the city-level Mexico City location: gambling no, forex yes. Same split. Killed — for this finding. The defect was real, and we fixed it the same day; it just isn't the cause.

Rival 4: the request quietly failed.

If the gambling SERP requests failed, there would trivially be no overview block. So we inspected the returned SERP item tree across three locales:

| topic | locale | AI Overview? | organic results | SERP healthy | |---|---|---|---|---| | gambling | Toronto, Ontario, Canada | no | 9 | ✓ | | forex | Toronto, Ontario, Canada | yes | 8 | ✓ | | gambling | Madrid, Spain | no | 10 | ✓ | | forex | Madrid, Spain | yes | 9 | ✓ | | gambling | Mexico City, Mexico | no | 8 | ✓ | | forex | Mexico City, Mexico | yes | 9 | ✓ |

Every gambling SERP came back with 8–10 healthy organic results. The request works. Google renders the page. It simply does not put an AI Overview on it. Six cells, six clean splits by topic. Killed.

What we are not saying

We can see what Google does. We cannot see why. We have no visibility into policy, intent, or whether this is a deliberate suppression rule, a quality threshold, an advertiser-safety heuristic, or an emergent property of the model. Anyone telling you which of those it is — including us — is guessing.

We are also not saying AI engines ignore gambling. They don't. ChatGPT, Claude and Perplexity cite gambling sources freely and abundantly; our Alberta experiment counted 1,584 citations across 310 runs on exactly these questions. This finding is about one surface: Google's AI Overviews.

And this is our prompt corpus, not a random sample of the web. Our prompts exist to measure sites we and our clients operate. A different prompt set could give a different number — though it would have to produce 1.4% against 73.6% by accident. We measured through one provider, across 23/06–08/07/2026, and Google changes overview triggering constantly. Lottery, bingo, poker and esports betting are essentially unrepresented.

What it means if you operate in this market

(Interpretation. Reasoned, not measured — held to a lower standard than everything above, and marked as such.)

Google AI Overviews is not a channel for commercial gambling content. It is a closed door. Any budget line item that reads "optimise for AI Overviews" on casino terms is money spent on a surface that does not render. Not "we rank poorly there" — it does not appear.

The door is ajar for law. Every single overview we saw on a casino property was a legal or regulatory question. That is where an overview can still appear — and it is precisely the content an independent publication is built to carry, and precisely the content a comparison page is not.

That is a real structural argument, arrived at accidentally, for splitting informational publishing from commercial pages: the informational surface can win a channel the commercial surface is structurally barred from. We did not set out to test that. The data walked into it.

And the other engines remain wide open. If you are trying to earn AI citations in gambling, ChatGPT, Claude and Perplexity are where that work pays. Google AI Overviews is, for now, not a place where the work can pay at all.

Method, in one paragraph

2,942 clean runs on the aio surface, 23/06–08/07/2026, runner 0.2.0, fetched via DataForSEO's google/organic/live/advanced endpoint with load_async_ai_overview=true (without that flag the overview renders after the response and you undercount). A run counts as "overview shown" when the returned SERP contains an ai_overview block. Errored runs excluded. From 3,127 clean AIO runs we excluded 185 — internal smoke tests, GetCited's own self-scans, and two sites whose vertical is genuinely ambiguous — and we made those exclusions before looking at their numbers, so we could not pick the exclusion that helped. The confound probes were run live on 10/07/2026 against the same endpoint. Every figure in this post is in the primary data file that ships beside it, and the three defects this experiment found in our own stack — a hardcoded language_code, an invalid Italian location string, and two granularities of Mexico City silently splitting our cohorts — are recorded there too. Two are fixed; the third is documented and deliberately left alone.

Corrections

If any number here is wrong, we will correct it in place, with a dated note, and say what changed. We published a Wikipedia citation table earlier today that was wrong twice over; we corrected it the same day and left the error visible. An imprint whose only asset is trustworthy figures has to correct in public, including against itself.

exp-014 is open. We will re-run this in 30 days. If the blackout lifts, we will say so here.