GTA 6 comes out on November 19, 2026. Ask three machines where that came from and you get three different answers.

The registered question, written before any data was pulled

Given one identical query string, how much do the domains cited by an answer engine overlap with the domains Google ranks organically?

Four outcomes were registered with probabilities before retrieval: the layers are the same channel (15%), related but distinct (25%), separate distribution problems (45%), or too unstable to compare (15%). The falsifier was named in advance: the separate-problems reading dies if two or more of the nine reference domains rank organically for the same string.

Method: DataForSEO AI Optimization live endpoint · ChatGPT gpt-4.1-mini-2025-04-14 and Perplexity sonar, web search enabled, three calls each · Google organic via DataForSEO, Houston, Texas, United States, depth 10 · all eight rows inside 29 seconds · metered cost $0.11 · run 3063e451

The result: zero domains in common

The registered falsifier did not fire. Zero of the nine reference domains ranked organically, against a threshold of two.

The last row is the one worth sitting with. Google was less consistent with itself across two phrasings of one question than Perplexity was across three repeats of the same string. Rewording the question changed five of six results on the same engine, in the same city, in the same minute.

Every domain each machine cited

Published in full because a summary of a citation list is not a citation list. Sites flagged dedicated exist to answer questions about this one game and publish little else.

One domain of seven survived all three calls. On call three, all four citations were purpose-built GTA wiki sites: gta6-wiki.org, gta6wiki.io, gta6game.wiki and gtacorner.com. Mainstream games press appeared in two calls and vanished from the third. Rockstar's own domain was never cited at all.

Ten of ten, in the same order, three times. This is the fourth separate measurement in which Perplexity has returned a fully stable citation set and ChatGPT has not. It is a measurement of these questions on these days, not a league table, and we would not publish it as one.

Every domain, every instrument, in one table

The twenty-seven distinct domains returned by four readings on one afternoon. Filter it. The control worth pressing is cited by more than one instrument, because of what it returns.

Showing all 27 domains.

No domain reaches two. Twenty-seven domains, four readings of one question inside half an hour, and the Instruments column never leaves 1. The only repeat anywhere is redbull.com, which Google returned for both of its own phrasings.

What the machines reach for instead: pages built for one question

Six of the 17 domains the two answer engines cited are sites that exist to answer questions about this one game. Both of Perplexity's top two positions are held by them. All four of ChatGPT's third-call citations are.

Of ChatGPT's seven domains, exactly one — pcgamer.com — is mainstream press. IGN, GameSpot, Eurogamer, Rockstar's own site and Take-Two's investor pages were registered as the expected match list before the pull. Rockstar was cited zero times out of three by ChatGPT on a question about Rockstar's own release date.

What people actually search for

Thirteen of the top thirteen keywords are release-date variants, each between 823,000 and 1,500,000 searches a month. Pre-order, leaks, map, price and delay follow. There is not one analysis-intent query in the top fifty.

Volumes are Google's advertising bands rather than counts, and keyword suggestion data is an advertising product that skews commercial. Read the ordering, not the magnitudes.

What the publisher spends

The step change is FY2023, a 4.08-times jump, and intensity has been flat to falling since. Take-Two closed the Zynga acquisition in May 2022, inside FY2023, and Zynga's model is paid user acquisition. The most recent year is lower than three years ago.

For scale: the median advertising intensity across 137 prepackaged-software filers, 61 pharmaceutical filers and 40 business-services filers for CY2024 was 2.01%, 3.22% and 3.30%. Take-Two sits at roughly six to nine times a cross-industry typical value — but the spread inside every category measured was so wide that the category itself predicts almost nothing, which is a separate CULT+MATH study.

The same query, three times in one hour, three different answers

Before reading any comment section we had to pick one, and picking it turned out to be the finding. Six YouTube searches were run three times inside one hour and the top ten by comment count recorded each time.

The instability arrives in discontinuities, not as steady churn. The reaction corpus replaced nine of ten members between two consecutive runs and then held perfectly. The roleplay corpus held perfectly and then lost two. Only Call of Duty survived all three intact.

A volatility rate computed from a single interval would have been wrong for both of the first two. So this table reports pairs rather than a score, and every corpus read on this page has its video IDs frozen in the specification rather than re-picked at run time.

What people say, before a game ships and after

The control row is the finding. gameranx appears twice. On an unreleased game its comment section returned the host's name, a tribute to a late actor, and a horse. On a game that had shipped two days earlier, the same channel returned campaign in 20% of comments, review, mission, and unprompted price talk — one commenter volunteering what they would pay for the campaign sold separately.

Because the channel is held constant, the difference is not the channel. Pre-release comment sections carried no commercial vocabulary at all. For anyone hoping to read purchase intent out of pre-launch conversation, that is the answer, and it is a negative one.

The audience answered the economic claim with their own working week

A television station said the launch could drain a billion dollars of US productivity. We read the comment section to see whether an economic framing draws an economically-minded audience. It draws something better and stranger than that.

Three of three corpora covering this game economically carry it. Zero of six everything else — including two corpora about industry economics, three gameplay analyses, and a control covering a game that had already shipped.

What people wrote, quoted at the 120-character cap the listening standard sets, with no author names:

• “I quit my job 2 years ago for this day”
• “A day off work. Rookie numbers. I took a week off.”
• “I’m 44 & I’ve already put in my vacation time over 3 months ago”
• “Put in to take the day off for this 3 months ago.”
• “I’ve already notified the boss I’m going to be sick on that day”

They did not debate the billion-dollar figure. They are the billion-dollar figure, reporting themselves, unprompted, with dates and notice periods attached.

The framing that is already working, for somebody else

A Detroit television affiliate framing a game launch as an economic event drew 1,742 comments. Three of the ten domains Perplexity cited about this game are financial press — and on the open-ended version of the question, CNBC, Yahoo Finance and Benzinga were among them.

The FOX 2 figure moved from 1,683 to 1,742 comments across four readings in one evening, which is why it is given with its reading dates rather than as a standing number.

What this study cannot tell you

• One city, one day, one query string. Houston, October 24, 2026. No claim about other markets and no trend claim.
• Organic results are a minority of the Google page. Depth 10 returned six and seven organic positions. The AI overview, the video pack and the paid units are not counted and are most of what a person sees.
• Platform-default budget-tier models. Both engines are read at the tier the API serves, which is not the model a consumer is served.
• Nothing here observes a purchase. Search volume is interest, comment share is attribution, model citation is belief. Not one row measures a sale, and advertising spend is an input rather than an outcome.
• The Zynga explanation is unread. Named as a hypothesis in the spend section and still unconfirmed against the filing.
• The frame comparison returned no registered verdict. Four candidate outcomes were registered before the pull and the result fell in a gap between two of them, so the test could not report what it found. The design fault is ours and the self-report table is labelled unregistered because of it.
• Three of the five economically-framed corpora concern this one game, 26 days from launch. A general audience with a dated event in front of them is an unusual condition and the self-report pattern may not survive outside it.

How this page changes

This is a standing study, not an article. The measurements above are re-run on a schedule and the page is corrected in place with the date of the correction shown. Corrections are published, never applied quietly, and the retraction in the corpus section is the first entry in that record.

Related reading, free and in full

• All standing studies — the index, with the method each one follows
• The IP Association Map — 89 measurements of whether a brand's own domain is cited when an AI is asked about it
• How much of a games publisher's revenue coming from one franchise is too much?
• Is live service or premium release the better model for a mid-size games publisher?