Can ChatGPT Read Game Outsourcing Studio Websites? An AI-Readiness Census of 520 Studio Sites (September 2026)
When ChatGPT names a game outsourcing studio, it mostly takes what it says from the studio's own website. So we asked a narrower question than "does ChatGPT recommend you": can it read your site, and are the facts it repeats about studios actually written there?
On 25 September 2026 we ran our free AI-readiness checker over the website of every outsourcing studio listed in our directory that day: 520 sites. This post reports what it found, in aggregate. No studio is named, and no single studio's result is published. Every figure is fixed to the run date.
The short version
- Access is not the problem. Robots.txt blocked ChatGPT's search crawler on 1 of 516 sites that answered. Three (0.6%) blocked any of the four AI search crawlers we test.
- Firewalls and JavaScript are a small problem. 22 of 516 sites (4.3%) refused our homepage request with a 403 or 429. Of the 491 homepages we could read, 51 (10.4%) had fewer than 50 words of text in the HTML, so a crawler that doesn't run JavaScript sees an almost empty page.
- The text is the problem. Of those 491 sites, 389 (79%) named at least one game service somewhere, but only 201 (41%) had a services page naming two or more. 367 (75%) stated a city or country, 216 (44%) named an engine or production tool in text, 327 (67%) named clients, games or shipped counts, and 242 (49%) stated a team size or founding year.
- Few have the whole set. 89 of the 491 (18%) stated all five facts. The median site passed 6 of our 8 checks.
How we ran it
- Who: every studio our directory listed on 25 September 2026 with a website: active, listed, not deleted, with a logo. That was 520 of the 521 listed studios.
- What we read: the homepage, robots.txt and up to four pages linked from the homepage (services, about, work, contact), as raw HTML. We didn't run JavaScript, because the AI search crawlers don't either. The requests identified themselves as
GameDevOutsourcingAIReadiness/1.0and asked for English. - What we checked: the eight website checks the free checker shows you (it adds a ninth about your listing here, which we left out). Robots.txt for six crawlers; a homepage that answers 200 without a noindex; at least 50 words of text in the homepage HTML; a services page naming concrete services; a city or country; an engine or tool; credits; and a team size or founding year.
- No AI model was called. This measures what is on the sites, not what ChatGPT says about them.
- Denominators: robots figures are out of the 516 sites that answered (a site with no robots.txt allows everything). Content figures are out of the 491 sites whose homepage answered 200 with HTML.
- Aggregates only: country and service rows appear only where at least 10 sites were checked, and a percentage only where its own base is 10 or more.
Do studio websites block AI crawlers?
Almost none do.
| Robots.txt blocks… | Sites | Share of 516 |
|---|---|---|
| OAI-SearchBot (ChatGPT search) | 1 | 0.2% |
| ChatGPT-User (ChatGPT fetching a page a user asked about) | 1 | 0.2% |
| PerplexityBot | 3 | 0.6% |
| Claude-SearchBot | 1 | 0.2% |
| Any of those four | 3 | 0.6% |
| Googlebot or Bingbot | 0 | 0% |
| A training crawler (GPTBot, ClaudeBot, Google-Extended or CCBot) | 14 | 2.7% |
443 of the 516 sites had a robots.txt with rules and 73 had none, which crawlers read as "everything allowed".
The 14 sites that block training crawlers are making a choice about model training, and that choice doesn't remove them from AI search: the search crawlers are separate tokens. 11 of those 14 still let all four AI search crawlers in.
What it means for you: if your robots.txt was written by your CMS or a template, it almost certainly isn't blocking ChatGPT. Check it once with the free checker and move on to the text.
Firewalls, noindex and JavaScript-only homepages
Three smaller access problems showed up.
- Firewalls. 22 of the 516 sites that answered (4.3%) refused our homepage request: 19 with a 403 and 3 with a 429. That is usually a bot rule at the CDN or host. Our checker isn't an AI crawler, so we can't tell from our own request which crawlers those rules let through, but a rule that refuses unknown bots often refuses AI crawlers too.
- Noindex. 3 of the 491 readable homepages (0.6%) told search engines to leave them out.
- Text drawn only by JavaScript. 51 of the 491 (10.4%) had fewer than 50 words of visible text in the raw HTML, and another 42 (8.6%) had between 50 and 149. A crawler that doesn't run JavaScript reads those pages as empty or nearly so.
What it means for you: if your site is built as a single-page app, or with a page builder that loads text by script, turn on server rendering or pre-rendering for at least the homepage and the services page. If your host has "block AI bots" or "bot fight" settings, let the verified search crawlers through.
Do studio websites have a services page?
This was the largest gap, and the one with the clearest evidence behind it.
In our ChatGPT test of 24 September 2026 (20 buyer questions, each asked three times with web search on), 68% of the lines naming a studio cited that studio's own website. Of those 255 own-site citations, 152 were a services page and 103 a homepage. Across the kinds of question in that test, ChatGPT stated what a named studio does in 57% to 77% of lines. Answers change between runs, so read these as one test, not a rule.
Against that, on 25 September 2026:
- 389 of 491 sites (79%) named at least one recognisable game service somewhere on the pages we read;
- 260 (53%) linked a page we could identify as a services page from the homepage;
- 201 (41%) had a services page that named two or more concrete services.
The rest either put their services in a homepage section, listed them only as images or icons, or described what they do in general words ("we bring games to life") that name no service a buyer would type.
What it means for you: give services their own page, or one page per service, and name each one the way a buyer searches: "2D character art", "Unity co-development", "console porting", "game QA testing", with a sentence on each. This is the cheapest fix on the list and the one closest to how ChatGPT cited studios.
Location, engines, credits and team size
| Fact on the site | Sites stating it | Share of 491 | How often ChatGPT repeated it (24 Sep 2026 test) |
|---|---|---|---|
| City or country | 367 | 75% | 85% of lines on country questions (three runs each), 92% on smaller-country questions (one run each) |
| Engine or production tool, in text | 216 | 44% | 3% to 32% of lines, depending on the question |
| Clients, games or shipped counts | 327 | 67% | 17% of lines on broad questions, 18% on country questions |
| Team size or founding year | 242 | 49% | size in words in 18% to 37% of lines; a number or a year in 4% or fewer |
| All five (including a services page naming services) | 89 | 18% |
Location was the fact ChatGPT leaned on hardest when a buyer asked about a country, and a quarter of sites never state it in text. Engines were missing from more than half, often because they appear only as logos, which a crawler can't read.
Team size and founding year are the weakest link to the answers: ChatGPT rarely repeated a headcount or a year. It did describe size in words ("boutique", "large") in about a third of lines on broad questions, and buyers ask for both, so one sentence covering them is still worth writing.
What it means for you: one sentence carries most of this. "Founded in 2016, we are a team of 35 artists and developers in Sofia, Bulgaria, working in Unity and Unreal Engine 5." Put it on the homepage, and name the games and clients under your portfolio images as text.
By country
Countries with at least 10 listed studio sites on 25 September 2026. Brazil had 10 sites but only 9 readable homepages, so its content figures are left out.
| Country | Sites | Readable homepages | Services page naming services | City or country stated | Engine named | Text only drawn by JavaScript | All five facts |
|---|---|---|---|---|---|---|---|
| United States | 75 | 70 | 39% | 70% | 46% | 5.7% | 8.6% |
| United Kingdom | 41 | 38 | 29% | 84% | 34% | 13% | 21% |
| India | 31 | 30 | 57% | 77% | 57% | 10% | 33% |
| Ukraine | 30 | 26 | 77% | 81% | 81% | 3.8% | 65% |
| Poland | 22 | 21 | 67% | 81% | 62% | 0% | 19% |
| Japan | 18 | 17 | 12% | 29% | 12% | 35% | 0% |
| Canada | 17 | 16 | 25% | 94% | 44% | 13% | 19% |
| France | 13 | 12 | 33% | 83% | 42% | 0% | 8.3% |
| Vietnam | 13 | 12 | 58% | 92% | 67% | 8.3% | 33% |
| Malaysia | 12 | 12 | 33% | 75% | 33% | 17% | 25% |
| Germany | 11 | 11 | 9.1% | 91% | 55% | 9.1% | 9.1% |
| China | 10 | 10 | 40% | 60% | 30% | 30% | 30% |
| Philippines | 10 | 10 | 50% | 70% | 70% | 20% | 0% |
Ukraine's studio sites were the most complete: 17 of 26 readable homepages (65%) stated all five facts, against 6 of 70 (8.6%) in the United States. Canadian and German sites almost always said where they were, and rarely had a services page naming services.
Japan's row needs a caveat. The checker reads English words, and most of these sites are likely written mainly in Japanese, so the text checks will miss facts stated in Japanese. Read it as what an English-language crawl finds, not as a verdict on the sites. The same applies, less strongly, to any site whose main language isn't English.
By main service
Each studio is counted once, under its main listed service. Crafts come first (a studio listing 2D art and full production counts as 2D art), so the generic categories hold the studios that list nothing more specific. Services with at least 10 sites:
| Main service | Sites | Readable homepages | Services page naming services | City or country stated | Engine named | Credits in text | All five facts |
|---|---|---|---|---|---|---|---|
| 2D Art | 202 | 191 | 53% | 80% | 52% | 72% | 31% |
| 3D Art | 133 | 128 | 25% | 70% | 39% | 52% | 12% |
| Audio & Sound Design | 45 | 44 | 43% | 75% | 46% | 77% | 16% |
| QA Testing | 39 | 36 | 64% | 75% | 44% | 86% | 11% |
| Game Design | 22 | 21 | 19% | 67% | 38% | 52% | 0% |
| Marketing | 17 | 15 | 6.7% | 87% | 6.7% | 100% | 0% |
| Animation | 14 | 13 | 54% | 85% | 46% | 85% | 7.7% |
| Porting | 13 | 13 | 39% | 62% | 46% | 54% | 0% |
3D art studios trailed 2D art studios on almost every text check, and 20 of 128 readable 3D art homepages (16%) had their text drawn only by JavaScript, the highest share of any service with 10 or more sites. Marketing's low services score is partly ours: the checker looks for game production services, and trailers and store assets are only partly on its list.
What this census doesn't tell you
- Whether ChatGPT will name you. Passing every check puts the facts where a crawler can find them. It doesn't make any assistant name a studio, and nothing on this site can.
- What a real AI crawler sees behind a firewall. Our checker sends its own user agent. A site that refused us may let OAI-SearchBot through, or not.
- Anything JavaScript adds. We read raw HTML on purpose. A site that renders everything in the browser may look complete to you and empty to us.
- llms.txt or schema markup. The checker doesn't look for either. The published tests we know of, both by Ahrefs, found that 97% of llms.txt files got no requests at all, and that pages which added schema markup gained no AI citations for it. We'd rather point you at the services page.
- Non-English sites fairly. See the Japan caveat above.
See your own site's result
- Anyone: run the free AI-readiness checker on your site. It takes a few seconds and shows the same eight website checks, with the fix for each.
- Listed studios: if you manage your listing, your dashboard shows your site's result from this census next to the share of studios in your country or service that passed each check, and lets you re-check after you fix something. It is private to you. If your studio is listed but you haven't claimed it, claim it free.
- What ChatGPT said: /ai-visibility explains what we learned from asking ChatGPT buyer questions on 24 September 2026, and /ai-recommendations shows which studios it named.
If you want this tracked, the AI Visibility plan asks 10 buyer questions about studios like yours every month, each 3 times with web search on, and shows whether ChatGPT named you, who it named instead, the sources it read and a fix list for your website and listing. What it won't do: make ChatGPT recommend you. Paying doesn't change what ChatGPT says.
We plan to rerun this census as the directory grows and publish each edition with its own date.
Sources: this census, run on 25 September 2026 over 520 listed studio websites and published in aggregate only; our ChatGPT test of 24 September 2026 (60 answers with web search on, plus 28 one-run answers on niche and smaller-country questions), written up on /ai-recommendations; Ahrefs' published studies of llms.txt requests and of schema markup and AI citations.