Is There a Sandbox in AI Search, and What Evidence Supports It?

What do people mean by a sandbox?

The "sandbox" is an old SEO theory. It holds that search engines deliberately hold new websites back from ranking well for a period, however good their content, until they have built some history. In AI search, the same idea takes the form of new brands or pages being slow to appear in answers even when they answer the question well.

The question matters because it changes what you do. If there is a fixed waiting period, effort in the first months is wasted. If the delay comes from specific mechanics, you can work on each one.

What is the evidence for Google?

• The 2024 documentation leak. Leaked Google API documentation, widely analysed in May 2024, listed a hostAge attribute. It was described as the earliest first-seen date of all pages on a host, used "to sandbox fresh spam in serving time". Read literally, that describes holding back new content judged to be spam, not all new sites.
• Google's response. Google said at the time that people should avoid inaccurate assumptions based on out-of-context, outdated or incomplete information. The documentation does not show how, or whether, each attribute is used.
• Google's statements. Google's John Mueller has said more than once that there is no sandbox, including for site moves.

Together, these suggest Google has mechanisms for treating very new content cautiously when it looks like spam. They do not show a blanket delay for legitimate new sites. Most of the lag new sites see can be explained by discovery, crawling and a lack of links and history.

How does this carry into AI search?

AI answers depend on several systems, each with its own lag:

• Google's AI Overviews and AI Mode use Google's index. Anything that slows a new site in Google slows it there.
• ChatGPT's search uses OpenAI's own crawler, OAI-SearchBot, and third-party search providers. Which providers matter is unclear: OpenAI called Bing "an important one" at launch, but a September 2026 analysis found Bing mainly in Deep Research. Make sure OAI-SearchBot can crawl the site first, and keep Bing indexing in place as well.
• Claude's web search relies on Brave for results; Anthropic lists Brave Search among its subprocessors. MERJ's analysis of Brave's discovery found that pages found through its users' browsing needed visits from around 20 distinct opted-in users before being picked up, or one if the page had appeared in another engine's results. For a brand-new page with little traffic, that is a real, sandbox-like threshold.
• Perplexity runs its own index and fetches pages live.
• Model knowledge reflects training data collected months before release. A brand founded after that point is unknown to the model until the next version, except through search.

How these retrieval systems fit together is covered in why AI search is made up of multiple retrieval systems rather than one engine.

What looks like a sandbox but usually isn't?

What does the evidence say about new and fresh content?

Freshness tends to help rather than hurt once content is found. Seer Interactive's 2026 analysis found about three quarters of pages cited by ChatGPT, Gemini and Perplexity had been updated within the past year. Ahrefs' 2026 analysis found that only about 38% of AI Overview citations came from pages ranking in Google's top ten. Pages outside the top results are cited regularly when they answer well. Neither finding fits a strict rule that new pages must wait. Both fit a picture in which discovery and quality decide.

How can you shorten the delay for a new site or page?

• Submit sitemaps to Google Search Console and Bing Webmaster Tools, and use IndexNow for Bing and other participating engines.
• Link new pages from pages already crawled, such as your homepage, navigation and popular articles.
• Allow AI search crawlers and user-triggered fetchers in robots.txt and your bot protection.
• Earn early mentions and links from established sites: partners, directories, press, industry lists.
• Create consistent profiles on the platforms assistants cite, with the same name and description as your site.
• Drive real visits through newsletters, social posts and communities. Visits help discovery in systems such as Brave's.
• Answer specific questions well, where established sources are thin, rather than competing head-on for broad topics.

How long GEO work usually takes is covered in how long GEO takes to show results. Why unique information matters is covered in how information gain improves AI search visibility.

How do you tell whether you are being held back?

Compare like with like. Take a new page and an older page on your site that answer similar questions equally well. If the older page is cited and the new one is not after a few weeks, check crawling and indexing first. Look for the new page in Google's and Bing's indexes, and in your logs for visits from each engine's crawler. If the new page is indexed everywhere and still passed over, the reason is more likely competition or content than age.

What might an illustrative case look like?

Consider an illustrative new cybersecurity consultancy whose site launched with twenty strong guides. After two months, Google's AI Overviews cited two of them, but ChatGPT and Claude never did. Its founders suspected a sandbox. Logs showed OAI-SearchBot had fetched only the homepage, and Bing had indexed four pages. The site had no links from other sites. The team submitted the sitemap to Bing, turned on IndexNow and linked the guides from its homepage. It was also listed in two industry directories and a partner's resources page. Within six weeks, Bing had indexed most of the guides, and ChatGPT began citing three of them.

What do people ask about an AI search sandbox?

How long should a new site expect to wait?

There is no fixed period. Discovery can take days to months depending on links, sitemaps and crawler access for each engine.

Do new pages on an established site face the same delay?

Much less. They are discovered through existing links and inherit the site's history.

Does buying an old domain avoid the delay?

It can bring existing links, but repurposing an expired domain mainly to manipulate rankings breaches Google's spam policies.

Is the delay longer for AI answers than for search results?

Often, for engines other than Google, because each index and model catches up separately.