Experiment 01: Can a Brand-New Domain Get Cited by AI Engines in 90 Days?

What does Experiment 01 test?

Experiment 01 tests whether a brand-new domain with 30 answer-first pages, and no link building, earns at least one AI citation for a non-brand prompt within 90 days. The hypothesis, prompts, engines and schedule were fixed on day 0 and do not change during the test.

The experiment answers one primary question and three secondary ones:

1. Primary: does at least 1 of 6 AI engines cite a growaisearch.com URL for at least 1 of the 48 non-brand prompts by day 90?
2. Secondary: on which day does each AI crawler first fetch the site, and which pages does it request?
3. Secondary: how many days pass between an engine’s first crawl and its first citation?
4. Secondary: how do mention rate and citation rate differ between the 6 engines over 13 weekly runs?

The design follows the testing methodology used for every experiment on this site: fixed inputs, a fixed schedule, every answer logged, and results published whether or not they support the hypothesis.

What was the starting point on day 0?

On day 0 the domain had no history: it was registered on 25 September 2026, and 30 guides, 2 free tools and 5 trust pages went live the same day. No links were built and no content was syndicated before the first run.

Every later change that could affect the outcome — a new page, a verified Search Console property, a first external link — is written into the weekly log below with its date.

Which prompts and engines are tracked?

The test tracks 50 fixed prompts on 6 engines: Google AI Overviews, Google AI Mode, ChatGPT search, Perplexity, Gemini and Microsoft Copilot. 48 prompts are non-brand questions a buyer or marketer asks; 2 are brand questions that measure entity recognition.

Each prompt is mapped to the growaisearch.com page written to answer it, so a citation of any other page on the site is recorded as well. The prompts cover 10 intent types:

Definition and how-to prompts make up 24 of the 50 because those are the questions the site’s guides answer most directly. The full list, with the expected page for each prompt, is in the prompt set download.

How is each answer recorded?

Each Monday, all 50 prompts run once on each of the 6 engines, giving 300 answers per week and 3,900 over 13 runs. For every answer the log records whether growaisearch.com is mentioned, whether it is cited, the cited URL and the position of the first mention.

Runs follow fixed conditions so that week-to-week changes come from the engines, not the setup:

• Signed out of every engine, in a private browser window.
• Same location and language settings on every run.
• Same weekday — Monday — for all 13 runs, plus a final run on day 90.
• Prompt text unchanged, pasted exactly as written in the prompt set.

The log also records how many sources each answer cites and the domains cited most often, so the result can be read against the competing sources. ChatGPT and AI Mode answers are also recorded by the Semrush AI visibility tracker, which Bikash uses daily; the other 4 engines are checked by hand. The fields match the five AI visibility metrics used throughout this site.

What is logged automatically?

A Cloudflare function on this site logs every request that 17 AI and search crawlers make for robots.txt, llms.txt and the sitemap. Crawlers fetch those files before any page, so the log dates each engine's first contact with the domain.

The logger reads the self-declared user agent of each request, so it counts GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Googlebot, bingbot and the other crawlers listed in the AI crawler directory. User agents are not IP-verified, so a spoofed bot is counted as the bot it claims to be; the weekly log notes any pattern that looks spoofed.

The logger runs only on those 4 files, never on the guides themselves, so pages keep answering in under 500 ms. Page-level crawls come from Cloudflare’s AI Crawl Control report for the domain, and visits from AI engines come from the site’s analytics; both are copied into the weekly log after each run.

What does the weekly log show so far?

The log holds three entries so far: the launch on day 0, then the domain going live and Search Console verification on day 1. Run 1 is on Monday 28 September 2026; each run adds a row with the date, citations and mentions per engine, and any change made to the site that week.

The weekly log sheet for each run is published here as a download after the run, so every figure in this table can be checked against the answers it counts.

What counts as a result?

The primary result is binary: yes if any engine cites a growaisearch.com URL for a non-brand prompt by day 90, no otherwise. The secondary results are dates and rates, reported per engine, with the number of runs each one rests on.

AI answers change between runs of the same prompt, so a single citation in one run is reported as a single observation, not as a stable position. A citation counts when the answer links a growaisearch.com URL as a source; a mention counts when the answer names Grow AI Search or growaisearch.com without a link.

What are the limits of this test?

The test covers one site, one topic and one location, so its result describes this domain rather than new domains in general. It measures what happened, not why: a first citation after a first crawl shows order, not cause.

Three limits are recorded before the first run:

• One observation per prompt per week. Engines vary between runs; a larger sample per week would narrow that noise but is not part of this design.
• Signed-out answers only. Personalised answers for signed-in users are not measured.
• Self-declared crawlers. The automatic log trusts user agents; spoofed traffic is flagged by pattern, not blocked.

Planned experiments on the experiments page reuse this prompt set as their control, so the Experiment 01 log also serves as their baseline.