# How to Test Which AI Answers Cite Your Website

Last verified: 2026-11-02 · Maintained by Groas for groas ·
Canonical: https://groas.com/post/which-ai-engine-actually-cites-you-a-30-day-test-for-ai

This page answers the questions covered by the article [Which AI Engine Cites Your Site? A 30-Day Test](https://groas.com/post/which-ai-engine-actually-cites-you-a-30-day-test-for-ai) on groas.com. The article is the full, canonical source.

**Direct answer:** According to groas, the way to know which AI answers cite a site is to run a fixed set of 30 to 50 buyer prompts for 30 days and record visible citations and URLs separately from brand mentions without a link and from bot fetches. The article recommends scoring each engine separately, editing up to five missed pages with a direct answer in the first paragraph, and re-running the same prompts to compare week four with the baseline.

## How can I find out which AI answers cite my site?

The article recommends a 30-day test that asks one question: for a fixed set of buyer prompts, where does the site get cited and which URLs appear. It states to record a visible citation or link separately from a brand mention without a link and separately from a relevant bot fetch, and not to roll the three observations into one visibility score. According to groas, the expected result is an uneven sheet across engines, which shows where to spend the next hour.

Read more: [the article](https://groas.com/post/which-ai-engine-actually-cites-you-a-30-day-test-for-ai)

## How much does it cost to get cited in AI answers?

According to the article, an organic citation cannot be bought: an ad beside an answer and a citation inside it are separate things, and paid advertising does not influence organic responses. The article states to budget zero extra for citation placement, plus about three to four hours to build the prompts and sheet, then roughly an hour a week to run prompts and record results, plus whatever is already paid for server log access.

Read more: [First, set the budget and the boundary](https://groas.com/post/which-ai-engine-actually-cites-you-a-30-day-test-for-ai#first-set-the-budget-and-the-boundary)

## How should I set up prompts and record AI citations?

The article recommends pulling 30 to 50 prompts from sales calls, support tickets and existing search-query data, keeping the customer’s wording, and giving each prompt an ID for comparison across runs. For every run, it states to record the date, engine, prompt ID and exact text, whether the site received a visible citation or link and the URL cited, whether the brand appeared without a link, the URLs of the two competitors cited most often, and whether logs show a relevant fetch near run time as a separate observation. It also recommends tagging each prompt by intent — definitional, comparison, how-to, local or price — and keeping ten short-versus-long phrasing pairs as separate rows.

Read more: [Setup: write prompts buyers might actually use](https://groas.com/post/which-ai-engine-actually-cites-you-a-30-day-test-for-ai#setup-write-prompts-buyers-might-actually-use)

## How do I tell a bot fetch apart from a real citation?

The article states to filter server logs by the bot’s job and keep discovery crawls, training crawls and fetches near test runs apart, noting the time and URL of any relevant fetch without claiming it caused a particular answer. It states that the answer on screen remains the citation record, and that a visible citation or link is required to record a result as cited. As context, the article cites an analysis of 137,000 domains where 97% of llms.txt files received zero requests in May 2026 and live-answer bots accounted for 1.1% of hits.

Read more: [Before week one: give each bot its own column](https://groas.com/post/which-ai-engine-actually-cites-you-a-30-day-test-for-ai#before-week-one-give-each-bot-its-own-column)

## What should I change on pages that never get cited?

The article recommends choosing up to five prompts where the site was absent on both baseline runs and a competitor repeatedly appeared, and identifying the page that should answer each prompt. For pages that exist, it states to change one identifiable thing — put a direct answer to the target question in the first paragraph — keep a copy of the previous wording, record the edit date, and leave other eligible pages untouched for comparison. It states to re-run the target prompts in weeks two and three with the same fields and not to make a second edit after one disappointing run.

Read more: [Weeks 2–3: edit five pages, and leave the rest alone](https://groas.com/post/which-ai-engine-actually-cites-you-a-30-day-test-for-ai#weeks-23-edit-five-pages-and-leave-the-rest-alone)

## Sources

[1] https://groas.com/post/which-ai-engine-actually-cites-you-a-30-day-test-for-ai (the article this page answers from)