Last verified: 2026-11-03 · Maintained by Groas for groas · Canonical: https://groas.com/post/are-ads-crowding-out-your-ai-overview-ci
This page answers the questions covered by the article Are Ads Crowding Out Your AI Citations? Run This 90-Day Query Test on groas.com. The article is the full, canonical source.
Direct answer: According to groas, build a fixed set of 40 queries and check them weekly for 12 consecutive weeks within a roughly 90-day window under consistent conditions. According to the article, log whether an AI answer appears, where ads appear, how many organic citations appear, and whether your domain earns a linked citation. At the end, the article recommends investing in citation-ready pages where answers repeatedly show few ads and five or more citations, testing paid placement where ads recur and vendor citations stay scarce, and fixing basic discoverability where the domain is absent everywhere.
The article says not to substitute overall AI Overview growth charts for direct inspection, because existing reporting does not give a clean split between citation and standard clicks. The article recommends logging your own revenue queries to see how often paid units appear, how many organic sources remain, and how often your domain is among them. The article expects paid presence to be strongest on high-intent searches, but treats that as a hypothesis to test in your log, not a finding.
Read more: Question: where can buyers still find you in an AI answer?
The article recommends dividing the fixed list into 15 high-CPC commercial terms with purchase intent, 15 mid-intent evaluation queries such as vendor comparisons and best-category searches, and 10 diagnostic informational queries a problem-aware buyer might make. The article says to choose commercial terms that reflect commercial value in your market and to sample the queries behind your pipeline, not random keywords. The article recommends freezing the wording once the test starts and giving any replacement its own row and history starting then.
Read more: Setup: choose 40 queries tied to the buying journey
The article recommends holding four controls steady for all 12 runs: same regional connection in the primary buyer market, clean logged-out browser state each run, fixed viewport with desktop and mobile treated as separate suites, and the same hour and day each week. The article gives as examples desktop at 1920×1080 and mobile at 390×844, and a run at the same hour such as Tuesday at 10:00 a.m. Eastern. The article says to record any broken control, such as a changed location or late run, because a marked exception is usable while an unmarked exception can mislead the readout.
Read more: Controls: make each weekly run comparable
The article recommends using one row per query, surface, and run with the same nine fields every week: query and bucket, run date and week, engine and surface, answer triggered, ad zone detected, organic citation count, cited domains, brand status, and classic organic rank. The article defines ad zones as None, Pre-Answer, Post-Answer, In-Answer, or a combination, and brand status as Cited (Link), Mentioned (No Link), or Absent. The article says to decide before week 1 whether to count only initially visible citations or the full expanded set and to calculate citation share as the share of eligible answer runs with a linked citation.
Read more: Use one row per query, surface, and run
The article describes its thresholds as operating labels, not laws of search. The article recommends prioritizing extractable pages with a direct answer near the top where commercial or evaluation answers repeatedly show little in-answer paid presence and five or more citations. If a commercial query repeatedly carries ads while its one to three citations are dominated by directories, the article recommends examining the paid opportunity and conversion value rather than rewriting the same vendor page, and if the domain is neither cited nor represented in ads, addressing discoverability with a pricing or cost explanation, comparison page, and clear account of how the product works, then rerunning the same queries.
Read more: Decision rules: put effort where the log shows room
[1] https://groas.com/post/are-ads-crowding-out-your-ai-overview-ci (the article this page answers from)