Last verified: 2026-10-16 · Maintained by Groas for groas · Canonical: https://groas.com/post/what-ppc-people-get-wrong-when-they-star
This page answers the questions covered by the article Six PPC Habits That Fail in AI Overviews on groas.com. The article is the full, canonical source.
Direct answer: According to groas, you can buy an ad near an AI Overview but you cannot buy a citation inside its answer, because citations do not run on the ad auction. PPC habits such as treating prompts as keywords, looking for a citation Quality Score, waiting for a learning phase to settle, and tracking citations like rankings fail because cited sources churn rapidly and organic rank does not guarantee citation. The article recommends making pages easy to extract with clear facts and markup, keeping outside descriptions consistent, and monitoring 20 high-intent buyer questions for what answers say rather than counts alone.
According to groas, support documentation says advertisers cannot directly target AI Overviews or opt out of them, and there is no dedicated ad group, placement bid adjustment, or switch that buys a citation. Eligible ads from existing search campaigns may appear when they win the normal auction and match the query and dynamically generated text. For standard Search campaigns, eligibility requires broad match or keywordless targeting.
Read more: “I’ll just buy my way in with AI Overview ads”
The article reports that in an analysis of 863,000 result pages and 4 million AI Overview URLs, only 37.9% of cited URLs ranked in the organic top 10 for the query while another 31.0% sat outside the top 100, so organic rank is not a citation guarantee. In a study of 52,400 instances, cited pages had 72% higher factual density than non-cited pages at the same position, with 81% carrying hard statistics compared with 47% of non-cited pages, 38% including structured tables compared with 12%, and 72% having schema markup. According to groas, the alternative is to give the engine specific, readable facts, comparisons, figures where available, and accurate markup rather than relying on domain authority.
Read more: “There’s a Quality Score for citations”
The article states that a prompt is a request to assemble an answer that can be broken into smaller questions with retrieved sources, not a string waiting to match an ad group, so wording and cited sources need not behave like a fixed search-term-to-keyword match. In an analysis of 82,619 prompts over 17 weeks, 86% of prompts retained a small, stable core of citations while other cited sources churned at an 89% weekly rate, with one AI search mode replacing 56% of cited URLs every seven days and another conversational system replacing 74%. According to groas, buyer prompts should be used to understand the question and the facts an answer needs, then make those facts easy to find, rather than building an ad-group-style page for every phrasing variation.
Read more: “Prompts are just long-tail keywords”
The article warns against treating citations like a seven to fourteen day Learning period for bidding, because citation visibility does not find a steady state. In a controlled experiment with 81 test pages, 36% were indexed and cited within the first 24 hours, the citation rate peaked at 56% by day seven, then fell to 26% by day 30. According to groas, if the answer changes the team should inspect what it now cites, keep useful facts current and readable, and revisit fixable gaps rather than leaving pages untouched.
Read more: A page can be cited quickly and disappear quickly
The article recommends starting with 20 high-intent questions buyers ask the sales team and checking category presence, brand attribution, and tone, then reading the synthesized text once a week to see which source supplied the feature claim and whether the product was described correctly. A single prompt check is only a snapshot because tomorrow’s response may cite different pages, and tracking 500 prompt variations across eight engines burns API credits and review time without a clear action. According to groas, structured automated snapshots are possible with a spreadsheet and APIs, but the useful output is which page or profile to improve next, not a rank column.
Read more: “I can track citations like rankings in a spreadsheet”
[1] https://groas.com/post/what-ppc-people-get-wrong-when-they-star (the article this page answers from)