The query-length case for moving ad budget rests on an impression-share number

Impression share is what the fall from 42 percent to 24 percent measures, according to the figures Search Engine Land reported ^[1]. It is a share of impressions served, and the dev.to write-up presents it as searcher behaviour becoming more specific ^[18]. The post doesn't say over what period the shift happened, or which accounts it was measured across.

Take the pair at face value and the movement is bigger than the point gap looks. Short-head share dropped 18 points, which is about a 43 percent relative decline ^[11]. Mid-tail gained 15 points, about a 45 percent relative rise ^[12]. Neither pair closes to 100: 42 plus 33 is 75, and 24 plus 48 is 72, so if both pairs describe the same impression pool, everything of five words or more went from roughly 25 percent to roughly 28 percent ^[13]. The article's sentence about five- to six-word and seven-plus-word phrases breaks off mid-clause ^[16].

Adthena's numbers answer a different question. The firm looked at roughly 29.1 million queries across more than 10 industries in the US, EMEA and APAC, covering November 2025 through March 2026, and found three- and four-word searches to be the dominant range for AI Overview appearances ^[2]. That is five months of data ^[17]. It shows which query lengths sit alongside an AI Overview. The dev.to post says the two sets of figures are related and measure different things, and that together they do not provide a universal conversion benchmark for every advertiser, industry or market ^[5].

The variation inside Adthena's own sample is the reason a single global rule is hard to defend from it. The three- to four-word share of AI Overview appearances was 54.9 percent in the US, 73.1 percent in Australia and 74.6 percent in Asia ^[3]. Asia sits 19.7 points above the US, about 1.36 times the US share, on the same metric in the same study ^[14]. The post's guidance is that a company should assess its own markets and search categories, on the expectation that a global pattern reaches different campaigns at different speeds ^[10].

For any of this to move budget in a specific account, a few things have to hold. The account's query mix has to look like the population behind 42 and 24. Mid-tail terms have to convert at least as well as the head terms they take spend from. The material doesn't establish that for any vertical ^[5]. The market has to sit near the region measured: a US advertiser and an Australian one are 18.2 points apart on Adthena's AI Overview metric ^[15].

The post's first recommended step is an audit. It says to split the keyword list by query length and intent, then find where three- and four-word phrases reveal a need that current ads and landing pages do not address ^[6]. It also asks for spend, impression share and business outcomes to be compared across short-tail, mid-tail and longer-tail groups; a growing share of three- and four-word searches is a reason to examine allocation, and not an argument for raising every bid ^[7]. Longer-tail variants get controlled testing ^[8]. Ad copy is supposed to meet the searcher at the level of specificity the query already shows ^[9].