Loading today’s stories
Build1 distinct publisher3 min readPublished
The rewrite rate cited by Search Engine Land leaves a page's own meta description on screen for under a third of tracked queries. The body of the page carries the rest.
The Engineer · Build desk
Compiled by The EngineerSomething wrong?How this is made
Google's snippet generator is not inventing text. It selects page copy it judges closer to the query than the tag you supplied [3]. The candidate pool named in the analysis is unglamorous: the page title, key headings, the introductory paragraphs, product and service explanations, and the answers to common questions [6]. Those are all things written for people who have already clicked. They now double as the copy shown to people who have not.
Take the rewrite rate at face value and the tag you wrote is on screen for roughly three tracked queries in ten [9], while the page body feeds results a little over twice as often as the tag does [10]. The tag was never a ranking input in the first place; its only job is persuading a click [12]. That job now has a ceiling on how often it gets a chance to do it, and the ceiling is set by Google, per query.
Two things keep this short of a rule. The figure comes from analysis cited by Search Engine Land drawing on Seer Interactive's findings, and the material does not say how many queries were tracked or how they were chosen [11]. Seer's own live tests, meanwhile, found cases where a hand-written description beat both AI-generated copy and Google's rewrite on click-through [4]. Both can hold at once: the tag performs when it is served, and it is not served most of the time. The source does not report the split, so there is no published figure for CTR on the queries that kept the tag versus the ones that did not.
Which is why the recommendation lands on triage rather than coverage, with the likely value of a click setting the level of human review [5]. The pages named as worth writing by hand are the ones with revenue attached: core service pages, product-category pages, location pages, pricing or booking pages, and evergreen guides that pull qualified traffic [7]. On those, the description should state the offer or answer accurately, complement the title rather than repeat it, and avoid promising information, pricing, availability or outcomes the page does not substantiate [14].
Everything else is a drafting problem, and generated copy is serviceable for it provided a human checks the priority pages and anything making a specific business claim, because templates drift into repetitive or inaccurate text [15]. The measurement side is what breaks the one-off audit: a single URL can show several snippet variations inside a month [2], so the recommended practice is keyword tracking that surfaces when a rewrite happened and which text Google chose [8]. For a local service business, a generic opening leaves Google assembling something vague, while an opening that names the service, the audience or location where relevant, and the customer outcome gives it something usable [13]. Read that way, each rewrite is a free report on the query language Google associates with the page, which is worth more than the tag it replaced.
Ranked by verification strength, evidence, and original report placement.
The research's conclusion is a more selective content strategy: invest editorial effort where clicks have meaningful commercial value, use automation to cover scale, and let the likely value of a click determine the level of human review.
Manually written descriptions are described as most defensible on core service pages, product-category pages, location pages, pricing or booking pages, and evergreen guides that generate qualified traffic.
The visible snippet in Google's results can be drawn from page text other than the meta description when Google considers that text more relevant to the specific query.
The page elements identified as the material Google can use for a snippet are the page title, key headings, introductory paragraphs, product or service explanations, and answers to common questions.
Because Google can create multiple snippets for one URL, a one-time metadata audit is insufficient; the research recommends SERP monitoring through keyword tracking to identify when a rewrite occurs and what text Google selected, which yields evidence about the query language Google associates with the page.
The 70% figure is presented as analysis cited by Search Engine Land that brings together Seer Interactive findings and testing of manual, AI-generated and Google-generated descriptions; the material does not state the size of the tracked query set or how those queries were selected.
Follow any of these and your For You feed starts watching them — no settings page required.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Thin, third-hand quantitative chain
One publisher, one article, and the load-bearing number is relayed from Search Engine Land's round-up of Seer Interactive testing with no query-set size, selection method or time window. The mechanism and workflow claims are internally coherent and unexceptional, but nothing in the cluster is independently corroborated or primary.
No adoption signal in cluster
The supplied material contains no release, deployment, benchmark run, pricing or usage disclosure — only advice about page copy and SERP monitoring. There is no evidence of how many sites, tools or teams follow this practice, so adoption cannot be measured without inventing facts.
Headline rate stated more firmly than its sourcing allows
The cluster's framing turns a single third-hand statistic into a settled state of the SERP ('Google rewrites the snippet on 70% of tracked queries'), while the same article concedes it cannot say how many queries were tracked or how they were chosen, and the CTR evidence for the recommended remedy is only 'cases' of lift. The mechanism claim — snippets can come from body text — is uncontroversial and not overstated, which keeps the gap moderate rather than severe.
No disclosed incentives
The material discloses no commercial relationship, funding, product placement or vendor interest for the author, the publisher, or the parties whose analysis is cited. Assigning an incentive score would require assuming motives the sources do not state.
Low
Confidence is capped by a single-publisher cluster, an unverifiable headline statistic and the complete absence of adoption or incentive evidence. What can be held with reasonable confidence is narrow: that the source states this mechanism and recommends this workflow.
build
Two thirds of US searches end without a click. Position is no longer a traffic report1 distinct publisher
product
Google's new Preferred Sources button hands publishers the recruiting job2 distinct publishers
build
Cloudflare's one-click AI block names GPTBot, not the bot that decides if ChatGPT cites you1 distinct publisher
product
Reddit joins the S&P 500 because a landlord got bought, not because the stock worked1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
dev.to
1 article · August 24, 2026