· via dev.to (home feed)
Semrush data shows AI Overviews on 88% of tracked commercial keywords
AI Overviews appeared on 88% of a 400-keyword commercial panel in August 2026, up from 45.75% a year earlier, and a dev.to analysis explains how to keep pricing pages quotable.

The numbers
According to a Semrush study published in October 2026, Google's AI Overviews appeared on 88% of a fixed panel of 400 commercial keywords in August 2026, up from 45.75% in September 2025. The panel covered banking, software and ecommerce terms on US desktop search, and weighted by estimated search volume the coverage climbs to 97.08%. The figures were highlighted in a dev.to post that also lays out what site owners should do about them.
The post adds the necessary caveat: this is one panel of 400 keywords, not a census of commercial search. Even so, the trend is steep. Queries shaped like "X pricing", "X vs Y" and "best X for small teams" now usually return an AI-generated answer above the list of links.
Where pricing pages go wrong
The dev.to analysis catalogues the ways modern pricing pages become invisible or ambiguous to systems that read HTML rather than interact with a page:
- Prices that exist only after JavaScript runs, because a monthly/annual toggle renders numbers client-side, and some AI fetchers never execute JavaScript.
- Only one billing period present in the markup, with the other appearing on click.
- Currency chosen by geo-IP, so a US crawler sees dollars while a European buyer sees euros, and the answer quotes whichever it happened to find.
- Plan limits buried in tooltip popovers rather than body text.
- Comparison tables built from generic container elements, so the row-and-column structure that makes them readable to people is lost to machines.
- No visible date, which leaves the AI unable to distinguish current pricing from an old blog post quoting an outdated price.
The last point compounds the others. As the post frames the dynamic: if a third-party review states a price in one clean sentence while your own page shows a from-price inside a client-rendered card, the review is simply easier to quote, even when it is wrong.
The recommended fixes
The post ranks remedies by impact. First, render every price in the server-side HTML for both billing periods, hiding one with CSS, while acknowledging that hidden text is not guaranteed to be read, so key facts should also appear in visible copy. Second, add one plain sentence per plan directly under the plan cards: a single paragraph naming each plan, both billing periods, the currency and the date, which solves several failure modes at once and helps humans who skim. Third, build plan comparisons with real table elements and a caption. Fourth, add structured data such as schema.org's SoftwareApplication with offer objects, keeping the numbers identical to the visible text, because a mismatch is worse than no markup at all. Fifth, search for your brand plus "pricing", see which sources the AI answer cites, and push outdated third-party pages, and your own old posts, to carry a dated pointer to the current pricing page.
A script to audit it
To check all of this, the post ships a small Node 18+ script that fetches pages without executing JavaScript and reports what a non-rendering reader sees: HTTP status, whether prices exist in the raw HTML, whether both monthly and annual wording appears, explicit currency, real table markup, plan names in headings, an FAQ section, a visible date, absence of noindex directives, and any JSON-LD structured data. A frequent result, it notes, is a currency symbol present in static text while every actual number is rendered client-side: the page looks fine in a browser and says nothing to a fetcher.
Comparison pages carry the same risks, the post notes, often with higher stakes, since "X vs Y" queries are increasingly summarised directly in the answer box.
Why it matters
Commercial keywords were the part of search that actually converted: pricing, comparison and "best X" pages sit at the bottom of the funnel. If AI Overviews now answer those queries directly, those pages lose clicks even when the answer is assembled from their content. That reframes routine front-end decisions, such as server-side rendering, table semantics, structured data and visible dates, as revenue questions rather than styling preferences. It also introduces an accuracy risk: when an AI answer quotes a stale third-party price, the vendor effectively loses control of how its own pricing is communicated. The panel is small and US-desktop only, but a jump from under 46% to 88% in roughly a year suggests the window to fix transactional pages is closing.
- #seo
- #ai-overviews
- #google-search
- #structured-data
- #semrush