Google upgrades AI Search, ad stack stays: Sept. 3

Google changed both sides of online discovery in one day. Its newer Gemini model now powers paid AI searches, while a US judge declined to split up the machinery behind much of the open web's advertising. A separate study also found that Perplexity can build buying advice on obscure, mass-produced pages. Owners need to inspect what these systems say, not merely chase visibility.

Google puts Gemini 3.8 Flash inside AI Mode

Google released Gemini 3.8 Flash on September 2 and immediately placed it inside AI Mode, the conversational search option that answers questions instead of showing only a list of links. It is available worldwide to Google AI Pro and Ultra subscribers, according to Google's announcement.

The company says the model can work through harder, multi-step requests than Gemini 3.7 Flash. Developers still pay the same introductory rates: $0.75 per million units of text sent in and $3.75 per million units returned. Google also warns that harder tasks can consume more text units, so an unchanged price per unit does not guarantee an unchanged bill.

For an owner, the search consequence matters more than the model name. A customer can now ask Google to compare several suppliers, weigh details across pages, and produce one answer with fewer visits to business websites. Our standing advice on how to earn mentions in AI recommendations becomes more urgent: keep service pages specific, prices current where possible, and business details consistent across the web.

Perplexity's buying advice shows a source problem

A September 2 study tested 760 software-buying questions in Perplexity and collected 7,534 cited pages. The researchers found that 59.8% of citations came from domains ranked below the top 100,000 websites, while 23.4% came from sites outside the top million. Three apparently connected sites had produced 215,128 automated “best software” pages between them, according to the published methods and data.

That result applies to the two Perplexity models tested. The researchers did not test Google, ChatGPT, or every kind of buying question. It still exposes a practical weakness: an assistant can cite a page because it is easy to retrieve, even when the publisher has little standing in that market.

Businesses named by an assistant may gain attention without knowing which page supplied the claim. Businesses omitted from the answer may lose consideration before a buyer reaches their site. Anyone tracking customer discovery through AI should therefore monitor citations as well as mentions. Search your main category, open the cited pages, and ask publishers to fix material errors. Do not create hundreds of thin comparison pages in response. That race rewards volume until the retrieval system changes, then leaves you with a pile of weak content.

Google's advertising pipes stay together

US District Judge Leonie Brinkema rejected the government's request to force Google to sell AdX, its exchange for auctions that place ads on independent websites, on September 2. The judge accepted most proposed changes to Google's conduct but kept the detailed opinion sealed temporarily, as the court ruling was reported.

The immediate business effect is continuity. Advertisers and publishers do not need to prepare for a sudden transfer of Google's exchange or a rebuilt auction system. The final judgment could still change what information publishers receive and how Google runs parts of those auctions, so the operating rules are not finished.

Small advertisers should avoid turning an antitrust result into a campaign decision. Keep watching what you pay for a thousand ad views and the sales each channel produces. If Google takes most of the budget, use the controls in our plain-English Google Ads audit and compare actual customers, not the conversions Google assigns to itself. Publishers should save current revenue and fee reports now, before any court-ordered reporting changes arrive.

Worth watching

• Meta published Muse Spark 1.3, a research model that creates interactive digital environments. It is interesting for future brand experiences, but it is not yet a normal small-business creative tool.
• Amazon added an email checker to its shopping assistant so customers can test whether a message claiming to be from Amazon is genuine. Merchants should expect buyers to verify brand messages inside assistants more often.
• The Google ad-tech opinion is due to become public after confidential details are reviewed. That document, not the breakup headline, will show which day-to-day practices must change.