Google quietly started testing a homepage without a search button. Similarweb revealed that 26% of ChatGPT answers now carry sponsored ads. Gemini crossed a billion monthly users. And a New Zealand study showed AI search engines still disagree 63.5% of the time. That was the week in AI search — here are the stories worth your time, plus one practical takeaway.
The search page is being rebuilt
The biggest structural story this week: Google is testing a homepage redesign that removes the classic search button for logged-out users, replacing it with Gemini AI shortcuts and a prominent AI Mode button. The test was spotted across Chrome, Edge, and Perplexity's Comet browser — an explicit shift from "type a query, press search" to "ask an assistant." At the same time, Google's official Gemini 3.7 Flash announcement confirms the model that now powers AI Mode for AI Pro and Ultra subscribers, with a rollout to the free tier expected later. Early SEO testing found mixed results on intent understanding — which is typical of a model swap in progress.
Why it matters: the entry point to the web is becoming an answer engine. Perplexity now handles more than 10% of search queries via direct answer citations, and ChatGPT's web search is taking branded traffic that used to belong to Google.
AI answers are getting ads
On August 17, Similarweb launched an AI Ads intelligence dataset covering ChatGPT, Google AI Mode, and Google AI Overviews — see the official AI Ads product page for the full methodology. The headline numbers from the dataset:
- 26% of ChatGPT responses (Free and Go tiers) carry a sponsored ad
- Nearly 30% of ad-eligible AI Mode queries show ads
- Over 40% of Google searches now trigger an AI Overview, often with ads
The dataset matters because none of these platforms maintain public ad libraries — this is the first independent look at how much of an "answer" is commercial.
How much can you trust an AI answer?
A study by The Optimisers ran 211 matched queries across ChatGPT, Google AI Mode, and Microsoft Copilot: the engines disagreed on the top answer 63.5% of the time. ChatGPT's answers changed for 85.6% of queries when personalization memory was off, and 34.8% of power-provider answers still named a discontinued brand. We covered this study ourselves on Monday with a practical verification workflow — read it here — but the short version is: for anything you'd act on, cross-check with a second engine and open the cited sources.
Assistants at billion-user scale
On August 12, Google announced the Gemini app passed 1 billion monthly active users — its fastest-growing product ever, with 63% of usage now voice-based and 150+ million images generated daily. The milestone was announced by Google CEO Sundar Pichai, and landed alongside Ticketmaster's new connected app inside Gemini. It follows ChatGPT crossing 1 billion weekly users just weeks earlier — two assistants at comparable scale, with completely different distribution strategies.
The models behind the answers
This week's model news was fast and cheap:
- Grok 4.6 (xAI) claims GPT-5.6 Sol / Claude Opus-level intelligence at half the price — $2/$6 per million tokens — with strong agentic and financial-tool benchmark scores (official announcement).
- OpenAI previewed GPT-5.6 "UltraFast" mode — up to 750 tokens per second peak, a claimed 14x speedup (official announcement).
- DeepSeek v4 Pro shipped as an upgraded open-weight release under an MIT license (official model releases).
- The Arena leaderboard (week 33) shows Anthropic models at #1-#3 with claude-fable-5 on top (ELO 1506), while Alibaba's qwen3.8-max reached #8 and Google's gemini-3.7-flash-high debuted at #9 — the first time three different model families sit inside the top ten.
- Developer preference is consolidating too: a ZDNET survey reports roughly three out of four developers now prefer Claude Code over OpenAI Codex.
Video summarization: the new frontier
Two pieces worth reading if you live in long videos:
- OpusClip ranked #1 in a 2026 field test of the category, for both summarizing and repurposing long videos.
- Fora Soft's engineering guide to the category — the useful mental model here is the divide between transcript-first tools (cheap, read captions) and multimodal tools (see frames and hear audio), with hybrid transcript-plus-keyframes designs becoming the 2026 consensus.
- The free-tier extension race continued: Video Summarizer for YouTube shipped v2.1.0 with one-click ChatGPT/Claude summaries.
Open-source and the community
Two newsletters remain the best weekly pulse for builders: Builder Radar (week of Aug 16) covers the week's most-starred projects (including graphify, which turns codebases and PDFs into queryable knowledge graphs), and Pollak's AI Intelligence Briefing has the densest daily context dump around.
What we're watching next
Three threads for next week: whether Google's search-less homepage test expands (and what it does to "Google it" muscle memory), whether the Similarweb ad data forces platforms to publish ad transparency libraries, and whether multimodal video summarization moves from engineering curiosity to default — because that's the category where the tools most people actually use are still getting better every week.
FAQ
Why do AI search engines give different answers to the same question? Each engine uses its own index, retrieval stack, freshness weighting, and personalization — a study this week found the top answer differed 63.5% of the time across ChatGPT, Google AI Mode, and Copilot.
Are AI answers with ads trustworthy? A sponsored placement is still an answer, but it's a commercial one — Similarweb found 26% of ChatGPT responses carry ads. Treat the sponsored part of an answer the way you'd treat a promoted link.
What changed for free users this week? Gemini 3.7 Flash is rolling into AI Mode for subscribers first; the homepage test affects logged-out users; and free tiers of ChatGPT and Google now show more ads in answers.
Is Gemini or ChatGPT bigger now? Both claim roughly a billion users: ChatGPT crossed 1 billion weekly users a few weeks before Gemini hit 1 billion monthly active users on August 12.
By the Glarity Editorial Team. The Glarity Editorial Team writes about AI search, video summarization, and getting more from your browser.
