01 The consumer data
A billion people now talk to the assistant
Google’s Gemini app crossed a billion monthly users. Most of them speak instead of type.
Google announced on August 11 that the Gemini app passed 1 billion monthly users, up from a disclosed 400 million in May 2025. The same post shared how people use it: 63% talk to Gemini directly, and Google says the voice-only group is growing.
The rest of the numbers sketch a search behavior that looks nothing like a keyword box. About one in five Gemini Live conversations adds a camera feed or screen share, 38% of school requests carry an attachment, and the app generates more than 150 million images a day.
Search Engine Journal’s Matt Southern adds the caveats Google left out. Google did not explain how it counts a monthly user, and its wording quietly shifted from 950 million “monthly active users” in July to 1 billion “monthly users” in August. Read the milestone as marketing with real usage underneath it.
For your firm, the behavior shift matters more than the round number. A typed search is “personal injury lawyer Phoenix.” A spoken question is a full sentence with a situation attached, closer to what a client says across a kitchen counter than what they type into a box.
What to do about it
Read your own site the way a spoken question arrives. A page built as a keyword list answers typed queries; a page that answers a whole question in its first two sentences answers spoken ones.
Start with your intake FAQs. The questions clients ask on the phone are the questions they now ask assistants out loud.
Source
Google, “Gemini app reaches 1 billion monthly users,” August 11, 2026 ·
blog.google
· Secondary: Search Engine Journal, Matt G. Southern, August 12
02 The research & evidence
ChatGPT drafts its guest list before it searches
A traffic analysis finds brand names inside ChatGPT’s first search query, written before any result comes back.
SEO consultant Suganthan Mohanadasan spent two months reading the network traffic behind his own ChatGPT conversations, and published the results August 14 in Search Engine Journal. In 21 of 27 conversations, ChatGPT’s very first search query contained brand names the user never typed. Nothing had been fetched yet, so the names came from the model itself.
The gap between the named and the unnamed is the finding. Brands written into ChatGPT’s own query showed up in the final answer 68.9% of the time; brands merely fetched during the search made it 2.1% of the time, roughly a 33-fold difference. In 86 cases, a brand got recommended without its website being fetched at all.
A second filter runs after the query, and it is harsher. Of 3,554 retrieved pages in his sample, 110 earned a citation, a rate of 3.1%, with position inside a domain’s group of results predicting nearly everything.
The caveats are in the piece itself: this is one account and a few hundred conversations, and the author calls every percentage a direction rather than a measurement. The mechanism, though, can be reproduced on any account in two minutes with browser developer tools.
It reads like a job interview where the hire was decided in the hallway. Most of what is sold as AI visibility work targets the retrieval stage, the 2.1% column; the 68.9% column is built by being written about, reviewed, and compared across the open web for years.
What to do about it
Ask ChatGPT the question your clients ask, five times, and note which firms it names on each run. Names that appear every time are your real competitive set inside the model; if yours never appears, the fix is not on your website.
That fix is coverage: directory listings, bar association profiles, local press, and the comparison pages your market reads. Route any AI-assisted content through a documented human review before it publishes.
Source
Search Engine Journal, Suganthan Mohanadasan, “ChatGPT Already Knows Who’s In The Running Before It Searches,” August 14, 2026 ·
searchenginejournal.com
· Secondary: the author’s original write-up
03 The platform shift
The bot reads pages that told it no
Crawler data shows ChatGPT’s fetch bot reaching disallowed pages, and OpenAI says the rule may not apply to it.
TollBit’s State of the Bots report for the first half of 2026 measured how AI bots treat robots.txt, the plain-text file where a site lists which crawlers may enter. On the European sites studied, about 15% of identified AI page-fetchers reached URLs the sites had explicitly disallowed.
A few agents account for most of it. ChatGPT-User, ByteDance’s Bytespider, and Youbot each reached disallowed pages on nearly half the European sites that had listed them, and ChatGPT-User reached the most sites of any bot. OpenAI’s own documentation says robots.txt may not apply to ChatGPT-User, because a human asked for the page.
The distinction underneath matters more than the outrage. The agent that decides whether you appear in ChatGPT search results is a different bot, OAI-SearchBot, and it does honor the file. A firm that blocks both has traded away its visibility while keeping a control the fetch bot is documented to walk past.
A robots.txt line is a posted sign, not a lock, and one visitor has been told the sign is not addressed to him. Structural enforcement is coming from the network layer instead: from September 15, Cloudflare will block training and agent crawlers by default on ad-carrying pages of newly added domains.
What to do about it
Decide what you are trying to control before touching robots.txt. If you want clients to find you through assistants, blocking the search bots works against you; if you want evidence of what the fetch bots take, the file cannot provide it.
Your server logs can, which is exactly what the next item is for.
Source
TollBit, “State of the Bots,” first-half 2026 report ·
tollbit.com
· Secondary: Search Engine Journal, Matt G. Southern, August 14
04 The operational change
A free meter for what the machines take
Microsoft Clarity now scores AI scraping against the visits it sends back, one operator at a time.
Microsoft added an AI Scrape-to-Referral Ratio card to Clarity, its free analytics tool, reported August 13 by Search Engine Land. The card compares each AI operator’s scraping activity against the visits it refers back, so the trade stops being a matter of opinion.
Microsoft’s own example shows a bot taking roughly 6,000 page scrapes for every one visit it referred. The report ranks operators by referral return and links straight into session recordings of AI-referred visitors, so you can watch whether those people scroll, read, or fill out your contact form.
This is the ledger the last story needs. Robots.txt records what you asked for; your analytics record what actually happened, and now the exchange rate between content taken and visits returned sits on a single card.
One caution before anyone weaponizes the number. As the other stories in this issue show, a recommendation can arrive with no fetch at all, so a lopsided ratio is an input to a decision rather than an automatic reason to block.
What to do about it
If your site runs Clarity, open Bot Analytics this week and read the ratio for each operator. If it does not, the tool is free and installs with one script tag, an afternoon task for whoever manages your website.
Note which assistants send visitors who complete your contact form. That list, from your own traffic, is your evidence for what AI visibility is worth in your market.
Source
Search Engine Land, Barry Schwartz, “Microsoft Clarity AI Scrape-to-Referral insights report,” August 13, 2026 ·
searchengineland.com
· Secondary: Microsoft Clarity AI Visibility Dashboard
05 The money
Your ad account starts narrating its own results
Google is adding agentic AI summaries and chat-built reports across Ads and Analytics.
Google is expanding Ask Advisor, its AI assistant inside Ads and Analytics, with what it calls agentic capabilities, announced August 10. Google Analytics homepages will open with an AI summary of what changed since you last logged in, with the option to receive those summaries by email.
Google Ads gets a redesigned homepage with personalized AI insight cards, and advertisers can ask natural-language questions, such as how competitors are affecting their share of ad views. Dashboards built from a plain text prompt, with automatic summaries explaining the trends behind the charts, arrive in Ads first.
For a firm without a marketing department, this is real convenience; interpreting an ads report was never going to be a partner’s second job. The commentator on this scoreboard, though, is on the home team’s payroll. The system summarizing your ad performance is owned by the company selling you the ads.
The defense is the one this brief keeps returning to. Keep an independent count of what Google cannot see: intake calls, consultations booked, clients signed, and judge campaigns against that ledger rather than the platform’s narration.
What to do about it
Treat the AI summaries as prompts for questions rather than verdicts. When a card says a campaign improved, ask improved against what, and check the claim against your own intake numbers for the same period.
The standing rule applies here as everywhere: state bar advertising rules govern ads managed with an AI assistant exactly as they govern ads a human wrote, including required disclaimers and review of any claims.
Source
Google, “New agentic capabilities in Google Ads and Analytics,” August 10, 2026 ·
blog.google
· Secondary: Search Engine Land, Anu Adegbola, August 10
