01 The market data
The channel that built you can unbuild you
A magazine lost the referral engine behind its peak audience, then rebuilt on four smaller ones.
Hello!, the celebrity and royal title, told the Press Gazette podcast that Google Discover helped carry it to a peak of about 40 million unique visitors a month globally. Then Discover’s algorithm shifted toward social posts, YouTube videos and “locally-relevant” content, and chief content officer Sophie Vokes-Dudgeon called the result the biggest requirement for a pivot the brand had faced.
Discover is Google’s scrolling feed of articles it thinks you might want without your asking — a recommendation surface, not a results page. The distinction matters, because traffic the reader never requested can be withdrawn without notice.
The rebuild is more instructive than the loss. Hello! plugged the advertising gap by syndicating to Apple News, MSN, Yahoo and Newsbreak, and treats that revenue as a bridge to its first paid membership rather than a permanent fix. Its roughly 35 editorial staff also stopped writing the off-brand stories that had performed well on Discover, gardening among them.
The counterintuitive number is the one that held. Google Search traffic did not collapse alongside Discover, because news-led coverage kept earning clicks that summaries could not fully absorb. Losing a recommendation channel and losing search are different injuries, and they need different treatments.
What to do about it
Open your intake log and count what share of new consultations trace back to a single source — one directory, one referral partner, one platform. If any single source is over half, that is not a channel, it is a landlord. Name the second source you will build this quarter and give it a number to hit.
Source
Press Gazette, "How Hello! fought back after saying goodbye to Google Discover traffic," August 6, 2026 —
pressgazette.co.uk
02 The research
Two question types carry nearly every brand mention
New open data suggests an AI visibility score measures the questions asked as much as the brands answered.
Dr. Peter J. Meyers at Moz published a free dataset of 50,000 query fan-out prompts, generated from 1,000 subtopics across 20 industry verticals using Google’s Gemini API. Query fan-out is the step where one typed question is silently expanded into many related questions before an AI writes its answer.
Across the set, 12.8% of generated prompts mentioned the primary brand or one of its two competitors. The distribution is the finding: two of the ten fan-out types accounted for 97% of all brand mentions — entity questions and comparison questions.
Meyers also names the mechanism behind the bias. When a brand appears in the prompt itself, the odds the model repeats it in the answer are close to 100%. A prompt list is therefore less like a thermometer and more like a ballot: whoever writes the questions has already narrowed who can win.
Take the limits seriously. These prompts were machine-generated by Gemini, not observed from real people, they averaged 8.2 words apiece, and relevance was scored by cosine similarity — a mathematical measure of how close two phrasings are — averaging 0.67 on a zero-to-one scale. This is evidence about how the measurement behaves, not proof of what your clients type.
What to do about it
If you are paying for an AI visibility report, ask the vendor for the prompt list behind the score. If most of those prompts are comparisons and brand-name questions, the number describes a narrow slice of how people ask — useful, but not a verdict. Ask what a problem-shaped question returns instead, the kind a worried person types at eleven at night.
Source
Moz, "What 50k Query Fan-Outs Reveal About Brands (+ Free Data!)," August 6, 2026 —
moz.com
03 The platform change
Google Maps grew hands this week
The local directory stopped listing options and started completing the transaction.
Google added agentic capabilities to Ask Maps, the conversational layer inside Google Maps. Ask it to order takeout on the way home and it finds open restaurants along your route and adds the dish to a cart on Square or Toast, with Uber Eats listed as coming soon.
“Agentic” here simply means the software takes multi-step actions on your behalf instead of handing you links. The same release compares live hotel prices and checks availability, surfaces local events with ticket links, and adds a transit board that updates minute by minute.
Two details matter more to a service business than the food. Personal Intelligence can connect Ask Maps to your Gmail for context about existing plans, off by default, with Calendar signalled next. And anyone can now suggest edits to a business listing conversationally, including uploading a photo of a storefront sign so Maps reads the new hours off the image.
The rollout is uneven and worth tracking rather than reacting to. Ordering, hotels, events and conversational edits are launching in the U.S. first, while the transit board and personalization ship wherever Ask Maps already runs, now including Australia, Brazil, Canada, Indonesia, Japan and Mexico. A directory that can act is a directory whose accuracy has consequences.
What to do about it
Pull up your Google Business Profile and check the three fields an assistant would act on: hours, service area, and whether the listed number reaches a human during those hours. Turn on notifications for suggested edits so a stranger’s photo of your door cannot quietly rewrite your hours. Put one named person in charge of reviewing every suggested change before it stands.
Source
Google, "Ask Maps gets more helpful with food ordering and more," August 6, 2026 —
blog.google
04 The deadline
On August 17 your intake form starts reporting
OpenAI switches on automatic form matching for existing advertising pixels unless you opt out first.
OpenAI’s conversion measurement documentation states that automatic advanced matching will be enabled for existing web pixels on August 17, 2026, which is eight days from today. It is already the default on any pixel created new.
A pixel is the small piece of tracking code an ad platform asks you to install so it can tell which ads produced which actions. With automatic advanced matching on, the OpenAI pixel detects supported customer information typed into your website forms, then normalizes and hashes it with SHA-256 in the browser before attaching it to a conversion event.
OpenAI states that raw customer information is not sent to OpenAI, and hashing before transmission is meaningfully better than sending plain text. That is a statement about transport, not about permission. Whether intake data may be used for ad measurement at all is governed by your engagement terms, your privacy notice and your state bar’s advertising rules, which apply to this channel exactly as they apply to a billboard.
The opt-out is short and reversible. Tools, then Conversions, then Data Source, then Edit pixel turns it off, and you can re-enable it later. The default is the risk here, not the feature.
Do this before August 17
If an OpenAI advertising pixel is on your site, decide before August 17 whether hashed intake-form data may be used for ad measurement, and write the decision down with the name of whoever approved it. If you are not certain a pixel is installed, ask whoever manages the site today rather than after the date. Route the question through the same review you would use for any other client-data decision.
Source
OpenAI Help Center, "Conversion measurement — automatic advanced matching," Updated August 2026 —
help.openai.com
· Secondary: Search Engine Roundtable, Aug 7
05 The money
Google now decides who may buy attention
Impression limits rather than ad disapprovals, which makes the failure mode silent.
Google published a policy update on August 5 expanding its Limited Ad Serving policy to cover all Google Ads. Rollout begins in August and completes by 2028.
The mechanism is the part to understand. Google limits impressions rather than disapproving ads, so an unqualified advertiser sees no red banner on any individual ad, only less delivery. Qualification is assessed on account maturity, advertiser verification status, policy compliance history, user reports, ad format usage and industry.
For an established firm with a verified account this changes little. For a new firm, a rebrand, or a practice that has just moved its advertising into a fresh account, it is a cold-start tax — and because nothing is rejected, it is easily mistaken for weak creative or a soft market.
Google’s stated remedies are unglamorous and specific. Complete advertiser verification, avoid generic ad copy, keep branding consistent between the ad and the landing page, and pin your domain to the first headline position if your brand is new or less well known. Limited accounts receive an in-account notification and may appeal.
What to do about it
Check advertiser verification status in Google Ads this week and finish it if it is incomplete. Then read your top three search ads as a stranger would: does the firm name appear before the offer, on both the ad and the page it lands on? Anything you change there still has to satisfy your state’s rules on comparative claims, results and required disclaimers.
Source
Google Advertising Policies Help, "Update to Limited Ad Serving Policy (August 2026)," Posted August 5, 2026 —
support.google.com
· Secondary: Search Engine Land, Aug 7
