01 The consumer data
The ad paid for the slot, not the answer
A sponsored ad and Google’s own summary named different companies on the same screen.
A search strategist ran an ordinary buying query — which plumber to call for a broken pipe — and screenshotted the result. At the top sat a sponsored ad with an $89 promotion, 24/7 availability, and a 4.9-star rating across more than 20,000 reviews.
Directly underneath it sat an AI Overview: the summary Google writes itself and places above the ordinary results. It named two entirely different companies, with no hedging language at all, and the advertiser paying for the position above it was never mentioned.
The same pattern showed up in a second test on a shopping query. Two of the three brands the summary recommended had no paid presence whatsoever, while the brand spending most aggressively for the top ad slot went unnamed in the summary.
You rented the billboard and the city hung a road sign under it pointing somewhere else. The ad reads as an offer and the summary reads as a verdict — and a verdict printed under an offer quietly undercuts it.
One caution: these are documented examples, not a measured sample. They prove the failure mode exists, not how often it hits you.
What to do about it
Open a private browsing window and run the three queries you pay the most for. Screenshot the ad and the summary beneath it together, and note whether they agree.
If the summary names competitors, your ad is arguing with Google’s own answer, and the fix is a better-documented practice-area page rather than a higher bid. State bar advertising rules govern the ad copy you write and give you no control over the summary beneath it — one more reason to keep comparative claims out of the ad.
Source
Search Engine Land (Sarah Stemen), “What happens when AI Overviews contradict paid search ads?,” August 5, 2026 —
searchengineland.com
02 The research
Cited almost anywhere, recommended almost nowhere
Six months of data separate being used as a source from being named as a choice.
Kevin Indig examined Semrush visibility data covering 1,094 US categories, five phrasings of the question in each, from January through June 2026. The breadth tests alone rest on 283,215 citations and 76,493 named-brand mentions.
Two things are counted, and the difference is the whole story. A citation is a link the model leaned on as a source; a mention is the model saying your name as the answer.
In categories far from a brand’s core expertise, 50% of appearances were citations and only 25% were mentions by name. In close categories, 74% were cited and 44% were named — while citation-only presence barely moved at all, 41% far versus 40% close.
Depth is what shifts the second number. Appearing in many categories carried no penalty for citations, but being named across many only helped when a brand surfaced in all five phrasings. Shallow presence spread across many topics is what costs you, not breadth itself.
Being cited is being quoted in a news story; being recommended is being the person the reporter calls first. Indig is careful about how far to push it, and so are we: these are associations after controls, not proof of cause, and he notes the effects are light enough that brand reputation may outweigh topical focus.
What to do about it
Pick your single most profitable practice area. Write down the five ways a prospective client would actually phrase that question out loud — the sentence, not the keyword.
Then check whether one page answers each of the five plainly, with a date, a jurisdiction, and a named attorney. That is what depth looks like to a machine, and it is a morning’s work rather than a content programme.
Source
Growth Memo (Kevin Indig), “Does topical focus make your brand more visible?,” August 3, 2026 · Semrush AI Visibility Toolkit, US ChatGPT data, Jan–Jun 2026 —
growth-memo.com
· Secondary: Full version, free to read, at Search Engine Land
03 The platform change
Google asked a searcher to sign in first
A limited test swaps the familiar “prove you’re human” puzzle for a Google account login.
A searcher clicking past the first few pages of results was stopped by a message: sign in to verify you’re a human and see more results. Google has used image and checkbox puzzles for this job for years; asking for an account login appears to be new.
Barry Schwartz, who has covered Google search daily for two decades, reported it and said plainly that he could not reproduce it. It is a limited test, and limited tests often never ship.
For most firms this changes nothing about how a client finds you; almost nobody clicks past the second page. It matters for the reports you receive. Rank-tracking tools read those deeper pages anonymously, and a login wall is a locked door on the stairwell to the back stacks.
If it rolls out, the ranking numbers in your monthly marketing report get harder to collect and harder to trust.
What to do about it
Ask whoever sends you a monthly ranking report one question: where does that data come from? If the answer is a tool that reads results pages automatically, ask what happens if Google puts a login in front of them.
A report you cannot trace is a report you cannot act on.
Source
Search Engine Roundtable (Barry Schwartz), “Google Search Tests Sign In To Verify Searches Are Human (Not Captcha),” August 3, 2026 —
seroundtable.com
· Secondary: Search Engine Land’s account, August 4, 2026
04 The deadline
October 5 changes who may advertise government services
Google will want proof a government authorized you, and a business license will not count.
On October 5, 2026, Google updates its policy on advertising government documents and services. Advertisers will have to demonstrate explicit government authorization for the specific document or service they promote.
The proof standard is narrow. Your domain must be linked from an official government website that names you as an authorized provider — a government-managed directory of vetted partners, or an official regulatory portal.
The list of what will not be accepted is the part to read twice. Commercial contracts, business and trade licences, entries on business registries, and articles or blog posts are all explicitly excluded, even when the registry itself is run by a government.
Two further constraints ride along. You may advertise only the services your authorization covers, and targeting is restricted to the relevant geography, with a carve-out for cross-border items such as electronic travel authorizations and Trusted Traveler programs.
For a firm this lands on immigration and visa work, travel documents, expungement and record sealing, and anything adjacent to filing a government form for someone. Google has moved from asking whether you are a real business to asking who sent you.
Do this before October 5
Pull every ad group that touches a government form, filing, or application, and go find the government page that names your firm as an authorized provider. If none exists, plan now for those ads to stop rather than discovering it on the morning of the fifth.
Your state bar advertising rules apply independently. Google’s approval is not a compliance review, and a documented attorney sign-off belongs in the workflow either way.
Source
Google Advertising Policies Help, “Update to Other restricted businesses: Government documents and services policy (October 2026),” effective October 5, 2026 —
support.google.com
· Secondary: Search Engine Land’s summary (Anu Adegbola), August 5, 2026
05 The money
The click may stop going to your website
OpenAI appears to be building an ad whose destination is a conversation, not a page.
Reporting this week describes an advertising format inside ChatGPT where clicking the ad opens a conversation with a business-specific assistant instead of loading the advertiser’s website.
The setup described has three parts. ChatGPT crawls the business’s own site to build a profile of common customer questions, the advertiser configures an assistant on top of it with product data and lead forms, and campaigns point at the assistant rather than a web address.
Treat all of it as unconfirmed. OpenAI has not announced this, it was spotted inside an ads interface by an outside observer, and nobody has documented what one of these ads looks like to a real user.
It is worth understanding anyway. For thirty years an ad has been a door into your building. This would make it a receptionist standing in somebody else’s lobby, answering in your name.
That carries a specific edge for firms. An assistant answering a prospective client’s questions is an intake channel, and bar rules on advertising, solicitation, and disclosing that the person is not speaking with a lawyer would apply exactly as they do to a phone script.
What to do about it
Do not build for this yet. Do write down the ten questions a prospective client asks before booking a consultation, with a plain answer to each, reviewed by an attorney before it goes anywhere.
That one document feeds an assistant if this ships, sharpens your intake script if it does not, and gets read off your website either way.
Source
Search Engine Land (Anu Adegbola), “OpenAI appears to be building chatbot-native ads that launch AI agents,” July 31, 2026 · reported, not announced by OpenAI —
searchengineland.com