Today's theme — Discovery became a market

The week the answer box grew a cash register

For three years, AI answers were a curiosity. In the past ten days, they picked up the three things that turn a curiosity into a channel: an audience that trusts it, data it can be held to, and a way to charge for it.

Here's the pattern hiding inside today's five stories. A national study found that consumers researching lawyers have quietly stopped double-checking what AI tells them. OpenAI signed a deal to pipe millions of real business reviews into ChatGPT's local answers. Its ad platform added conversion bidding. Google announced it's retiring the dashboard where local service businesses have managed their best leads for years.

And in the middle of all that, a group of researchers published a survey of 45 studies concluding that most of what's being sold as "AI optimization" hasn't actually been proven to work.

Both things are true at once, and that's the whole story. The channel is real and the playbook is young. The businesses that win the next 18 months won't be the ones who bought the loudest tool. They'll be the ones who got the boring fundamentals right while everyone else was chasing tactics that a peer-reviewed survey can't yet defend.

01 The consumer data

Your future clients stopped fact-checking the robot

Google's share of attorney research fell nearly 15 points in a single year — and the people driving it aren't 24-year-olds. They're 52 and affluent.

iLawyer Marketing has run the same national survey for four years: if you needed a lawyer for something serious, where would you look? This year's answer moved more than it ever has — Google fell from 86.7% to 71.9%, the steepest single-year drop in the study's history, while ChatGPT climbed from 28.1% to 41.9%. Yelp fell by nearly half. Bing lost more than half its share. Reddit, Instagram, and Facebook all declined. The only category that grew was answer engines.

But the number that should actually stop you is a different one. Last year, 94% of people who used ChatGPT said they'd also check Google to validate the answer. This year, that fell to just over 70%.

Think about what that means in human terms. It's the difference between a new associate whose memos you read line by line, and one whose memos you skim and sign. Nothing about the associate changed overnight. Your relationship to their output did. As the study's author put it, the story isn't that consumers discovered AI — it's that they stopped double-checking it.

And the demographic finding upends the usual assumption. The heaviest AI users for legal research were consumers aged 45 to 60 in households earning over $125,000 — which is to say, the exact people behind the highest-value matters most firms are trying to win.

What to do about it

Don't read this as "abandon SEO." Read it as the opposite. The same researchers note that answer engines don't invent recommendations from nothing — they lean heavily on web search signals, and firms ranking on page one of Google are significantly more likely to be named by every major answer engine.

Your Google ranking is now doing two jobs: earning the click from the 72% who still search, and serving as the credential that gets you named to the 50% who ask an assistant instead. Pull back on one and you lose both.

Source iLawyer Marketing, "What Online Sources Do People Use to Research and Find Attorneys in 2026?" (n=1,110 U.S. adults) — ilawyermarketing.com · Announcement release: openPR

02 The research

The most useful paper in AI search says most of AI search is unproven

A review of 45 studies finds the evidence for "AI optimization" is real, narrow, and much smaller than the marketing around it.

On July 15, a critical survey landed on arXiv reviewing nearly three years of Generative Engine Optimization research. It examined 45 studies published between November 2023 and July 2026, and it argues that GEO isn't one ranking problem at all — it's a chain of separate stages: whether search gets triggered, whether you're crawled, whether you're retrieved, whether you're reranked into the context window, whether you're cited, whether you're absorbed accurately, and whether anyone acts on it.

Picture a relay race where you can only see the finish line. If your team loses, the tools on the market will confidently tell you the last runner was slow — but the baton might have been dropped four legs back, before anyone ever saw you. That's the measurement problem the paper is describing.

Its central finding is bracing, and worth memorizing before your next vendor call. The famous gains from the original 2023 GEO paper hold only within its experimental setup, where a source was already present in a fixed context — they establish neither organic discoverability nor durable traffic effects. Across the whole body of work, the authors conclude that already-retrieved content can causally change how it's cited or used, but no reviewed technique shows stable, longitudinal, cross-platform results. Commercial audits also show low overlap between which sources different engines pick, and substantial run-to-run variability.

This is not a reason to sit out. It's a reason to be precise about what you're buying.

What to do about it

Trust the boring stuff. The things the evidence does support are unglamorous: being crawlable, being retrievable, and putting extractable, well-structured evidence on the page.
Distrust single-run screenshots. If a report shows one query on one day, it's a coin flip presented as a trend. Ask for repeated measurements across paraphrased prompts.
Ask the honest question. "Which stage of the pipeline does this tactic act on?" If nobody can answer, you're buying decoration.

Source "Optimizing Visibility in Generative Engines: A Critical Survey of Generative Engine Optimization (2023–2026)," arXiv:2607.14035, July 2026 — arxiv.org/abs/2607.14035

03 The data deal

ChatGPT just bought a library card for local reviews

Yelp is licensing its ratings, photos, and business data to OpenAI — and "Request a Quote" is coming inside the chat window.

Yelp is licensing reviews, photos, and business information to OpenAI, giving ChatGPT a real-time source of local recommendation data. Yelp's branding and links will appear when its content is used, though OpenAI controls how that's presented — and Yelp's "Request a Quote" feature is being added, letting people contact local service providers directly through the assistant.

For two years, the knock on AI assistants was that they were brilliant about the world and hazy about your neighborhood. This is the fix. The model stopped guessing about local businesses and took out a subscription to the local paper of record.

Notice what that changes structurally. Your visibility is no longer only a function of what's on your own website. It's a function of what a licensed third-party database says about you — and whether that database has your name, address, practice areas, and reviews right. Reviews become dual-purpose: they shape what a prospect thinks, and they shape how an assistant describes you when asked to recommend someone.

The deal is non-exclusive, which is the real signal: this is a template, not a one-off. Expect more licensed data pipes, not fewer.

What to do about it

Audit every third-party profile you have — Yelp, Google Business Profile, bar directories, Avvo, Martindale — the same way you'd audit a filing. Wrong suite number, a stale practice area, a partner who left in 2023: those aren't cosmetic anymore. They're now training material for the description an assistant reads aloud to your next client.

And if "Request a Quote" arrives inside ChatGPT, treat it like a new intake channel with a real response-time clock attached, not a novelty.

Source Axios (exclusive), "Yelp partners with ChatGPT to surface reviews," July 23, 2026 — axios.com · Confirmation and analysis: Search Engine Land

04 The platform change

Google is closing the dashboard your best leads live in

Local Services Ads move into Google Ads starting this month. Your historical performance reports do not come with them.

Google is moving Local Services Ads into Google Ads as Performance Max campaigns built for pay-per-lead goals, with bidding, reporting, and account-management changes beginning in August 2026. The standalone LSA dashboard is going away.

Most of what matters commercially survives the move. Pay-per-lead billing, Search-and-Maps-only placement, and keywordless targeting all carry over, and despite the Performance Max name, these campaigns won't spill out into Display, YouTube, or Gmail the way a standard PMax campaign would. Your Google Screened badge, lead history, and reviews transfer automatically.

The mechanics underneath do change. Manual bidding and vertical-level cost targets are being retired in favor of a single campaign-level target, weekly budgets convert to a daily average, and Better Business Bureau callouts are being removed.

But here is the line to act on. Lead histories transfer; historical performance reports do not. Once your account moves, the original dashboard becomes inaccessible. That's a filing cabinet being emptied on a date you don't choose — account administrators get 14 days' notice, then a reminder at seven.

The first wave in August covers U.S. home and storefront services; professional-service categories including legal are expected in the later 2026 waves, with international accounts in 2027. Later doesn't mean safer. It means you have a window most advertisers won't get.

Do this before your migration email arrives

Export every LSA performance report you'd want for a year-over-year comparison — by month, by service category, by lead type. Do it this week, not on notice day.
Write down your current cost-per-lead by practice area on paper. After migration, automated bidding replaces your manual controls, and you'll want a "before" number to argue with.
Expect up to two weeks of unstable performance after the switch and don't make staffing or spend decisions off that window.

Source Google Ads Help, "Local Services Ads transition to Performance Max campaigns with pay-per-lead goals" — support.google.com · Coverage: Search Engine Journal · Law-firm specific: Justia Onward

05 The money

The answer box now has a bid button

ChatGPT Ads added conversion bidding, budget pacing, geographic exclusions, and a bulk API — the plumbing that turns an experiment into a channel.

On July 24, Search Engine Land reported a batch of ChatGPT Ads updates that, individually, sound like release notes. Advertisers can now create optimized cost-per-click campaigns by selecting a Conversions objective, with delivery automatically steering toward clicks more likely to convert while still billing per click. Daily budgets are shifting to a rolling seven-day average model, so spend can fluctuate day to day within overall limits. The release also added geographic exclusions, AppsFlyer and Adjust integrations, advanced matching for conversion measurement, and an asynchronous bulk API.

Together they mean something bigger than any one feature. An ad platform without conversion optimization is a billboard: you rent attention and hope. An ad platform with it is a direct-response machine. That's the line ChatGPT Ads just crossed, roughly six months after ads first appeared inside the product.

For any local service business, this is the moment the assistant your clients are already trusting became a place your competitors can buy their way into the frame.

What to do about it — and one caution

If you test it, start with a small budget and a single, well-instrumented conversion event. Conversion bidding only performs as well as the signal you feed it, so wire up server-side tracking and let the campaign learn before you judge it.

For regulated professions: an ad inside a chat window is still an advertisement. State bar advertising rules on comparative claims, results, testimonials, and required disclaimers apply to the ad copy and the landing page exactly as they would on a billboard. Route it through the same review workflow. A new channel is not a new rulebook.

Source Search Engine Land, "ChatGPT Ads adds conversion bidding, geo exclusions and bulk campaign tools," July 24, 2026 — searchengineland.com · OpenAI documentation: Conversion-optimized campaigns