01 The consumer data
AI tripled and search grew anyway
Shopify’s quarterly numbers show AI referrals up 197% and organic search up 12% in the same period.
Shopify published merchant traffic data on August 13 covering the second quarter. AI-referred sessions to merchant storefronts grew 197% year over year. Organic search traffic to those same stores grew 12% on a much larger base.
The mix still favors search. Organic sent more traffic than every AI platform Shopify tracks put together, and Shopify’s chief technology officer, Mikhail Parakhin, wrote that the pie itself got bigger rather than one slice eating another.
The quality difference is the part worth borrowing. In specification-heavy categories, where buyers research before committing, AI-referred shoppers converted at roughly twice the rate of organic visitors, and AI brought in about 1.3 times more first-time customers.
One number points at something you control. Shoppers converted at twice the rate when the AI read Shopify’s structured catalog data rather than scraped third-party feeds. Clean first-party facts beat letting a machine reconstruct you from someone else’s copy.
Read it with the caveat Shopify left in. The company did not disclose how many merchants or transactions the analysis covers, and this is retail rather than professional services. Take the direction and leave the decimal.
What to do about it
Write the plain facts about your practice somewhere you own: practice areas, courts and counties you appear in, languages your staff speaks, how a consultation actually works. Directories reconstruct that badly, and the Shopify number suggests the reconstruction costs you.
Do not read this as permission to stop doing organic search work. On Shopify’s own data, the older channel grew while the newer one tripled.
Source
Shopify, “AI and organic search are doing different jobs: What Shopify’s data shows,” August 2026 · shopify.com · Secondary: Search Engine Land, Danny Goodwin, August 13
02 The research & evidence
The model’s reasons are not its reasons
An audit of seven models found that being listed first was worth $11 a visit, and the models almost never said so.
Syeda Anshrah Gillani and Mirza Samad Ahmed Baig posted a randomized audit to arXiv on August 14. It asked language models to choose one physician from a list of five synthetic profiles, with rating, fee and name randomized independently, and produced 40,068 scored responses across seven models.
The obvious levers behave as you would expect, with a size attached. Raising a profile’s rating from 3.9 to 4.7 lifted its chance of being chosen by 31.4 percentage points. Raising the fee from $90 to $190 cut that chance by 20 points.
Then the finding that should stop you. Being listed first, with nothing else attached, was worth $11 per visit in fee-equivalent terms. Names signaling different genders and ethnicities shifted the outcome too, by 1.3 to 2.9 points, and in the opposite direction from what audits of human decision-makers predict.
None of it showed up in the explanations. The models mentioned gender or ethnicity in at most 0.03% of their stated reasons, and one reasoning model failed the study’s auditability check outright. Ask the system why it chose, and you get a tidy answer with the actual mechanism missing.
The authors are careful about limits and you should be too. The profiles were synthetic and text-only, and the panel was six small open-weight models plus one proprietary model, which the paper says under-represents the assistants patients really use, and this is doctors rather than lawyers. The structure still transfers: a rating and a price move the choice, and so does where a name sits in the list, which no explanation mentions.
What to do about it
Treat your review rating as a mechanism rather than a vanity number. This paper puts a size on it, and asking satisfied clients for reviews through a process your bar rules permit is the cheapest lever in the issue.
Position in a list is not something you or any vendor controls. A tool promising placement in an AI answer is selling something it does not control, and no outcome may be promised to a prospective client in any case.
Source
Gillani, S. A. and Baig, M. S. A., “Whose doctor does the AI recommend? An algorithm audit of reputation and demographic signals in large language model-assisted physician choice,” arXiv:2608.14399, submitted August 14, 2026 · arxiv.org · Secondary: full text
03 The platform shift
Google pushed a spam update this morning
The August 2026 spam update went live today, worldwide, in every language, with several days still to run.
Google announced the August 2026 spam update on its Search Status Dashboard at about 12:30 p.m. Eastern today. It applies globally and to all languages, and Google says the rollout may take a few days to complete.
What it is not is as useful as what it is. This update does not target link spam, and it does not target the site reputation abuse policy. Google announced no new spam policies alongside it, so the rules being enforced are the ones already published.
The recovery timeline is where firms misjudge the stakes. Google’s documentation says a site that fixes a violation may improve only after its automated systems learn over a period of months, and Google runs periodic refreshes rather than reassessing on demand.
For a law firm the exposure is rarely deliberate. It is the forty near-identical city pages a vendor built in 2022, or a blog that quietly filled with generated copy nobody reviewed. Movement you notice this week should be checked against Search Console data from August 18 onward rather than against a hunch.
What to do about it
Open Search Console and compare the seven days ending August 17 against the days that follow. Look at clicks and impressions by page, not just the site total, because a spam update usually moves a section rather than a whole domain.
If an agency built pages for you at scale years ago, read three of them today. If they read as forty copies of one page with the city name swapped, that is the thing to fix, and no shortcut recovers it quickly.
Source
Google, “Released the August 2026 spam update,” Search Status Dashboard, August 18, 2026 · status.search.google.com · Secondary: Search Engine Roundtable, Barry Schwartz, August 18
04 The operational change
A setting your ads rely on is being retired
From September, language targeting stops applying to Google Search campaigns.
Google told advertisers on August 13 that campaign-level language targeting stops applying to Search campaigns starting in September. Search Engine Land reports the change lands late in the month; Google has published the month, not the day.
The replacement is automatic matching. Google will use the language of the ad itself, the language of the search term, and the user’s own language settings to decide what to show. Someone browsing in Spanish who types an English query may get either.
Performance Max shifts with it. Language settings will no longer apply to Performance Max ads on Google Search, though they still guide YouTube, Display, Discover and Gmail, and Shopping ads inside those campaigns are unaffected.
For a firm advertising in a bilingual market this is not a minor preference. If separate English and Spanish campaigns have been kept apart by that setting, Google’s own prioritization now decides which campaign’s ad is the better language match.
Google says no action is required for existing campaigns, which is accurate and incomplete. The developer note is blunter: setting a language criterion on a Search campaign will start returning an error, so anything automated against that field breaks rather than quietly degrades.
What to do about it
Before September, check whether any Search campaign depends on language targeting to stay in its lane. Then make each ad and its landing page unambiguously one language, because the language of the ad is now the primary signal Google reads.
State bar advertising rules follow the ad into whichever language it runs in. Any required disclaimer, disclosure or qualifier has to appear correctly in the Spanish version too, and a machine translation is not a compliance review.
Source
Google Ads Developer Blog, “Google Ads language targeting changes starting September 2026,” August 2026 · ads-developers.googleblog.com · Secondary: Search Engine Land, Anu Adegbola, August 13
05 The compliance line
Your AI notetaker just met the ethics rules
A New York City Bar opinion says the default for recording people who are not your clients should be not to.
The New York City Bar Association’s Committee on Professional Ethics issued Formal Opinion 2026-2 on August 5, publicized on August 17. It covers AI tools that record, transcribe and summarize conversations with people who are not clients of the firm.
The holding is settled law pointed at new software. Attorneys may not record surreptitiously, must disclose the intent to record, and must get permission from every participant. The committee notes that these tools complicate consent, because the technology may lock people out of the discussion if they decline.
The intake implication is the one to sit with. The committee’s default is not to record absent a specific reason in the specific instance. A recording of a prospective client who never hires you, sitting in a folder the whole firm can open, creates a risk of information sharing and a potential conflict under Rule 1.18.
Competence gets stretched in a way that will surprise people. Rule 1.1 is read to include knowing how to turn the recording function off. If your meeting platform admits a notetaker to every call by default, that setting is now an ethics question rather than an IT preference.
This is a city bar opinion and binds no one outside its advisory reach, but ethics committees read each other and the fact pattern is not local. Check what your own state has said before assuming your rules differ.
What to do about it
Open your meeting platform’s settings today and find out whether an AI notetaker joins automatically. That single default is doing more work than most firms realize.
Then decide in writing whether intake calls get recorded. If they do, put the consent language into the script your staff already reads, and store recordings of declined prospects where the rest of the firm cannot browse them.
Source
New York City Bar Association, Committee on Professional Ethics, “Formal Opinion 2026-2: Ethical Issues Affecting Use of AI to Record, Transcribe, and Summarize Conversations With Persons Who Are Not Clients,” August 5, 2026 · nycbar.org · Secondary: press release, August 17
