01 The market data
Traffic fell a fifth, revenue barely noticed
A publisher lost 22% of its sessions last quarter and 1% of the money those sessions made.
People Inc, the publisher behind People and Better Homes & Gardens, says it is not blocking Google’s crawlers yet, for a mechanical reason. Google’s AI crawling cannot be separated from its search crawling, so switching off one switches off both.
The numbers behind that patience are the real story. Google now accounts for about 21% of all traffic to the company’s 19 core brands, down from roughly two-thirds at its peak. Sessions across those brands fell 22% last quarter — and session-based revenue fell 1%.
A 22% drop in visitors that costs 1% of the revenue those visitors produced is not a rounding error in the traffic. It is a rounding error in the dependence. 43% of the company’s digital revenue now comes from something other than website sessions — events, subscriptions, licensing — and licensing alone rose 23%.
Chief executive Neil Vogel’s framing is worth borrowing verbatim: “we are nearly out the other side of search being a material driver of value for us. But we’re not there yet.” Total revenue was down 2% year on year to $416.7m, with digital up 6% and print down 16%. The shrinking part is the old business, not the audience.
What to do about it
Pull your last twelve months of new matters and mark each one with where it actually began — a search, a referral, a directory, a past client. Then work out what share started with a Google click. That single percentage is your exposure, and a more useful number to move this year than any ranking.
Source
Press Gazette, "People Inc not blocking Google ‘at the moment’ as it rolls out digital subscriptions," August 5, 2026 —
pressgazette.co.uk
02 The research
Most local businesses do not exist to AI
A complete market census found that being good had almost nothing to do with being mentioned.
A paper posted to the arXiv preprint server on 7 August did something most studies cannot afford to do: instead of sampling businesses, it counted all of them. The author enumerated 4,776 cafés, restaurants and bars across two bounded markets in Bali, then collected 2,208 search-grounded answers from ChatGPT, Claude, Gemini and Perplexity across 96 persona-conditioned questions, over seven days, under a pre-registered protocol.
Sampling tells you who showed up; a census tells you who was never asked. 85.6% of the venues were never recommended by any system. Among venues with fifty or more ratings — established, easily findable places — 72.6% still never appeared once.
What predicted getting into an answer at all was documentation, not quality. Having your own website raised the odds 1.92 times; review volume 1.64; publicly listed price information 1.54; mentions on third-party sites 1.44. Star rating had no measurable effect on whether a business appeared — though it did predict finishing first once a business was already inside the set.
Two secondary findings deserve as much attention. Invention was rare, at 0.08% of mentions, but the systems recommended permanently closed venues 93 times, which the author summarises as staleness rather than hallucination being the practical failure mode. Agreement between the four assistants was also low, so “visibility in AI” is four separate questions.
Treat the result as suggestive rather than settled. The study covers two tourist markets in Bali, the odds ratios are observational, not causal, and the author notes that movement between runs is sampling noise. What travels beyond Bali is the mechanism: these systems select from what is documented.
What to do about it
Audit your documentation before you audit your rankings. Three questions: do you own a website that states your practice areas plainly, is your fee structure described in public language, and are you mentioned on sites you do not control? Then check every listing you can find for a closed office or an old address, because these systems recommended dead businesses far more often than they invented live ones.
Source
arXiv (preprint, not yet peer-reviewed), "Invisible to the Machine: Auditing AI Restaurant, Cafe, and Bar Recommendation Against a Complete Market Census," submitted August 7, 2026 —
arxiv.org
03 The platform change
The default model changed under your clients
OpenAI swapped the model most free users get, and benchmarked the change on legal prompts.
On Thursday OpenAI updated GPT-5.6 Sol for paying subscribers and, in the same announcement, made GPT-5.6 Luna the default model for free users — the tier most of your prospective clients are on. Nobody has to opt in. The machine answering “do I have a case” simply changes underneath them.
The accuracy claim is unusually on-topic for this audience. In an internal evaluation of financial, medical and legal prompts requiring factual detail, OpenAI reports that responses containing at least one factual error were about 62% less common with Luna and 68% less common with Sol than with the previous model.
Read the qualifier before you read the number. This is an internal evaluation, with no published sample size, no released prompt set and no confidence interval — a vendor grading its own homework and then reporting the grade. The direction is plausible; the magnitude is unaudited.
The practical consequence has little to do with accuracy anyway. A default-model change alters which sources get retrieved and cited, and it arrives without notice to the people using it. If you track how AI assistants describe your firm, movement this month may be the model moving rather than anything you did.
What to do about it
Re-run the five questions a prospective client would actually type — the “how much does this cost,” “how long does this take,” “do I need a lawyer for this” questions — in a logged-out session this week, and save the answers with today’s date on them. You want a reading taken after the model changed, not one from July, or you will spend the autumn comparing against a baseline that no longer exists.
Source
OpenAI, "Improving GPT‑5.6 Sol in ChatGPT—and expanding access to GPT‑5.6 Luna for free users," August 6, 2026 —
openai.com
04 The deadline
Google will want a government’s permission slip
From October 5, advertising help with government documents requires proof most firms cannot produce.
Google posted a policy update on 4 August that takes effect on October 5, 2026. From that date, only an “authorized provider” may run ads promoting the direct acquisition of government documents and services. This is less an advertising policy than a licensing board with a start date.
To qualify, your domain must be linked from an official government website and explicitly named there as authorized to provide that specific document or service. Google also lists what will not count: commercial contracts, business and trade licenses, entries on business registries, and articles or blog posts — even ones hosted on a government site.
Two further conditions are easy to skim past. Your domain must promote only the specific document or service the authorization actually covers, and your ads must target only the geographic region that authorization covers, with a carve-out for inherently cross-border categories such as electronic travel authorizations.
For an immigration practice this is the item of the week. A firm advertising help with visa applications, work permits or naturalization paperwork sits close to this policy even if it thinks of itself as a law firm rather than a document service, and Google reserves the right to deny or revoke certification at any time. Your state bar’s advertising rules apply to this channel exactly as they do to a billboard — comparative claims, results and required disclaimers get no platform exemption.
Do this before October 5
If any live campaign of yours mentions obtaining, filing, renewing or expediting a government document, list those campaigns this week and hold each one against the new definition. Then ask the question Google will ask: is there a page on a government website that names your domain as authorized? If the answer is no, plan the rewrite now rather than meeting the disapproval in October — and route the rewrite through whoever handles your firm’s advertising compliance review before it goes live.
Source
Google Advertising Policies Help, "Update to Other restricted businesses: Government documents and services policy (October 2026)," posted August 4, 2026 —
support.google.com
· Secondary: Search Engine Land, Anu Adegbola, Aug 5
05 The money
The lead-buying rulebook was rewritten quietly
Local Services Ads now carries named penalties, an image ban already in force, and an exit from manual bidding.
Google renamed its Local Services rulebook from “platform policies” to “Local Services Ads requirements”, and the rename came with substance. The document now sets out tiered consequences — email warnings, impression limits, then temporary or permanent account suspension — alongside concrete violation examples such as changing prices on arrival, asking for reviews on site, and ignoring customer disputes.
Sitting in the same policy set is a rule whose date has already passed. Since August 2, AI-edited and AI-generated images have been disallowed in Local Services Ads in the European Union and in New York state, and you attest to compliance simply by uploading an image.
That lands on law firms directly, because Local Services Ads carries per-professional headshots and names law among its supported categories. A retouched or AI-cleaned attorney headshot running in New York is now outside policy — as are, listed separately on the same page, images that violate attorney–client privilege.
The structural change underneath is where the money actually moves. Local Services Ads is migrating into Performance Max campaigns with pay-per-lead goals, and manual bidding and industry-level target CPA are being deprecated — trading a manual gearbox for an automatic on the same road. Phase one, in August, covers select US home and storefront trades such as plumbing, HVAC and roofing; legal is not in it, and account administrators get an advance email 14 days before their migration date.
What to do about it
Two jobs this week. If you advertise in New York or the EU, open every image on your Local Services profile and confirm none has been through an AI edit, including background cleanups a designer may have run for you. Then read the requirements page end to end against your own intake practice, because the consequences now escalate in defined steps rather than starting and ending with a warning.
Source
Google Local Services Help, "Local Services Ads requirements," image policy effective August 2, 2026 —
support.google.com
· Secondary: Google Ads — the pay-per-lead migration
