01 The market data
Nine in ten marketers use AI and one in seven sees results
Orbit Media’s twelve-year blogging survey finds AI adoption at 92.4% and strong results at 14%, with no relationship between the two.
Orbit Media has surveyed content marketers every year since 2014, and this year’s edition, reported by Search Engine Land on Thursday, is the bleakest in the series. Just 14% of 1,042 marketers said their blogs delivered “strong results,” which is six points below the previous twelve-year low and nearly half the 26% reported in 2022.
The obvious suspect is the tool everyone adopted. AI use reached 92.4% this year, and it did make the work faster: the average post now takes 3 hours and 20 minutes to write, down from more than four hours in 2022, which Orbit estimates saves the typical marketer about 50 hours a year. But using AI showed no relationship with stronger results, in either direction. AI made the kitchen faster. It did not change the recipe.
What the survey does link to results is the work people stopped doing. Marketers who regularly worked with subject-matter experts were 2.6 times more likely than the benchmark to report strong results, yet the share doing so fell from 25% in 2017 to 7% this year. Keyword research, original research, formal human editing and regular use of analytics followed the same pattern: still tied to results, less common than a year ago.
One more finding matters for a firm that has been told traffic is dead. Marketers who measured qualified leads and deals rather than visits were more likely to report strong results, and keyword research still predicted success even as fewer people bothered with it. Ann Handley of MarketingProfs called the report “a wake-up call” about being far more discerning over which efforts get cut. Two cautions: the results are self-reported, and Orbit sells web design and content services, so read the direction rather than the decimals.
What to do about it
If your firm publishes articles, ask who reviewed the last five before they went out, and whether a lawyer or partner with the actual expertise touched them. The survey’s strongest predictor of results is the expert in the room, and it is the step most firms cut first once the drafting got cheap.
Keep the human edit as a documented step, not a courtesy. It is what the data says works, and for attorney advertising it is also what the bar rules already expect of anything that goes out under your name.
Source
Search Engine Land, “Content marketing success falls to 12-year low: Survey,” by Danny Goodwin, September 10, 2026, reporting Orbit Media’s 2026 Blogging Statistics survey (1,042 content marketers, self-reported results) · searchengineland.com
02 The evidence
The visibility score you were sold measures the vendor’s questions
A new arXiv paper argues that an AI-visibility score is defined by which prompts were chosen and how they were weighted, and that a citation alone proves nothing about contribution.
An “AI visibility score” is the number every agency in this field now sells, Forever Cited included: how often your firm appears, is cited or is named in generated answers. A paper posted to arXiv on Sunday by Olivier Martinez takes that number apart. “The prompt corpus selects the situations evaluated, while weights determine their relative importance. Together they define an ‘answer market’ that need not represent actual user demand.”
In plain terms, whoever writes the test questions decides what the score measures. Ask a hundred questions about personal injury in Phoenix and you have a score for that; ask a different hundred and the same firm scores differently, with no change to the firm. Prompt wording alone “can alter retrieval, competing sources, and generated answers.” A visibility score is a poll, and the vendor wrote the questions and picked the neighbourhood.
The paper’s second point is sharper. Most tools use a language model to decide whether an answer mentions you, and “its instruction can change the score assigned to an unchanged answer.” The same answer, the same firm, a different score, because the grader was told to look differently. And on the number everyone reports: “A citation alone does not establish a source’s contribution.” Being footnoted is not the same as having shaped the answer.
What Martinez proposes is disclosure rather than despair: a framework that states the situations tested, the exact prompt wording, the execution conditions, the weights and the scoring rule, and reports a range of admissible scores when the weights are a matter of choice. It is a 16-page critical survey with no new experiments, and it says so, so treat it as an argument rather than a finding. The argument holds either way: a visibility number without its question list is a number without a denominator.
What to do about it
Ask any vendor reporting your AI visibility for three things: the full list of questions they ask, how those questions were chosen, and whether the number would move if a different person wrote them. A vendor who cannot produce the list is reporting a market they made up.
When you see your own number, read it as a range over a fixed set of questions rather than a point. Movement inside the same question set means something. A comparison across two different sets means nothing.
Source
arXiv, “Measuring GEO Visibility: Prompt Corpora Define the Answer Market,” Olivier Martinez, submitted September 6, 2026 (16-page critical survey, no new experiments) · arxiv.org
03 The structure
Your plain keyword ad can now show up inside the AI answer
Google confirms a test that lets exact and phrase match Search campaigns serve text ads in AI Mode, a placement that until now required its automated products.
Until this month there was one way to buy an ad inside Google’s AI Mode: hand the keywords to Google’s automation. AI Max, Performance Max or broad match with Smart Bidding were the only routes in, which meant an advertiser who wanted to appear in the conversational answer had to give up control over which searches triggered the ad. On September 4 that changed, at least as a test.
Google Ads Liaison Ginny Marvin confirmed it after advertiser Anthony Higman noticed his campaigns appearing there. “Exact and phrase match keywords are eligible to serve text ads in AI Mode,” she wrote, adding that it is “a small experiment that recently kicked off” and is “limited to cases where there is explicit and direct user intent.” Higman says he confirmed the behaviour “through multiple campaigns and tests.”
For a professional firm this is the part of AI Mode that was missing. A firm’s search budget usually lives on a short list of exact phrases, “divorce lawyer” plus a city, because broad match and automation spend it on searches nobody meant. Those tightly held campaigns were shut out of the AI answer by design. Now they are let in when the searcher’s intent is plain. The old keyword ad got a seat at the new table, but only when the guest states their order clearly.
Two cautions. Google has named no countries, no end date and no performance data, and it still positions AI Max and Performance Max as the products for conversational searches and for the newer formats such as Highlighted Answers. And the ad rules did not change with the placement: an ad served inside an AI answer is attorney advertising in every state that regulates it, with the same disclaimer, comparison and testimonial rules as a billboard.
What to do about it
Ask whoever runs your Google Ads to check whether any of your exact and phrase match campaigns are showing AI Mode impressions yet, and to report that traffic’s cost per click and lead quality separately from the rest. Small numbers at first, but they are the first direct measure of what a plain keyword is worth inside the answer.
Do not loosen your match types to chase this. The test only serves on explicit intent, which is exactly where exact match already lives, and the compliance review your ads went through still covers them here.
Source
Search Engine Roundtable, “Google Ads Testing Serving Restrictive Match Types In AI Mode,” by Barry Schwartz, September 4, 2026 · seroundtable.com · Search Engine Land, “Google tests traditional Search campaigns in AI Mode,” by Anu Adegbola, September 7, 2026 · searchengineland.com
04 The one to check today
Google can lose faith in your whole site, not just the bad pages
John Mueller says mass-produced pages can cost a site Google’s trust as a whole, and that getting it back “tends to take time & significant effort.”
Every firm has been pitched the package: a page for every practice area in every town you serve, generated in bulk, hundreds of them, live by Friday. The trade calls it programmatic SEO. On Bluesky this week Google’s John Mueller described where it usually ends: “It’s easy to spin something up with many pages, it’s hard to provide real value to users.”
The sentence that should stop you comes next. “Our systems have possibly lost faith in your site providing good value to users based on the old pages.” Not the pages: the site. Mueller says the tactic often produces something that is “either spam, borderline spam, or low quality,” and the judgment attaches to the domain as a whole.
That is why deleting the pages is not the end of it. “Resolving this tends to take time & significant effort to show the value,” he wrote. The bulk pages can be gone in an afternoon. The lost trust stays for months. A site is judged like a firm rather than like a page: one associate’s careless work ends up on everyone’s letterhead.
This is a Bluesky reply, not a policy document, and Mueller says “possibly” rather than “certainly.” It still explains a pattern many firms have met: the good pages stopped ranking after the cheap ones arrived. Every AI answer engine that draws on Google’s index inherits the same judgment.
Do this today
Search Google for site:yourfirm.com and count the pages. If the number is far larger than the pages you would put your name to, find out who built the rest, what they say, and whether a lawyer ever read them. City-by-city pages that differ only in the place name are the usual culprit.
Do not mass-delete before you have the list. Decide page by page what gets rewritten by someone who knows the subject, what gets consolidated, and what gets removed, and keep a record of the decision. That record is the “significant effort” Mueller is describing, and it is also your supervision file if the pages were AI-drafted.
Source
Search Engine Roundtable, “Google Can Lose Faith In Sites Based On Low Value Programmatic SEO Pages,” by Barry Schwartz, September 7, 2026, quoting John Mueller of Google on Bluesky · seroundtable.com
05 The money
Amazon now sells the ads under ChatGPT’s answers
A U.S. pilot lets advertisers buy ChatGPT placements through Amazon’s ad platform, in the same week OpenAI quietly barred rival AI tools from advertising there.
Two stories on Thursday, both reported by Search Engine Land, describe the same thing from opposite ends: ChatGPT’s answer page becoming ordinary ad inventory. First, Amazon is letting advertisers extend campaigns from its ad-buying platform, Amazon DSP, into ChatGPT, as a managed service, U.S.-only at launch, with Delta Vacations among the first participants. Ads are bought per click or per thousand impressions and appear as labelled text and image units beneath the answer.
The division of labour is the interesting part. Amazon handles the buying and campaign management; OpenAI keeps control of where and whether an ad is served. Advertisers get aggregated impressions, clicks, cost per result, CPM and CPC, and Amazon can bring its shopping and streaming data to the targeting. ChatGPT’s answer page is shelf space, and Amazon just became the wholesaler. The store owner still decides who is allowed in the door.
The second story shows the store owner using that power. The Information reported, and Adobe confirmed, that OpenAI has updated its advertising policies so that competing image and audio generation tools can no longer run campaigns inside ChatGPT. Adobe’s Doug Wyatt said his team was told “campaigns for standalone image and voice generation products would no longer be approved.” Video tools are still allowed. The change was passed to advertising partners directly rather than announced.
Neither story carries a number about scale, and the Amazon pilot is a “select group” of U.S. advertisers, so this is a direction, not yet a market. The direction is clear enough. One of the largest advertising platforms anywhere is now selling the space under an AI answer, and the company that owns the answer has already shown it will exclude a category when that category competes with it. For a firm the question is what happens when the excluded category is not image tools but a directory, or a rival, that sells the same advice the assistant gives.
What to do about it
Nothing to buy yet: the pilot is closed and aimed at consumer brands. What you can do is get the measurement in place. Add ChatGPT and the other assistants as named referral sources in your analytics and on your intake form’s “how did you hear about us” list, so that when paid placement does reach professional services you know what the unpaid answer was already sending.
If an agency proposes ads in any AI assistant, apply the same rule as for Google: state bar advertising rules follow the ad into the new channel, and the sponsored label belongs to Amazon or OpenAI rather than to you, so your required disclaimers still have to be in the copy itself.
Source
Search Engine Land, “Amazon pilots ChatGPT Ads through its DSP,” by Anu Adegbola, September 10, 2026 · searchengineland.com · Search Engine Land, “OpenAI reportedly blocks rival AI tools from advertising in ChatGPT,” by Anu Adegbola, September 10, 2026, reporting The Information · searchengineland.com
