01 The consumer data
Confident, and unable to define the job
Six hundred marketers were asked what they are doing about AI search. The gap between their certainty and their understanding is the finding.
Scribewise and Scrunch surveyed 602 US marketing and public relations professionals and published the results this week. 84% said they are confident they can influence how their brand appears in AI answers. 56% said they are unsure how AI search optimization differs from ordinary SEO.
Those two numbers describe the same people. 51% are not sure their current approach is the right one, and 47% do not know what content the AI platforms are drawing on to build their answers.
The measurement habits are thinner than the confidence. 45% test their visibility across AI platforms, but 71% are not tracking share of voice against named competitors and 67% are not looking at AI bot traffic at all. Bot traffic is the plainest signal available: when an AI crawler pulls a page off your site, someone was prompting about your category.
Then the finding that explains why so little of this reaches the person paying for it. 62% say leaders outside marketing still judge them on traditional SEO metrics or web traffic. Among agencies selling this as a service, 71% spend more time explaining it than executing it.
One caveat the coverage mostly skipped. The survey was fielded May 19 to June 2, with a stated margin of error of plus or minus four points at 95% confidence, and it measures what marketers believe about their own work rather than what that work produced.
What to do about it
Before you buy a tool or hire anyone, write down the number you expect it to move and the competitor set you will measure against. Share of voice against three named firms in your practice area is a measurement. “Are we showing up in ChatGPT” is a feeling.
And if your managing partner grades marketing on organic sessions, have that conversation before the strategy, not after the first quarterly report.
Source
Scribewise and Scrunch, “Moving fast, flying blind: 2026 AI search survey,” 602 respondents, fielded May 19–June 2, 2026 · scribewise.com · Reporting: Search Engine Land, August 24, 2026
02 The research
Forty-five studies, and one honest conclusion
A critical survey of the whole generative-engine-optimization literature finds almost nothing that survives being measured twice.
This brief covered this paper on August 12 and is coming back to it on purpose, because it is the frame for everything else on this page. Olivier Martinez posted a critical survey of generative engine optimization to arXiv on July 15, reviewing 45 studies published between November 2023 and July 2026. It is older than every other item here and runs as context.
Generative engine optimization, or GEO, is the practice of trying to get your content quoted inside AI answers. The survey argues it is not a single ranking problem. It runs through a chain of stages, from search activation and crawling through retrieval, reranking, citation and absorption, and most of those stages cannot be observed from outside the system.
On what actually works, the survey is narrow and specific. Topical relevance and where your content sits in the model’s working context are the most reproducible levers. Generic optimization heuristics transfer poorly between platforms, and rewriting a page to sound more citable can make it harder to retrieve in the first place.
The line to carry into your next vendor call: no reviewed technique showed a stable, longitudinal, cross-platform causal effect on organic discoverability. Content already sitting inside the model’s context can be made more likely to get cited. Getting it into that context is the part nobody has shown they can control.
The survey also records what the commercial audits keep finding, which is low overlap between the sources different platforms pick and substantial run-to-run variability. Which brings us to the next item.
What to do about it
Treat any claimed percentage lift in AI citations as conditional until you have seen the protocol behind it. The survey’s own recommended standard is repeated measurements, paraphrased prompts, controls and human validation. Ask for it by name.
The lever the literature does support is the unglamorous one. Be the most topically relevant page on the internet for a specific question a client actually asks you.
Source
Olivier Martinez, “Optimizing Visibility in Generative Engines: A Critical Survey of Generative Engine Optimization (2023-2026),” arXiv:2607.14035, submitted July 15, 2026 · arxiv.org
03 The platform shift
Reddit lost 86% of its citations in four days
The most-cited source on the internet fell off ChatGPT’s list over a weekend, and the firm that measured the fall will not guess at a cause.
Promptwatch, which samples answers from the live interfaces of the major AI platforms, reported that Reddit’s share of ChatGPT Search citations fell 86.4% between August 14 and August 17. Search Engine Land covered the report on August 19.
The before and after is stark. Reddit averaged 3.83% of ChatGPT Search citations from July 18 to August 7. By August 14 the share was under 1%, and it averaged 0.52% through August 17.
There was an earlier step down on August 8, the day Promptwatch says ChatGPT Search changed how it fans a single query out into sub-searches. Google’s products showed nothing comparable. Reddit’s share in AI Overviews drifted from roughly 2.5% in early July to about 2.1% in August, a slope rather than a cliff.
Promptwatch is careful about what its own data can carry. It says the numbers show when the shift happened and not why, calls the findings provisional, and will not rule out a collection problem on its own side. That caution is a reason to take the number seriously.
If you run a law firm, the point here is not Reddit. A source with millions of indexed pages and a commercial data relationship lost most of its citation share over a weekend, with no announcement and no explanation. Your four practice-area pages have less standing than that.
What to do about it
Build a baseline you re-check on a schedule instead of a screenshot you took once. Same prompts, same wording, same day of the week, written down with the date.
And do not run the whole program off one platform’s behaviour. Gemini and ChatGPT cite different things, so a strategy tuned to whichever assistant you personally use is a strategy tuned to a sample of one.
Source
Promptwatch, “Reddit citations are dropping in ChatGPT,” August 2026 · promptwatch.com · Reporting: Search Engine Land, August 19, 2026
04 The operational change
Google’s AI report undercounted you for a week
The Generative AI performance report logged short from August 13. Google has confirmed it. Check your dates before you present a number.
Search Console’s Generative AI performance report shows how often your pages surfaced in AI Overviews, AI Mode and the generative features inside Discover. Site owners began reporting a cliff-edge drop around August 12 and 13. Google confirmed it was a logging error rather than a change in visibility.
The entry Google added to its data anomalies page is unambiguous: a logging error caused a decrease in impressions on the Generative AI performance report in Search for data starting August 13, 2026, and the issue affects data logging only. A second entry covers Discover on the same date.
John Mueller confirmed the problem publicly and said an annotation would appear inside Search Console. Barry Schwartz reported on August 20 that impressions had bounced back, which puts the window of bad data at roughly August 13 through August 19.
The bug is not the interesting part. What matters is that at least one practitioner watching the line fall assumed their AI citations had collapsed and went hunting for a cause on their own site. A fuel gauge reading empty on a full tank does more damage than a gauge reading nothing at all.
Do this before your next report
If you pulled an August figure out of the Generative AI report for a partner meeting or a client update, pull it again. Anything dated August 13 through August 19 is understated.
Annotate the gap wherever you track this, and record where the correction came from. A number you cannot account for six months from now is a number that gets used against your judgment.
Source
Google Search Console Help, “Data anomalies in Search Console,” entry dated August 13, 2026 · support.google.com · Reporting: Search Engine Roundtable, August 17, 2026
05 The money
ChatGPT Ads started counting the clicks nobody made
Automated bidding, platform targeting, and a view-through metric that will make the platform’s own numbers look larger next month.
OpenAI’s advertising platform shipped three changes on Monday. A “Maximize results” bid strategy hands bid decisions to the platform’s own automation, advertisers can now restrict a campaign to the iOS app, the Android app or the web, and the platform expanded into Brazil and Mexico.
The change to watch is the quiet one. ChatGPT Ads Manager now reports one-day view-through conversions at campaign, ad group and individual ad level. A view-through conversion is recorded when somebody converts within a day of seeing an ad without ever clicking it.
Nothing improper is happening. View-through attribution is standard across the ad industry. But the effect is predictable: reported conversions will rise without anything about your campaign changing, because a category of outcome that used to be invisible is now added to the total.
On a small budget that matters at renewal. Ask whether the conversion figure you are being shown is click-through, view-through, or both, and require the two to be reported separately before you compare this month against last.
One more thing for firms under bar advertising rules. A new ad surface is still advertising. Comparative claims, statements about results, testimonials and required disclaimers apply on ChatGPT exactly as they apply on Google, and the copy belongs in the same compliance review as anything else the firm publishes.
What to do about it
Before your next report, split click-through from view-through in whatever dashboard you use. If your agency cannot produce both, that is this week’s question rather than next quarter’s.
Treat the automated bid strategy as an experiment with a hard budget cap rather than a setting you switch on. It optimizes toward the goal you picked, and the goal you picked now counts conversions nobody clicked.
Source
Search Engine Land, “ChatGPT Ads adds automated bidding, platform targeting,” Anu Adegbola, August 24, 2026 · searchengineland.com
