Insights › AI Search Watch
AI Search Watch: The Weekly AEO & GEO Briefing
Every week, this page tracks what changed in answer engine optimization (AEO), generative engine optimization (GEO) and the AI models themselves — and translates it into what it means for your visibility, whether you run a local shop or an enterprise. The rules of being found are being rewritten in public, one model release at a time. Miss a shift, and the citation slots in your market get taken while you are not looking. This briefing exists so that never happens to you.
Why a weekly watch?
- The answer layer is small and moving. ChatGPT cites on average fewer than 7 sources per answer (Zhang, He & Yao 2026, N = 21,143 citations). Who fills those slots changes as models and their retrieval pipelines change.
- Model updates move visibility overnight. Engine behavior differs by platform and language (Chen et al., University of Toronto, 2025) — a new model version can reshuffle who gets cited without any change on your site.
- Clicks keep migrating. Google users click less when an AI summary appears (Pew Research Center, 2025). Every week that trend compounds.
Edition — week of August 10, 2026
AEO / GEO: your own website is 2.9% of the citations
The first large-scale baseline for AI brand visibility has landed. Pratyush Kumar (arXiv:2606.20065) analysed 102,025 prompt responses and 149,912 citations across 102 brands and five engines — ChatGPT, Gemini, Perplexity, Claude and Grok — between March and May 2026. Two numbers reframe the work. First, a brand-stature ladder: global household names appear in 72.9% of relevant answers on their first run, established mid-market brands in 43.6%, niche and small brands in 11.4% — roughly 30 percentage points per step (Kruskal–Wallis H = 38.32, p = 4.8×10−9). Second, and more useful: of all citations recorded, only 2.9% point at the brand’s own domain. About 75% go to corporate and third-party brand pages, and the single most-cited content format is the ranked “best-of” listicle, at 21.0% of all citations. The consequence is uncomfortable for anyone selling on-site optimization alone: engines mostly cite pages about you that you do not own. One more finding worth pinning up — sentiment flipped in 45.5% of runs against 6.8% for mention status, making how you are framed roughly 6.7× noisier than whether you are mentioned at all. Our full analysis →
Models & engines: a quiet week, and the control that is actually landing
No engine shipped a documented change to how it retrieves or cites this week. That is worth saying plainly rather than inflating a vendor blog post into news. The change still working its way into accounts is Google’s, and its origin is regulatory. On 3 June 2026 the UK Competition and Markets Authority imposed a conduct requirement giving publishers, in its words, “effective tools to prevent their content being used to power AI features in search, such as AI Overviews” (UK Government). The same requirement obliges Google to attribute publisher content with clear links inside AI-generated results, and to let publishers opt out of model fine-tuning separately. Google has nine months to implement, with compliance reports every six months in the first year, and the CMA expects important parts of the controls to reach publishers well before that deadline. In Search Console this surfaces as a site-level setting covering AI Overviews, AI Mode and AI Overviews in Discover, rolling out UK-first to a subset of owners; opting out is not used as a ranking signal for ordinary Search. If the setting has reached your property, the decision is live for you now. If it has not, it is coming.
What it means for you this week
Small and mid-sized businesses. If you are a niche or small brand, the baseline says you appear in roughly 11% of relevant answers before you do anything. That is a starting line, not a failure. The highest-leverage move the data supports is not another rewrite of your homepage: ranked listicles carry 21% of all citations and third-party brand pages about 75%, while your own domain carries 2.9%. Audit the “best X for Y” roundups, review sites and community threads in your category, and find out what they say about you before you spend on your own copy. And separate sentiment from mention in whatever you track: with framing flipping in 45.5% of runs, one bad-looking answer is noise, not a crisis.
Large accounts. Two workstreams. Your incumbency is real but conditional: Kumar puts household names at 72.9%, and a companion audit (Chu & Hou, arXiv:2606.17443) finds well-known brands recommended 100% of the time when specifications are identical — a dominance that collapses once a rival holds a rating advantage of less than +0.1 stars. Treat that as a monitoring requirement, not a moat. Second, if you publish into the UK, the CMA controls are a cross-functional decision rather than an SEO one: opting out of the AI surfaces protects content from grounding answers but forfeits the impressions those surfaces generate. That trade belongs to legal, communications and search together.
Edition — week of August 3, 2026
AEO / GEO: the “+40%” everyone quotes does not mean what they think
The first critical survey of the GEO field has landed, and it is blunt about the industry’s favourite number. Martinez (arXiv:2607.14035) reviewed 45 studies published between November 2023 and July 2026. The widely quoted “up to 40% more visibility” traces back to a single measure in Aggarwal et al.: position-adjusted word count rising from 19.3 to 27.2 for quotation addition, roughly +41% in relative terms. It does not mean 40% more clicks, and it does not mean a 40% better chance of being retrieved — it means a source already handed to the generator earned a larger share of the answer text. The survey lists the generalised claim that “GEO increases visibility by 40%” among those it rejects as unsupported. What does replicate is narrower and more useful: topical relevance and position in the model’s context are the most reproducible levers. Its verdict on the market is one line: “claims about GEO return on investment clearly outstrip the academic evidence.” Our full analysis →
Models & engines: the EU AI Act’s transparency rules went live
Since 2 August 2026, the bulk of the EU AI Act applies — including the Article 50 transparency obligations (European Commission). Systems that interact directly with people — chatbots, AI agents, on-site assistants — must make clear to the user that they are talking to an AI, at the latest at the first interaction. AI-generated or manipulated content carries marking and disclosure duties. The scope reaches past Europe: providers established outside the EU are covered when their system’s output is used in the EU. High-risk obligations under Annex III also begin to apply, with Article 6(1) deferred to 2 August 2027. This is not a ranking change, but it is an answer-layer change: if you run a support bot, an AI answer widget, or publish AI-assisted content into an EU market, disclosure moved this week from good practice to legal obligation.
What it means for you this week
Small and mid-sized businesses. Treat any flat “+X% AI visibility” quote as a question: at which stage, measured how, over how many repetitions? Before buying optimization, verify the cheap thing the research actually depends on — that the page is crawlable, indexed and genuinely on-topic for the query you want. Everything these studies measure assumes the engine already has your page. And if you run a chatbot on your site, add the AI disclosure now: it takes minutes and it is now required in the EU.
Large accounts. Two separate workstreams. Compliance should inventory every customer-facing AI system and AI-generated content stream against Article 50, including vendors outside the EU whose output reaches EU users. Marketing and SEO should reset reporting: replace the single “AI visibility score” with citation share per engine, sampled repeatedly, with the denominator shown. The survey’s seven-stage pipeline is the reason — one number cannot tell you whether the problem is crawling, retrieval, context position or citation.
Edition — week of July 27, 2026
AEO / GEO: repair beats rewrite
The strongest recent result in GEO research: targeted, diagnosed repairs lifted citation rates by over 40% while modifying only 5% of content — while blanket optimization touched five times more content for worse results, and could even harm long-tail pages (Tian et al., arXiv:2603.09296). If your agency proposes “rewriting everything for AI,” show them this. Our full analysis →
Models & engines: structure is the safety net
Realistic end-to-end testing (SAGEO Arena, Kim et al., arXiv:2602.12187) shows naive body-text optimization often degrades retrieval — and that structural information (titles, headings, markup) is what mitigates it. For small businesses this is good news: fixing structure is cheap and does not require a content team. Our full analysis →
What it means for you this week
Businesses of every size: before producing new content, verify your existing money pages are structurally clean (title, metas, heading hierarchy, schema) and carry sourced evidence. That is the highest-leverage hour of SEO/GEO work available right now.
What this briefing covers — and what it does not
| Covered weekly | Not covered |
|---|---|
| AEO and GEO research findings, verified at the source | Unverified “leaks” and hype threads |
| AI model and engine updates that affect who gets cited | Model benchmarks irrelevant to search visibility |
| Impact for businesses of every size — SMB to enterprise | Invented statistics: no source, no mention |
Frequently asked questions
How often is AI Search Watch updated?
Weekly. Each edition covers the changes of the past week in AEO, GEO and AI models, with sources for every claim.
Who is it for?
Anyone whose business depends on being found: SMB owners, marketers, agencies and enterprise SEO teams. Each edition ends with what the week’s changes mean in practice, by business size.
Where do the findings come from?
Peer-reviewed and preprint research (verified on arXiv), platform announcements and measurable engine behavior. If a number cannot be traced to a source, it is not published.
How do I know if these changes affect my site?
Run the free CapstonAI visibility audit: it shows whether ChatGPT, Gemini and Perplexity currently mention your brand, and which of the week’s shifts are relevant to your gaps.
Check what this week’s changes mean for your site — free audit →
Related: GEO scientific research · All insights · Free tools
