Welcome to our weekly briefing on AI Search. The Prompt Marketing industry (aka AI Search / GEO / AEO / LLM SEO) is filled with developments and news. Every week we track the latest developments on how AI systems find, cite and recommend brands. Our News section is designed to filter the noise and pick out the developments that actually change something. This week: the category finally gets a measurement standard, research shows ChatGPT quietly prefers your English pages, and ChatGPT Ads starts behaving like a real ad platform.
Money is moving into AI visibility faster than measurement can keep up
On 3 August the IAB published Measuring Visibility in the AI Era, the first serious attempt to standardise how brand visibility inside AI answers is measured. The context is uncomfortable. More than twenty vendors now sell AI visibility tools, each with its own methodology, and they routinely produce different and sometimes contradictory numbers for the same brand. Only 16% of brands track AI visibility systematically at all.
The guidelines do three useful things. They define a shared vocabulary across four levels (presence, prominence, portrayal and persuasion), they split data into two quality tiers, and they list what a vendor has to disclose before a number deserves to be trusted. The tier split is the part worth remembering: directional data is fine for spotting trends and keeping an eye on competitors, while decision-grade data has to meet standards for sample size, reproducibility and documented methodology before it should be allowed to move a budget.
The timing is no coincidence. Digiday reported on 7 August that marketers now put roughly 24% of their search or content budgets into AI visibility work, with 82% spending something and 43% spending more than a fifth of their budget. In the same round-up, 52% of B2B marketers rated AI answer engines their most effective distribution channel, against 29% for organic search.
Our view: the money arrived before the measurement did. If you buy a single AI visibility tool and treat its score as truth, you are making decision-grade calls on directional data. Ask any vendor for the disclosures the IAB now lists. If a score cannot survive that question, use it to watch direction and nothing more.
ChatGPT reaches for your English pages, Google does not
The most useful research this week is Pieter Serraris' analysis of multilingual websites, published 5 August. He measured which language version the AI systems actually fetch across dozens of multilingual sites, and the three big engines behave completely differently.
ChatGPT over-indexes English by a median factor of 2.6, fetching the English version 65% to 79% of the time. Copilot sits at 1.07, effectively neutral. Google AI comes in at 0.79, which means it leans slightly towards local language pages. Sites that do run an /en/ folder saw higher AI citations across the board: 122% for ChatGPT, 52% for Copilot and 28% in aggregate for Google AI, although the Google figure drops to a 9% median per site once outliers are removed.
The caveat in the research matters as much as the numbers. Only add an English version if you can genuinely maintain it. A stale English page can become the version cited to your customers instead of the local one you actually keep up to date.
For European brands this is the practical finding of the week. If ChatGPT matters in your category, the English version of your site is not a nice-to-have for international visitors, it is the page the model is most likely to read about you. That sits oddly against what we see in the market ourselves. Our own sweep of business communities across English, Dutch, German and French found the AI Search conversation is currently concentrated in French, following the AI Overviews launch there on 22 July, and close to silent in Dutch and German. The engine reads you in English while your buyers are still talking in their own language.
Being cited is not the same as being recommended
Kevin Indig published a study with Semrush on 5 August covering 1,094 categories with five prompt variants each, using US ChatGPT data from January to June 2026. It puts numbers on something we end up explaining in almost every client conversation.
In categories far from a brand's core expertise, 50% of appearances are citations and only 25% are named recommendations. In topically relevant categories those figures rise to 74% and 44%. Getting quoted outside your lane is achievable. Getting recommended outside it is not.
Two further numbers stand out. In June 2026 only 15.2% of categories had a clear owner, and 53.7% were open fields with several contenders and no winner. The effect also varies by sector: finance and real estate teams can push into adjacent categories and treat citations as an early signal, while legal and healthcare stay constrained even at full coverage.
Read alongside the IAB guidelines, this is a strong argument for measuring recommendations separately from citations. A visibility score that rolls both into one number will tell you that you are winning categories you are merely being quoted in.
ChatGPT Ads starts behaving like an ad platform
OpenAI shipped a batch of ChatGPT Ads updates reported on 7 August. Conversion-optimised cost-per-click is in beta for product feed campaigns, a multi-product carousel format is in testing, and dynamic URL parameters can now append campaign, ad group and ad IDs to landing page URLs. Brazil and Mexico were set to go live within the week.
One item comes with a deadline. Automatic advanced matching is now the default on new web pixels, and on 17 August OpenAI switches it on for existing pixels unless the advertiser opts out first. If you are running ChatGPT Ads, that is a decision to make this week rather than next.
The dynamic URL parameters are the quiet win here. Until now ChatGPT ad traffic has been awkward to isolate inside analytics, which made it hard to argue for budget. Being able to segment it properly is what turns the channel from an experiment into a line item.
Also noted
• Google told The Verge that Reddit gets no special preference in Google Search or in its AI search features, in a piece on the wave of promotional posting aimed at AI systems.
• Search Console does surface AI Mode queries in the main performance report, just not in the generative AI report where you would expect to find them.
• Google reshuffled its AI leadership on 5 August: Jeff Dean left after 27 years and Demis Hassabis moved to Chair of Google DeepMind and Chief Scientist of Alphabet.
That is the week. If you want to know what any of this means for your own AI visibility, our door is open.