
The AI advantage also lives in architecture and infrastructure (and in costs)
Fifteen seconds feels like magic… until you look at the bill. Z.ai is an agent on infra; Cocobot is architecture and infra for an agent. Different jobs, same cost compass.

AI scraping — copying web content at scale to train models — is back at the center of a legal fight. It is not only a media-law story: it hits B2B companies (selling to other businesses) that depend on organic traffic and on being cited when someone asks an assistant.
Somos Gente Digital's take is blunt: the debate confirms what already matters for GEO and AEO. GEO (Generative Engine Optimization) is how AI assistants find and cite you. AEO (Answer Engine Optimization) is designing for answer engines — search, chatbots, copilots — not only ten blue links. If you only recycle other people's web, you lose when those systems stop sending visits.
TechCrunch summarizes newly unredacted filings in The New York Times lawsuit against OpenAI and Microsoft: a Microsoft executive described AI scraping as labor theft at historic scale, and internal signals show AI products can substitute visits to the original sources.
The answer for companies is not fighting the Microsoft headline. It is publishing operable own knowledge — processes, cases, stack, clear FAQs — and designing sites that answer without substituting themselves. At SGD we work SEO·GEO·AEO as a channel of a B2B product, not as ad spend.
Scraping here means collecting pages at scale — often without clear permission — to feed model training or grounding (using someone else's text as context). It is not a visitor reading your blog; it is mass copy so a system can answer without the person opening your site.
That matters because your B2B web's value is not only “being indexed.” It is that the buyer arrives, trusts you, and moves through your process. If the answer stays inside the chatbot, the channel thins out.
We attribute carefully: much of the new material comes from The Times' own brief, not every sealed exhibit. TechCrunch reports, among other points, that an internal Microsoft document (Brent Hecht, January 2024) described a “doom loop”: fewer clicks to sources, a weaker web, weaker models. In the same piece, Microsoft data on Copilot as an “answer engine” shows steep click-through drops for the Times domain versus traditional Bing search.
OpenAI, per the same reporting, has internal notes describing products like ChatGPT as “largely substitutive” of the original publisher. Nadella, in sworn testimony cited by TechCrunch, agreed that chatting with bots can substitute going to the underlying source.
We do not invent quotes or court outcomes. The business pattern in plain language: if the assistant answers “right there,” many visits never arrive. For a publisher that is existential. For a LatAm B2B team publishing generic recycled content, it is the same loop in another costume: you become invisible in the “organic” channel you thought you owned.
GEO and AEO are not keyword magic. They are a content-product decision:
If your blog is a collage of what already exists on ten sites, models and engines will treat you as noise. If your blog is the operable manual of your B2B product, you have something to cite — and a reason to click.
We are a human web/tech B2B agency team. We do not sell “winning a lawsuit against Microsoft.” We sell building the channel: classic SEO where the click still matters, GEO/AEO where the citation and the answer matter, and a site that converts.
In practice: processes, cases, stack, and clear FAQs; content architecture with Prismic or another headless CMS when you need editorial control; and front ends that answer without eating the next click into your product. For the software method, see software development that works; our services map; or how we work with Prismic.
If you are also accelerating with AI in the codebase, the same judgment applies: speed without a domain map is another way to recycle without understanding. We cover that in vibe-coding without a map.
Copying web content at scale — pages, articles, datasets — to train or ground models. It is not a visitor reading your site; it is mass extraction that can power answers without sending traffic.
GEO (Generative Engine Optimization) is how AI assistants find and cite your content. AEO (Answer Engine Optimization) is optimizing for answer engines (AI search, chatbots, copilots), not only the classic link list.
Not magically. The filings reinforce a market risk already visible: click substitution. Your operational response is publishing operable own knowledge and designing the site as a product channel — not fighting the headline.
It can be part of a technical policy, but it does not replace having something worth citing. Blocking without differentiated knowledge does not build GEO/AEO.
One clear idea per URL, citable FAQs, honest CTAs, and editorial architecture (often headless) so marketing publishes without breaking the product. The assistant can summarize; the buyer still needs your process, your demo, and your team.
No. In LatAm B2B they coexist: SEO where the click still matters, GEO/AEO where the answer and citation set the first impression. SGD treats them as one product channel.
If your site still lives by recycling what is already on the web — or by publishing without a map of real buyer questions — let's talk. At SGD we build SEO·GEO·AEO as a channel of a B2B product, not as ad spend.
OWN CONTENT · GEO · AEO

Fifteen seconds feels like magic… until you look at the bill. Z.ai is an agent on infra; Cocobot is architecture and infra for an agent. Different jobs, same cost compass.

Practical guide for owners, IT, and marketing in Colombia: B2B criteria, rescue vs ads signals, 2026 checklist, and questions for the first call.