We Stopped Writing Blog Posts. We Started Writing for Machines. Here's What Happened.
Inside the 620-day GEO content engine that replaced Creative Agent's blog. The architecture, the citation share gains, the search-traffic shift, and what the data taught us about writing for AI answer engines.
In 2024 the Creative Agent content engine looked like every other UK agency content engine: a keyword list, a brief template, a writer rota, a publishing cadence, a monthly traffic deck. It worked. It also stopped working — slowly at first, then quickly — as Google AI Overviews and ChatGPT browsing started absorbing the answers our blog posts existed to provide.
So we stopped writing blog posts. We started writing for machines. 620 days later the system has changed enough that it deserves a full write-up.
Why the old engine broke
The keyword-led blog model was built for a SERP made of ten blue links. When the SERP becomes an AI-generated paragraph plus three cited sources, the unit of competition shifts from rank to citation. A piece that ranks fifth on a SERP is invisible inside an AI answer. A piece that gets cited inside the answer wins, regardless of where it would have ranked classically.
The supply side made it worse. Generative content cost collapsed, every competitor flooded the same long tail, and Google's update cycle started actively suppressing low-distinctiveness content. The old engine was producing more, ranking less, and being cited almost never.
What the new engine does instead
Three architectural changes carry most of the weight.
Entity anchoring. Every page is tied to a clean schema entity — Organization, Person, Product, Service, FAQPage — that the AI engines can read, disambiguate and attribute. The Creative Marketing Group Organization graph and the Kasim Javed Person graph carry the citation weight that individual blog posts used to have to earn alone.
Structural cleanliness. Every page has a TL;DR, explicit Q&A blocks, numbered lists where lists belong, and stat boxes with attributable sources. LLMs lift these almost verbatim. The cleaner the structure, the higher the lift rate.
Citation engineering. Third-party citations now matter more than internal link equity. The 620-day rebuild includes a continuous PR and citation-acquisition operation that ensures the open web has multiple independent sources confirming the same entity-level claims. The AI engines triangulate on this.
From the first machine-first piece to the current production GEO engine — Creative Agent's full editorial system rebuild.
What the data showed
The leading indicator that moved first was citation share inside ChatGPT, Perplexity and Google AI Overviews — measurable through manual sampling, the brand-mention tracking inside the major AI assistants, and traffic referral data. SERP rank movement followed later. AI assistant referrals showed up as a distinct, measurable channel inside the analytics — not a rounding error.
The visitors arriving from ChatGPT and Perplexity converted at materially higher rates. They had already been pre-qualified by the AI assistant, which had effectively performed the discovery and shortlisting work that classic SERPs used to leave to the user. The assistant arrives at the site already convinced.
What broke along the way
Not everything worked. The first six months over-indexed on volume. We published too many machine-first pieces too quickly, and the diminishing returns showed up inside the citation data faster than we expected. The fix was harder selection — fewer, denser, more defensible pieces with stronger entity anchors and better third-party citation backing.
The internal team also had to change. Writers who optimised for keyword density were the wrong fit; writers who could carry an argument and cite cleanly were the right fit. The org chart shifted from a publishing team to an editorial-and-citation team.
The replicable system
The full nine-step playbook is documented at /insights/9-step-geo-playbook. The strategic discipline report — what GEO means at industry scale — is at /seo-in-the-geo-era. The operating doctrine that sits behind both is /insights/kj-doctrine-10-rules. Together they are the system. Not a one-off case study.
If you want to apply this to your own brand, book a call. The first hour is normally spent on entity audit — and that hour usually changes the next ninety days of the content roadmap.
Frequently asked
What is a GEO content engine?
A content operation built specifically to be cited by AI answer engines (ChatGPT, Perplexity, Claude, Gemini) and to dominate Google AI Overviews. It replaces keyword-led blog publishing with entity-anchored, schema-saturated, structurally-clean content that an LLM can lift verbatim and attribute correctly.
How is this different from SEO?
Classic SEO optimises for SERP rank inside a list of ten blue links. GEO optimises for citation share inside an AI-generated answer. The architectural difference is real: GEO requires Organization and Person schema saturation, third-party citation graphs, structurally clean Q&A blocks, and entity disambiguation that classic SEO never required. Read the full discipline report at /seo-in-the-geo-era.
Did organic traffic drop?
Top-of-funnel SERP impressions softened in line with broader UK AI Overviews data — see /uk-ai-overviews-traffic-tracker. But mid-funnel branded search and AI assistant referral traffic both grew, and the visitors arriving from ChatGPT and Perplexity converted at materially higher rates because they had already been pre-qualified by the AI assistant.
How long did the rebuild take?
620 days from first machine-first piece to the current production engine. Most of the time was system design, schema architecture, and the rebuild of the editorial operating model — the writing itself accelerated dramatically once the system was in place.
Is this replicable for other UK brands?
Yes. The architecture is replicable, the schema patterns are documented, and the operating model — entity ownership, citation engineering, structural cleanliness — applies to any UK brand with a defensible point of view. The nine-step playbook lives at /insights/9-step-geo-playbook and the doctrine that informs it is at /insights/kj-doctrine-10-rules.
Where do I start?
Book a call at /kasimjaved. The first hour is normally spent on entity audit — does your brand exist as a clean entity inside the open knowledge graph that the AI engines read from? If the answer is no, that is the first fix and everything else compounds on it.
See what AI-first email looks like
AutoEmails is the AI Email Operating System we built — autonomous lifecycle marketing in production.
Related insights
All researchGenerative Engine Optimisation: The New SEO Nobody in the UK Is Doing Properly
The definitive UK explainer on GEO — how to win citations from ChatGPT, Perplexity, Claude and Google AI Overviews, and why traditional SEO no longer fills the funnel.
How to Make ChatGPT Recommend Your Brand: The 9-Step GEO Playbook
Highly shareable practical guide — schema, Wikipedia, third-party citations, structured FAQs, entity SEO, author markup. The piece every UK marketer will save.
Why I Killed Our SEO Team and Built a 620-Day GEO Engine Instead
Insider story of Creative Agent's content automation system. The economics, the architecture, the results — honest about tradeoffs.
