Why I Killed Our SEO Team and Built a 620-Day GEO Engine Instead

In Q3 2025 I made the call that defined the next five years of Creative Agent: I dissolved the traditional SEO function and rebuilt it as a generative-engine optimisation (GEO) engine. The decision was internally unpopular, externally controversial, and — fourteen months on — clearly the right one. This piece is the honest version of what happened, including the parts the trade press never asks me about.

1. The pattern that forced the decision

For eighteen months prior, the SEO team had been doing excellent, careful, link-earning, schema-marking, content- pruning work. The metrics looked healthy on the dashboard. Rankings held. Then in Q2 2025 the floor moved.

Google AI Overviews started taking the top of the SERP for our money keywords. Click-through rates on rank #1 fell from 32% to 11% inside one quarter. Meanwhile, the inbound leads that came in by mentioning "ChatGPT recommended you" or "Perplexity cited your piece" went from one a month to one a week. The data was screaming the same message: the surface that produced our pipeline had moved, and the team built for the old surface was operating in a market that no longer existed.

2. What I dissolved and what I built

The legacy team was four people: senior SEO strategist, two operators, and a technical SEO. The new structure is three people: a senior GEO strategist (one of the original two operators, now elevated), a senior content-architecture lead (the original senior SEO strategist, role expanded), and a systems engineer who runs the agent layer. Two roles were removed.

The agent layer does the work that the operator layer used to do. It generates the long-tail Q&A pieces from a structured brief. It schema-marks every asset. It builds the internal linking. It tracks AI-search citation acquisition daily. The humans set strategy, write the pillar pieces, and review the exception screen.

3. The 620-day engine in plain terms

A 620-business-day (≈30-month) forward content plan organised into seven editorial kinds:

• Pillars (~12 a year): flagship narrative pieces — opinion-shaped, citation-bait. Authored by the founder.
• Disciplines (~24 a year): deep explainers per marketing discipline. Co-authored.
• Research (~12 a year): "State Of" reports with original UK data — most-cited format by AI engines.
• Essays (~12 a year): historical/cultural anchors. Slow burners that earn links forever.
• Sector pieces (~16 a year): UK geographic/industry explainers.
• Product pieces (~24 a year): Autothink portfolio storytelling.
• Long-tail Q&A (~1,000+ a year): structured answers to specific user questions, optimised for AI retrieval. The volume layer.

The roadmap is built once, reviewed quarterly, and shipped against daily. Total volume: ~1,200 pieces a year. Total headcount: 3 people plus the agent layer.

4. The numbers, honestly

Cost: Operator cost per published asset down 47%. Total content-team cost down approximately 35% (the remaining team is more senior, more expensive per head).

Conventional organic traffic: Down 38% year-on- year. Google AI Overviews are the principal cause; the engine did not stop the bleed, because nothing stops the bleed.

AI-search referred traffic: Up 4.2x. Smaller absolute volume but converts at roughly 3x the rate. Net effect on qualified inbound: +6x.

Brand-search uplift: +29% — driven by ChatGPT/Perplexity citations triggering branded follow-up searches.

5. The architecture, simplified

Three layers:

1. Roadmap layer. The 620-day plan, stored as structured data. The single source of truth for what gets shipped, in what order, against which entity, with what schema.
2. Generation layer. Agentic — pulls from the roadmap, drafts in brand voice, applies schema, internal links, and entity markup. Outputs go to senior review queue, not direct publish.
3. Citation-tracking layer. Monitors which AI engines cite which pieces, on which prompts, with what attribution. Feeds back into the roadmap re-prioritisation.

The mistake most agencies will make when they try to copy this is starting from the generation layer (the AI). The right entry point is the roadmap layer. The roadmap is the IP. The generation is the commodity.

6. The hard part

The technology was the easy part. The hard part was telling people their roles were disappearing. I will not pretend the severance conversations were clean. They were the worst part of the year for me personally and the part I was least prepared for.

What I would do differently in 2026: start the conversations eighteen months earlier, not three. Offer transition apprenticeships into senior strategy roles for anyone willing. Be honest about which roles will not exist by Q4 — vagueness is crueller than directness, every time.

7. What you should do, depending on who you are

If you run an in-house SEO team — start modelling a GEO-first roadmap now, even if you keep the legacy team in parallel for twelve months. The transition takes eighteen months; you do not have eighteen months after the board notices the bleed.

If you are an SEO professional — re-tool. The strategist, content-architect and entity-SEO roles are expanding. The operator roles are not. The single best skill to develop in 2026 is GEO content strategy: understanding what AI engines cite, why, and how to engineer a corpus to be cited.

If you run an SEO agency — re-price your book onto outcomes (citation acquisition, qualified-lead attribution from AI search) before procurement does it for you. The retainer-model warning piece applies to SEO retainers first.

To talk through this transition for your own team or agency, book a call. For the practical GEO playbook, read How to Make ChatGPT Recommend Your Brand.