For twenty years, search optimisation followed a reliable arc. A platform launches a black box. Agencies sell access to the mystery. The platform then publishes guidelines and free tooling, the mystery shrinks, and an industry of specialists settles in around the edges.
That is exactly what appears to be happening with Generative Engine Optimization right now. Google has AI Overviews and AI Mode. OpenAI has search baked into ChatGPT. Microsoft has Copilot. Perplexity, Anthropic and a dozen others are building answer engines. And each of them — or the vendors orbiting them — is now shipping the same three products:
- "Are we mentioned?" — brand visibility diagnostics scoring how often an assistant names you versus your competitors
- "Fix it for me." — one-click optimisation: schema markup, entity cleanup, knowledge-graph entries, content rewrites
- "Here's your report." — quarterly or annual packages, typically priced between a few thousand and five figures a year
Read that list again and it reads like 2005: Webmaster Guidelines, the AdWords launch, the first enterprise SEO dashboards. The comfortable conclusion is that GEO will mature exactly as SEO did — platforms build the tools, agencies deliver the service, brands make the content, everyone earns a living.
That conclusion is wrong, and the reason is not the tools. It is the distribution mechanism underneath them.
SEO handed you a hundred links. AI hands you three answers.
Start here, because everything else follows.
```
SEO: search "CRM software for small teams"
└─ 100+ results across ten pages — user compares, clicks, bounces
GEO: ask an AI "what's the best CRM for a 12-person sales team"
└─ 3 recommendations, with reasoning
└─ "Want me to compare pricing?"
└─ no page two, no "more results"
```
In the SEO world, distribution was a list
Search "CRM software" and you get pages of results. Paid ads take the top slots, but organic positions sit below them. Rank fifth, tenth, twentieth — you still get traffic.
The entry points were open (there were enough of them), the choice was the user's (they scanned, compared, opened several tabs), and the market was fragmented — agencies working head terms, long-tail specialists, vertical sites, local players all had a viable business. Nobody could own every keyword, and nobody held the top slot forever.
In the GEO world, distribution is a verdict
Ask an assistant and it does not return a list. It synthesises three recommendations with reasons attached, then offers to take the next step for you.
There is no page two. There is no "more results." The model filters, compares and concludes on the user's behalf — whoever it says is good, is good. And the set of brands it is willing to say that about is very small.
This is the brutal part. SEO was a ranking contest, where everyone competes for position. GEO is a qualifying round — fail it and you do not exist. The evidence that this shift is already underway is not anecdotal:
- Gartner predicted in early 2024 that traditional search volume would drop 25% by 2026 as AI assistants absorb those queries.
- Ahrefs, working from aggregated Google Search Console data, found that when an AI Overview is present the click-through rate for the top-ranking page falls by 58% (position two: −50.8%, position ten: −19.4%) — up from the 34.5% they first measured in 2025.
- Pew Research found that users click a traditional result in 8% of visits when an AI Overview is present, versus 15% without one — and click a cited source just 1% of the time.
- A July 2026 study by Fractl and Search Engine Land, covering over a million high-volume keywords, put the decline at 29% — worse than Gartner's forecast — but found total volume roughly flat. Demand has not disappeared. It has moved, into surfaces that return one synthesised answer instead of ten links.
| Dimension | SEO (list distribution) | GEO (verdict distribution) |
|---|---|---|
| Output | Dozens of pages, 100+ links | 3 answers with reasoning |
| Slots | Abundant, paginated | Very few, no pagination |
| Who decides | The user compares and picks | The model filters and concludes |
| Nature of contest | Ranking contest (compete for position) | Qualifying round (compete for admission) |
| Cost of losing | Position 5 or 10 still earns traffic | Not selected = invisible |
| Market shape | Fragmented, many viable vendors | Concentrated, winner-take-most |
Big platforms are not democratising the tools. They are consolidating the rules.
Now revisit the analogy everyone reaches for — "Big Tech entering GEO is like when search engines launched paid search."
The analogy holds. It is also worse than that.
When Google launched AdWords in 2000, the platform took control of pricing for "who appears above whom." But organic ranking survived. Free traffic and paid placement checked each other, and a three-way market emerged: platforms set rules, agencies executed, brands produced content.
The GEO equivalent is not a pricing mechanism. It is control over who gets named at all.
With only three slots, organic effort struggles to dislodge brands already embedded in a model's training data, its retrieval index, or a licensing deal. The likely end state is a platform holding the organic admission standard in one hand and a paid priority channel in the other — the same party acting as referee, competitor, and gatekeeper of who is allowed on the field.
Three global dynamics make this sharper than the SEO era ever was.
1. The referee also owns the stadium. In the US, the Department of Justice's search case and the EU's Digital Markets Act were both built for a world where the harm was *ranking* — demoting rivals, self-preferencing a comparison service. That legal vocabulary fits a list. It fits a verdict badly. When an assistant simply does not mention you, there is no demotion to point at, no lost position to audit, no comparable harm to plead.
2. The supply side is organising. On 1 July 2025, Cloudflare — which sits in front of roughly a fifth of the web — made blocking AI crawlers the default for new domains and launched pay-per-crawl, reviving HTTP 402 so publishers can charge for access. Their own crawl-to-referral numbers: Google 14 crawls per referral, OpenAI 1,700, Anthropic 73,000. Condé Nast, TIME, The Atlantic, the Associated Press, Gannett, Reddit and Pinterest all backed the move. A licensing market is forming, and licensing markets favour whoever already has scale.
3. Every market is running the same play. In China, 360, Alibaba, Tencent and Baidu are packaging GEO for SMEs. In the US, it is Google, OpenAI, Microsoft and Perplexity. Different names, same structure, same endgame: whoever owns the model owns the answer, and whoever owns the answer sets the price of admission.
None of this is tool democratisation. It is rule consolidation, entry centralisation, and value capture at the platform layer.
The endgame: winner-take-most
Chain the logic together and the conclusion is not subtle.
SEO's foundation was list distribution → plenty of slots, opportunity spread thin → organic and paid coexisted for two decades → a diverse ecosystem of vendors.
GEO's foundation is verdict distribution → a handful of answers, no page two → selected means you take everything, unselected means close to nothing.
And the platforms hold all three levers at once: the model, the selection standard, and the optimisation tooling. You either grind against their criteria, or you buy your way into priority — and the entity selling the second option wrote the first one.
There is no middle ground. In SEO, ranking fifth still paid. In GEO, ranking fourth is functionally identical to ranking four-hundredth.
So here is the call: GEO will not repeat SEO's arc of broad participation and a long tail of specialists. It looks much more like a membership club — a few admitted brands capturing nearly all the value, and a few platforms holding the door. Big platforms are not arriving to lower the barrier. They are building the AI recommendation layer, from day one, as a system where a small number of parties decide who gets in.
SEO was a thousand roads into Rome. GEO is three bridges, and the platforms own the bridges. That is the real endgame.
What this means if you are on the buying side
The conclusion is not "do nothing." It is "stop budgeting for this as if it were SEO."
- Optimise to be quotable, not just rankable. Short, self-contained statements of fact with a source attached get lifted into answers. Buried qualifications do not.
- Make your entity unambiguous. If a model cannot tell you from a competitor with a similar name, it defaults to whoever it already knows. Structured data and consistent entity descriptions across the web are the cheapest fix available.
- Track mentions, not just positions. Rank tracking will not tell you that an assistant stopped naming you. Sample real prompts every month and log whether you appear at all.
- Assume the criteria will change. Every one of these surfaces is being tuned in public. Any vendor promising durable placement is selling something they do not control.
Key takeaways
- The analogy holds: Big Tech entering GEO ≈ search engines launching paid search.
- The mechanism differs: SEO was list distribution (many slots, paginated); GEO is verdict distribution (a few answers, no page two).
- The consequence: organic and paid coexisted in SEO for twenty years. In GEO, being selected takes everything and being excluded takes nothing — there is no middle.
- The direction: selection standards and admission both sit with the AI platforms, so market value concentrates at the platform layer instead of spreading across a service ecosystem.