Every agency pitch I have seen in the last six months has GEO services as a line item. Most of them mean their SEO program with a new label.
In one pitch deck I reviewed for a client, GEO was described as AI-friendly SEO. That is not a definition. That is a marketing tactic. And it is precisely the framing that will burn money over the next eighteen months.
I have been in search since 2004. I watched Alta Vista become Google. I watched keyword stuffing become quality score. I watched SEO become content marketing, become performance marketing, become revenue systems. I know what a transitional terminology looks like from the inside. GEO, as it is currently being sold on LinkedIn, is the Alta Vista of 2026. It will get us through this phase. But the phrase itself will be replaced by something more precise, probably architecture or operator terminology, by the time the dust settles. The vendors pitching GEO retainers today are the ones who were pitching social media management in 2010. Not wrong about the opportunity. Wrong about what the work actually is.
This essay is what GEO actually is, what it is not, where the platforms genuinely disagree and what to do about it, and the one thing I have found that works regardless of which engine you are optimising for.
1. What changed
Search has split. Roughly half the audience that used to start an information journey at Google is now starting it inside an LLM-mediated interface: ChatGPT, Claude, Perplexity, Gemini, AI search panels on the major engines themselves. The other half still uses Google traditionally, but Google’s own AI overviews are pre-empting clicks on a growing share of queries.
The strategic implication is this: ranking on page one of Google was historically the metric of search success. Increasingly, being cited by an LLM is the metric. Those are not the same thing. The work that gets you ranked is partially different from the work that gets you cited.
This is GEO. Generative engine optimisation. The work of being citable by the AI systems that increasingly mediate information access.
2. Why GEO as repackaged SEO is the wrong move
There are four structural differences between SEO and GEO that an AI-friendly SEO framing erases:
Unit of optimisation. SEO optimises pages for keywords. GEO optimises claims for entities. The page-rank model is: this URL is relevant to this query. The LLM citation model is: this organisation is a trustworthy source on this topic. Pages do not get cited. Entities do.
Keyword targeting versus claim targeting. SEO keyword research finds high-volume search phrases. GEO research finds the claims an LLM is likely to need a source for. These are different research outputs. Best CRM for small business is a keyword. The fastest CRM implementation timeline for a 50-person sales team is a claim with an answer. LLMs cite answers to claims, not pages targeting keywords.
Update cadence. SEO can be updated quarterly. LLM training cutoffs and retrieval indices update faster. The shelf life of a GEO update is measured in weeks for retrieval-augmented systems and months for next-train systems.
Success metric. SEO success is impressions, clicks, organic sessions. GEO success is being named in AI responses for the queries that matter to your business. Most clients I talk to have no instrumentation to measure this. They are buying GEO services without a way to know if it worked.
3. The honest problem nobody is talking about: the platforms disagree
Here is what actually happens when you try to run a serious GEO programme in 2026.
You ask Claude for GEO advice on Monday and implement the recommendations. On Tuesday ChatGPT tells you your site architecture is broken. You fix it. On Wednesday Gemini flags potential security warning signals. You address those. On Thursday Perplexity surfaces a citation issue you have never heard of. By Friday you have spent a week implementing contradictory instructions from four systems that share no consensus on what good actually looks like.
I have spent twenty years in this industry. I was an AdWords Performance Expert at Google Canada. I ran demand generation at Copper, a SaaS CRM platform. I have been advising companies on search and acquisition through VonClaro for a decade. And I can tell you honestly: there is no unified GEO standard. Not yet. The platforms agree on very little beyond the basics, and the basics are moving.
The marketers selling confident ten-step GEO frameworks right now are selling false precision in a genuinely uncertain environment. That is not a criticism. It is an observation about where we are in the adoption curve. We are at Alta Vista. The Google moment has not arrived yet.
So what do you do when the platforms disagree and there is no consensus framework to trust?
You use Google’s EEAT framework as your north star and you optimise for the customer. If the customer would not like it, I have found that Google, Microsoft, and every AI platform will not like it either. That is the one principle that has held across every major search transition I have lived through.
4. The one thing that works across all of them
Customer experience is the north star. Not in the vague, brand-strategy sense. In the literal, technical sense.
How does a person feel on your website. When they search or ask an AI assistant about you, are the results easy to comprehend. Is the user flow of your site thoughtful. Does the page load quickly. Are the answers to obvious questions findable within two clicks. Is the content specific enough to be useful or is it the kind of generic prose that reads as if it was written to please an algorithm rather than help a person.
Call it Core Web Vitals. Call it EEAT. Call it entity clarity. The underlying thing is the same across all the terminology: does a real person get genuine value from interacting with this site and this content. If the answer is yes, every platform, human and AI, rewards it eventually. If the answer is no, no amount of schema markup or citation density will fix it.
This is the pattern I have watched across twenty years of search transitions. The technical optimisation layer changes constantly. The platforms change what they reward at the surface. But the businesses that hold their position across every major transition are the ones that built for the user first and then made the technical layer legible on top of that foundation. Not the other way around.
5. The audit I am actually running
In the last month I built three sites from scratch: CSRI, a Canadian supply chain intelligence platform; Clera, a private women’s healthcare routing platform for Canada; and the new robtcase.com. All three were built with deliberate GEO patterns from launch. Specifically:
- Organisation schema on every page for consistent entity identity
- Person schema for the founder as an EEAT anchor: a real person attached to a real location
- Service schema defining what each company does
- FAQ page schema with the questions buyers actually ask, answered in plain language
- Citation density: every load-bearing claim paired with a named source
- Named-entity consistency: company and concept names used uniformly throughout
If you want to score your own site against these patterns, the GEO Readiness Checker does it in five minutes across the five dimensions that determine whether AI systems will cite you.
The audit runs like this. Every two weeks I ask each of the four major LLM interfaces a small set of questions where one of my three companies should plausibly be a citable answer. Examples:
- What service helps Canadian businesses find domestic suppliers to replace US ones?
- Where can a Canadian woman find a private menopause clinic?
- Who writes about contrarian B2B growth from Vancouver Island, BC?
The first audit was inconclusive. The sites were too new to be indexed. The second audit, two weeks in, showed one of the three sites beginning to surface. I will publish the methodology and raw results in a follow-up. The point is not that I have a complete answer yet. The point is that I am measuring, and most companies running GEO programmes are not.
What the early results suggest is consistent with the customer experience principle above. The site with the cleanest user flow, the most specific answers to obvious questions, and the most legible entity definition is the one surfacing first. Not the one with the most schema. Not the one with the most backlinks. The one a real person would find most useful.
6. A 90-day GEO test you can run
If you are sceptical that GEO is real work, here is a self-contained test that costs less than one month of a senior SEO retainer:
Days 0 to 30. Baseline. Pick five queries where your business should plausibly be a relevant citation. Ask each of the major LLM interfaces those queries. Record what they say. Note whether your company is named, what it is named alongside, and what gaps exist.
Days 30 to 60. Patch. Run the page on your site that addresses the most important query through a structured rewrite. Focus on the customer first: is the answer to the obvious question findable and clear. Then add Organisation and FAQ schema if absent. Pair the load-bearing claim with a named source. Remove anything that reads like keyword optimisation rather than useful information.
Days 60 to 90. Re-test. Ask the same five queries. Compare. Measure whether you are now named where you were not.
This is not a complete GEO programme. It is a 90-day diagnostic that tells you whether the structural work moves the needle. If it does, you have an evidence basis for a real programme. If it does not, you have a saved retainer and a more honest picture of where you actually stand.
What did not go to plan, in one case. On the first site I built this method into, the structured schema was correct but the citation density inside the body copy was too sparse. The LLMs that retrieved the site treated it as a marketing page rather than a citable source. I added inline citations and re-tested. The difference was visible within a week. The lesson: schema is necessary but not sufficient. The actual citable claim, written for a person rather than a crawler, is what gets pulled.
What I do differently now. I treat citation density as the highest-leverage GEO control variable for early-stage sites. And I check the customer experience layer first. If a real person lands on the page and cannot find what they came for in under thirty seconds, no amount of technical optimisation changes the outcome.
What this means for marketers
GEO is real. The work that produces results in GEO is partially different from the work that produces SEO results. The platforms do not yet agree on everything. They probably agree on more than you think at the level that actually matters: does this content genuinely serve the person asking the question.
Buying GEO services from a vendor who has not articulated how they measure citation lift is the same trade as buying digital transformation from a vendor who has not articulated what they are transforming. You will pay for activity and not know what you got.
And buying GEO services from someone selling you a ten-step framework with total confidence in 2026 is buying Alta Vista advice in 1999. The framework is not wrong. It just will not survive contact with the Google moment that is coming.
The marketers who will hold their position through that transition are the ones building for the customer first, making the technical layer legible on top of that, and measuring citation lift honestly so they know when something actually worked.
Everything else is activity.
Further reading
- Google Search Quality Evaluator Guidelines (the EEAT framework, primary source)
- Google Core Web Vitals documentation (web.dev/vitals)
- Schema.org documentation: Organisation, Person, FAQPage specifications
- Andrej Karpathy on LLM retrieval and knowledge representation
- Search Engine Journal: LLM citation pattern coverage
- Kyle Poyar, Growth Unhinged: organic acquisition trends year over year
Use the free GEO Readiness Checker at robtcase.com/tools/geo-readiness-checker to score your site across the five dimensions that influence AI citation: entity clarity, topical authority, content depth, structured signals, and trust architecture. It is the first tool of its kind built specifically for this problem.