Strategy
GEO Grows Up: Notes on a Discipline Finding Its Footing
Two years ago, generative engine optimization was a hustle. Agencies sold "AI visibility audits" that amounted to stuffing FAQ pages with conversational phrasing and hoping an answer engine would quote you. It was 2011 SEO cosplay, and most of it did nothing.
That era is ending, and the discipline that is replacing it is genuinely interesting. We spent much of Q4 doing GEO strategy work for two clients, including a substantial engagement with Northwind's retail team, and the patterns we are seeing deserve a proper write-up.
The shift: from ranking to being cited
Classic SEO optimized for a position on a page. GEO optimizes for being the source an answer engine trusts, quotes, and links when it synthesizes a response. Those are different games with different physics.
The biggest practical difference is that answer engines reward being the origin of a fact, not the loudest repeater of it. When Northwind publishes original sizing data, return-rate statistics, or care instructions that exist nowhere else, assistants cite them by name. When they publish a rewritten version of the same generic buying guide as every competitor, they vanish into the synthesis, uncredited.
This changes what content strategy even means. The question is no longer "what keywords do we own?" It is "what does this brand know that nothing else on the internet knows, and have we published it in a form a machine can verify and attribute?"
What actually moves the needle now
After a year of testing, our honest list is short.
- Original, verifiable information: proprietary data, real measurements, named methodologies. Answer engines are increasingly good at tracing provenance.
- Consistent entity hygiene: the brand described the same way across your site, your structured data, your press coverage, and the knowledge graphs. Contradictions cost you citations.
- Genuine expertise pages with named humans attached. Bylines with real credentials outperform anonymous brand voice in citation frequency, in every test we have run.
- Being talked about elsewhere. Citations beget citations. PR and GEO are converging faster than either industry wants to admit.
Notice what is not on the list: volume. Publishing more mediocre pages is now actively harmful, because it dilutes the entity and gives engines low-confidence material to judge you by.
The measurement problem is real, and maturing
The awkward truth of 2024-era GEO was that nobody could measure it. That is improving. Share-of-citation tracking across the major assistants is now table stakes, and the better tools can attribute assistant-referred sessions with reasonable confidence. It is still fuzzier than paid search ever was. But "fuzzier than paid search" describes most things that matter in brand building, and we have never accepted that as a reason not to invest.
What we tell clients: treat GEO metrics like brand-tracking metrics, not performance metrics. Directional, quarterly, tied to a strategy. Anyone selling you daily GEO dashboards with precise ROI is selling you 2013 again.
Why this is a strategy problem, not a content problem
Here is the opinionated part. GEO keeps getting parked in the content team's backlog, and it does not belong there. The core questions are positioning questions. What is this brand the definitive source for? What claims can it make that are both distinctive and verifiable? What should a machine say when a human asks who does this best?
That last question is essentially the old positioning question with a new judge. Brands with sharp, true, differentiated positioning are winning citations almost by accident. Brands with mushy positioning are discovering that answer engines are brutally honest editors: they compress you down to whatever is actually distinct, and if nothing is, you get compressed to nothing.
GEO growing up means it stops being a channel tactic and becomes a strategic forcing function. The brands that treat it that way in 2026 will spend the next five years being the answer. Everyone else gets to be a footnote in someone else's.
Building something this could apply to?
We take on a small number of flagship projects each quarter.
Start a project