If I can only fund one, SEO or GEO?

Fund the layer that is broken, which is usually the shared one. If AI crawlers cannot read your site, that same problem is costing you in both channels and fixing it is not a choice between them. Once the site is readable, split by where your buyers ask: if they arrive through comparison searches, weight SEO; if they arrive already holding a shortlist someone else wrote, weight GEO.

Do good rankings eventually produce AI mentions?

Sometimes, and not reliably. Across 100 audits we found no correlation between organic performance and being named in an answer: brands with strong traffic and zero AI presence were common, and one agency in our competitor set had roughly 350 monthly visits while still appearing on the AI shelf. Ranking helps because ranked pages are often the sources engines cite, but it is not a substitute.

Does the same content work for both?

Partly. Long comprehensive pages win rankings; short liftable passages win citations. The practical answer is to keep the depth and add the answer-shaped surface: a 60-word definition near the top, a comparison table, a price signal, an FAQ whose markup matches the visible text word for word.

What metric replaces rankings in GEO?

Two, and they are different. Brand recall — does the engine know and describe you correctly. Category recall — does it name you when a buyer asks who to pick. In our dataset those were 93% and 4.5% respectively, which is why reporting only the first one is misleading.

GEO vs SEO

SEO competes for a position on a page of results. GEO — generative engine optimization — competes for your name inside the sentence an AI writes when a buyer asks who to pick. They share a technical foundation and diverge on everything above it: the crawler, the competitor set, the baseline, and what counts as success.

The most useful thing we can add to this comparison is not an opinion but a measurement: across 100 audits, organic performance and presence in AI answers did not correlate. Brands with healthy traffic and no AI presence were common, and so was the reverse.

side by side

the same site, two different exams
	SEO	GEO
the goal	Rank a URL	Be named in an answer
the surface	A results page of links	A written answer with two or three names
the crawler	Googlebot, which renders JavaScript	GPTBot, ClaudeBot, PerplexityBot and others — several render nothing
the competitor set	Whoever ranks on the keyword	Whoever the model names — often different companies
the unit of work	Keywords	Prompts
the baseline	Positions and sessions	A logged prompt panel with verbatim answers and cited sources
what wins	Depth, links, relevance	Readability, a verifiable entity, liftable passages, independent sources
feedback speed	Weeks to months, fairly stable	Days, and volatile — single runs prove little
failure mode	Page eleven	Absent from the sentence, or present with a warning attached

what they genuinely share

Roughly a third of any GEO audit would appear in a competent SEO audit too, and that overlap is the cheapest work in either discipline: one canonical host, clean redirects, valid markup, pages that return their content, reasonable Core Web Vitals. If that layer is broken you are paying twice for the same problem.

There is a second, less obvious overlap. Many of the sources engines cite are themselves ranked pages — roundups, comparisons, directories. So ranking is not irrelevant to GEO; it is one of the routes into the citation pool. That is why we run both rather than treating GEO as a replacement.

where they stop overlapping

The entity layer has no SEO equivalent. An SEO audit does not fail a site for being unverifiable as a company. A model does: with nothing to confirm you are a specific organisation, it either hedges or names someone it can confirm. Close to 100% of audited sites had no entity anchor.

The measurement has no SEO equivalent. There is no rank tracker for “does Claude mention us when asked who to hire”. The panel has to be run and read, and because answers are volatile it has to be a series.

Trust becomes machine-readable or it does not exist. In regulated categories we have audited licensed operators whose licence lived in an image while the complaint threads about them were plain text. SEO does not care. The answer engine does, and it will tell your buyer to verify you before paying.

which to fund first

Three cases, and they cover most situations.

The site fails a crawler check. Fix that first regardless of which channel you care about — it is the shared foundation and everything above it is wasted spend until it works.

Buyers arrive by comparing options themselves. Weight SEO. You need to be one of the documents they compare, and depth still wins there.

Buyers arrive holding a shortlist. Weight GEO, because the shortlist was written by a model reading sources you are not in. This is the case that has grown fastest in our own client work.

What does not work is choosing based on which channel is fashionable. The audit exists to answer this with your numbers rather than with a trend.

how to check your own split in ten minutes

Request your homepage with an AI crawler user-agent and read the body. Shell or 403 means the shared layer is broken.
Ask two engines your category question without naming yourself. Note who gets named and which sources are cited.
Ask the same engines about your brand directly. If they describe you correctly but never recommend you, you are in the 93% / 4.5% gap.
Open the cited sources. Whether you can realistically appear in them tells you what the next quarter of work actually is.

Definitions for anything unfamiliar are in the glossary; the full dataset is in 100 GEO audits; and if you would rather have it run properly, that is the audit.