Andy Hendricks

SEO and GEO share (some) inputs, but they don’t share measurement 

Andy Hendricks

Hiker viewing 2 trails, one for SEO and one for GEO

Every GEO tool I’ve seen in the last six months treats it as SEO’s natural evolution, like they are the same discipline to capture users in the new arena. It’s a clean story and clients like it, but it’s not the whole truth. The inputs do overlap, but GEO measurement and reporting is a different world from SEO.

What SEO and GEO Share

What makes a site rank well in Google also makes it legible to an LLM: content depth, schema, entity clarity, crawlability, E-E-A-T, freshness. If a search crawler can make sense of you, a model can too.

So if you’ve been doing SEO, you’ve already done a chunk of the GEO work. This is the overlap agencies lean on.

Where GEO Diverges

Third-party presence carries most of the weight. McKinsey found that brand-owned sites account for only 5–10% of the sources AI-powered search pulls from. The other 90+% is Reddit, G2, Capterra, Wikipedia, YouTube, trade press, etc.

Brand mentions matter more than backlinks. Ahrefs studied 75,000 brands and found branded web mentions (linked or unlinked) correlate with AI Overview visibility at 0.664. Total backlinks came in at 0.218. Correlation isn’t causation and big brands tend to score well on both, so don’t read this as a settled law. But the pattern held up across ChatGPT, AI Mode, and AI Overviews in Ahrefs’ December follow-up.

Original data gets cited. Opinion doesn’t. The Princeton GEO paper (Aggarwal et al., KDD 2024) ran nine optimization strategies across 10,000 queries and found that adding statistics lifted AI visibility by up to 41%, and adding quotations by 28%. Proprietary numbers and benchmarks get pulled into generated answers. Models want something quotable and original.

Formatting needs to be liftable. LLMs prefer Q&A blocks, inline stats, and clean sentences, so the model can grab content without having to rewrite it. The same article reorganized into extractable chunks performs differently than the flowing-prose version. Princeton’s data backs this up too, pages lower in the ranking see the biggest gains from citing sources and structuring content for extraction.

GEO Measurement is Heavily Limited Compared to SEO

SEO reporting is based on first party data with hard numbers. Search Console shows query-level impressions, clicks, CTR, position. SEMrush shows competitor movement and SERP features. GA4 shows organic traffic, engagement, conversions. When you make a change, you can see the impact over the next several months. Every input has a measurable output.

GEO is a different story. Most LLM interactions happen inside a private user session. There are no impressions to aggregate, no SERP to rank on, no referral string that tells you which prompt produced the visit.

Here’s what we can and can’t see from AI Traffic:

  • Partially measurable, through prompt sampling: citation frequency, share of voice against competitors, how the brand gets characterized.
  • Visible but uninterpretable: AI referral traffic shows up in GA4. We just don’t know which prompt sent it, which makes it hard to act on.
  • Not measurable: true citation volume, why a brand was or wasn’t cited, anything resembling a rank position.

Nate Elliott at eMarketer put it well: “Almost every GEO response is different from every other GEO response.” Prompt sampling gives you an idea, but not a number.

HubSpot and SEMrush both have AI visibility scores now, but they really are just automated prompt sampling with a dashboard on top. A custom set of prompts built from a client’s actual ICP will usually teach you more than the platform score will.

Summary

SEO is a measurable program. Strategy in, metrics out, attribution all the way through. Commit to the numbers and report against them.

GEO is a directional program. You can commit to the inputs: third-party presence, original research, extractable content, entity clarity, and you can show direction through prompt sampling and referral trends. However, you can’t commit to citation volume the way you commit to a keyword ranking, because what you’re measuring is just a sample. Anyone promising guaranteed, ranked GEO metrics right now is selling a level of certainty the data doesn’t support.

What’s Next

First-party AI data is finally starting to become available. Bing Webmaster Tools added Copilot query and citation data earlier this year, which makes it the first search console to show a real AI-side signal instead of a proxy. Ahrefs’ Brand Radar, Semrush’s AI Toolkit, Profound are all tracking citations across ChatGPT, AI Mode, and AI Overviews with more data than anything we had a year ago. None of it is in the Search Console yet, but the proxy layer is thickening, and the measurement story in twelve months will probably look different from today.

Sources

, , , , , , , ,

Leave a comment