Chapter 11: GEO Foundations

GEO cold start in four steps

1. Days 1–3 — open the doors. Check robots.txt and any edge firewall: the search and user-triggered bots need a 200, not a challenge page. Register the Bing webmaster tools and enable instant indexing. Publish an /llms.txt that describes what the site contains and where the canonical pages are.
2. Days 4–7 — build a baseline of 50 prompts. Twenty category-recommendation prompts, fifteen comparisons against named competitors, fifteen problem-solving prompts — drawn from sales calls and converting queries, not invented at a desk. Run each on the major answer engines, record whether you appear, at what position, in what tone, and — the useful part — which external sources get cited most often for those prompts.
3. Weeks 2–3 — restructure the pages that matter. For the top twenty pages, put a 40–60 word answer directly under each question heading, replace pronouns with the actual entity name so a chunk can stand alone, add a key-takeaways block, add a static HTML comparison table rather than a rendered widget, add first-party numbers, and ship the schema that matches the page type.
4. Week 4 onward — build the consensus layer. Take the list of sources the engines cited most in step two and earn your way onto them: directory and review profiles, the roundups that already get quoted, community threads with genuine practitioner answers. Being absent from the sources a model trusts cannot be compensated for on your own site.

Interview questions and answer frameworks

Q1: Is SEO dead now that everyone is talking about GEO? How do you combine them?

Framework: No — GEO runs on top of SEO, and the part that died is the part that deserved to: keyword-stuffed text written to attract a click and answer nothing. The reason is architectural. The major answer engines retrieve first and generate second: a query triggers a search over an index, the top candidates are fetched and read, and an answer is synthesised with citations. If you cannot be crawled, indexed, and ranked well enough to enter that candidate set, you are not in the conversation at all. So the practical model has three layers: retrieval — technical health, Bing as well as Google, topical clusters and links, so you reach the candidate pool; synthesis — structure each page so a passage can be lifted cleanly: a definition in the first sentence, concrete numbers, tables, unambiguous entities, no dangling pronouns; consensus — be present and accurate on the third-party sources those systems trust, because a model repeats what the web says about you, not what your homepage claims.

Q2: How do you measure GEO when there is no click to count?

Framework: Measure presence and accuracy, not sessions. Fix a prompt panel and a cadence; record share of answers where you appear, position within the answer, whether you are the primary citation, and whether the description is correct. Then track the second-order signals: branded search volume, direct traffic, and sales calls that mention a competitor comparison — those are where answer-engine exposure shows up commercially. State the limits honestly: models change between sessions, panels drift, and none of this attributes to revenue the way a session does. A GEO report that pretends to have the precision of a click report is worse than one that says “trend, method disclosed”.

Q3: A brand is invisible in AI answers but ranks well organically. Where do you look first?

Framework: Three candidates, in order of frequency. Extractability — the page ranks because it is comprehensive, but no single passage answers a question cleanly; the model prefers a page it can quote in one sentence. Entity ambiguity — the brand is described inconsistently across the web, so the model is not confident what it is or what it does. Source ecology — the models cite review sites, forums and roundups for that category, and the brand is missing or negatively represented there while its own site is immaculate. Check by reading the actual answer and its citations: who is being quoted, and what do those pages say about you?

Q4: What would you refuse to do in the name of GEO?

Framework: Anything that fabricates consensus. Seeding fake reviews, writing astroturf forum threads, editing knowledge bases to say something untrue, or generating dozens of “independent” comparison sites that all recommend one product. These work briefly and then become the evidence that gets you distrusted, because the same systems that read the web to praise you can read it to check you. The defensible version of the same effort is slower and permanent: publish something worth citing, make it easy to quote correctly, and be accurate everywhere you are described.